Reimplement rebase --merge via interactive machinery
Subsystems (2)
v2 cover Reimplement rebase --merge via interactive machinery
Now that the rewrite-interactive-rebases-in-C series have finally merged to master, this series deletes git-rebase--merge.sh and reimplements the --merge behavior on top of the interactive machinery.
Differences since v1:
- Updated code to hook into builtin/rebase.C instead of git-rebase.sh
(No range-diff provided, because it has been months since v1, and v1 was only RFC and was only discussed at a high level.)
Elijah Newren (2):
git-rebase, sequencer: extend --quiet option for the interactive
machinery
rebase: Implement --merge via git-rebase--interactive
.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/rebase.c | 15 ++- git-legacy-rebase.sh | 57 +++++------ git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 6 +- t/t5407-post-rewrite-hook.sh | 1 + t/t9903-bash-prompt.sh | 2 +- 15 files changed, 67 insertions(+), 318 deletions(-) delete mode 100644 git-rebase--merge.sh
Signature · 1 line
2.19.1.858.g526e8fe740.dirty
PATCH v2 1/2 git-rebase, sequencer: extend --quiet option for the interactive machinery
While 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Succesfully rebased..."
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)
builtin/rebase.c +1 -4
@@ -181,10 +181,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts))) opts->flags &= ~REBASE_NO_QUIET ;else opts->flags |=REBASE_NO_QUIET ;
git-legacy-rebase.sh +1 -1
@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&
git-rebase--common.sh +1 -1
@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \
sequencer.c +13 -10
@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")
@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {
@@ -2307,6 +2307,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;
@@ -2374,9 +2377,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)
@@ -3450,10 +3450,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());
@@ -3684,8 +3685,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);
sequencer.h +1 -0
@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;
Signature · 1 line
2.19.1.858.g526e8fe740.dirty
Re: [PATCH v2 1/2] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Elijah,
Quoted 8 lines · Elijah Newren
While 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
Quoted 4 lines
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Succesfully rebased..."
Makes sense.
Quoted 6 lines
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the
state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.
I am slightly concerned that some creative power user could have written scripts that rely on this behavior.
But only *slightly* concerned.
The patch looks correct.
Ciao,
Dscho
Quoted 137 lines
Signed-off-by Elijah Newren <[email protected]>--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)builtin/rebase.c +1
-4@@ -181,10 +181,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts))) opts->flags &= ~REBASE_NO_QUIET ;else opts->flags |=REBASE_NO_QUIET ;git-legacy-rebase.sh +1
-1@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&git-rebase--common.sh +1
-1@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \sequencer.c +13
-10@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {@@ -2307,6 +2307,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;@@ -2374,9 +2377,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)@@ -3450,10 +3450,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());@@ -3684,8 +3685,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);sequencer.h +1
-0@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;Signature · 1 line
2.19.1.858.g526e8fe740.dirty
PATCH v2 2/2 rebase: Implement --merge via git-rebase--interactive
In reply to Elijah NewrenInteractive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore.
Delete git-rebase--merge.sh and have the --merge option be implemented by the now built-in interactive machinery.
Note that this change fixes a few known test failures (see t3421).
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
--- .gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/rebase.c | 10 +- git-legacy-rebase.sh | 55 +++++----- git-rebase--merge.sh | 164 ------------------------------ t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 6 +- t/t5407-post-rewrite-hook.sh | 1 + t/t9903-bash-prompt.sh | 2 +- 12 files changed, 50 insertions(+), 302 deletions(-) delete mode 100644 git-rebase--merge.sh
.gitignore +0 -1
@@ -124,7 +124,6 @@ /git-rebase--am /git-rebase--common /git-rebase--interactive -/git-rebase--merge /git-rebase--preserve-merges /git-receive-pack /git-reflog
Documentation/git-rebase.txt +3 -14
@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS -------------------- -git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) - -Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date
@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options: * --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: - * --[no-]autosquash * --rebase-merges * --preserve-merges
@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoff
Makefile +0 -1
@@ -628,7 +628,6 @@ SCRIPT_LIB += git-parse-remote SCRIPT_LIB += git-rebase--am SCRIPT_LIB += git-rebase--common SCRIPT_LIB += git-rebase--preserve-merges -SCRIPT_LIB += git-rebase--merge SCRIPT_LIB += git-sh-setup SCRIPT_LIB += git-sh-i18n
builtin/rebase.c +5 -5
@@ -118,7 +118,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;
@@ -475,10 +475,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;
@@ -1156,6 +1152,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) { +imply_interactive (&options,"--merge" ); + } +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );
git-legacy-rebase.sh +25 -30
108 lines not on this page · raw message
git-rebase--merge.sh deleted +0 -164
164 lines not on this page · raw message
t/t3406-rebase-message.sh +1 -6
@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF - test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actual
t/t3420-rebase-autostash.sh +6 -72
102 lines not on this page · raw message
t/t3421-rebase-topology-linear.sh +5 -5
@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase failure -p
t/t3425-rebase-topology-merges.sh +3 -3
@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {
@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {
@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "
t/t5407-post-rewrite-hook.sh +1 -0
@@ -120,6 +120,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
t/t9903-bash-prompt.sh +1 -1
@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected && git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&
Signature · 1 line
2.19.1.858.g526e8fe740.dirty
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
Hi Elijah,
Quoted 8 lines · Elijah Newren
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore.
Delete
git-rebase--merge.shand have the --merge option be implemented by the now built-in interactive machinery.
Okay.
Quoted 1 line
Note that this change fixes a few known test failures (see t3421).
Nice!
Quoted 5 lines
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very
Quoted 8 lines
different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.
Ideally, the test would be fixed, then. If it fails for other reasons than a real regression, it is not a very good regression test, is it.
Quoted 4 lines
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.
This concerns me.
In bigger repositories (no, I am not talking about Linux kernel sized ones, I consider those small-ish) there are a ton of files, and checkout and commit (and even more so reset) sadly do not have a runtime complexity growing with the number of modified files, but with the number of tracked files (and some commands even with the number of worktree files).
So a larger number of commit/checkout operations makes me expect performance regressions.
In this light, could you elaborate a bit on the differences you see between rebase -i and rebase -m?
Quoted 2 lines
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
We should be able to make that test pass, still, by writing out a special file (e.g. $state_dir/opt_m) and testing for that. Users are oddly upset when their expectations are broken... (and I actually agree with them.)
Quoted 6 lines
Documentation/git-rebase.txt +0
-0@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 514
Quoted 10 lines
Documentation/git-rebase.txt +0
-8-------------------- -git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) -
Could we retain this part, with s/three/two/ and /git-merge/d? *Maybe* s/interactive backend/interactive\/merge backend/
Quoted 23 lines
Documentation/git-rebase.txt +2
-5-Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options: * --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: - * --[no-]autosquash * --rebase-merges * --preserve-merges
Documentation/git-rebase.txt Flags only understood by the interactive backend: · line 534
Quoted 8 lines
Documentation/git-rebase.txt +1
-1@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoff
The rest of the diff is funny to read, but the post image makes a lot of sense.
Quoted 25 lines
builtin/rebase.c +1
-5@@ -118,7 +118,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;@@ -475,10 +475,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;
builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1161
Quoted 10 lines
builtin/rebase.c +4
-0@@ -1156,6 +1152,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) { +imply_interactive (&options,"--merge" ); + } +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );
Okay, this makes sense.
Quoted 47 lines
git-legacy-rebase.sh +5
-5@@ -218,6 +218,7 @@ then state_dir="$apply_dir" elif test -d "$merge_dir" then + type=interactive if test -d "$merge_dir"/rewritten then type=preserve-merges@@ -225,10 +226,7 @@ then preserve_merges=t elif test -f "$merge_dir"/interactive then - type=interactive interactive_rebase=explicit - else - type=merge fi state_dir="$merge_dir" fi@@ -469,6 +467,7 @@ then test -z "$interactive_rebase" && interactive_rebase=implied fi +actually_interactive= if test -n "$interactive_rebase" then if test -z "$preserve_merges"@@ -477,11 +476,12 @@ then else type=preserve-merges fi - + actually_interactive=t state_dir="$merge_dir" elif test -n "$do_merge" then - type=merge + interactive_rebase=implied + type=interactive state_dir="$merge_dir" else type=am
git-legacy-rebase.sh then · line 497
Quoted 9 lines
git-legacy-rebase.sh +2
-4@@ -493,21 +493,13 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fi -if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')
Why are we no longer guarding this behind the condition that the user specified *any* option intended for the am backend?
Quoted 8 lines
git-legacy-rebase.sh +3
-4+if test -n "$incompatible_opts" +then + if test -n "$actually_interactive" || test "$do_merge" then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" - fi
git-legacy-rebase.sh then · line 505
Quoted 10 lines
git-legacy-rebase.sh +1
-6- fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" - fi + die "$(gettext "error: cannot combine am options ($incompatible_opts) with either interactive or merge options")" fi fi
The rest of this hunk makes sense.
Quoted 10 lines
git-legacy-rebase.sh +1
-1@@ -672,7 +664,7 @@ require_clean_work_tree "rebase" "$(gettext "Please commit or stash them.")" # but this should be done only when upstream and onto are the same # and if this is not an interactive rebase. mb=$(git merge-base "$onto" "$orig_head") -if test -z "$interactive_rebase" && test "$upstream" = "$onto" && +if test -z "$actually_interactive" && test "$upstream" = "$onto" && test "$mb" = "$onto" && test -z "$restrict_revision" && # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/null
git-legacy-rebase.sh then · line 714
Quoted 7 lines
git-legacy-rebase.sh +4
-0@@ -716,6 +708,19 @@ then GIT_PAGER='' git diff --stat --summary "$mb" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head" +then + # If the $onto is a proper descendant of the tip of the branch, then + # we just fast-forwarded.
This comment is misleading if it comes before, rather than after, the actual checkout -q command.
Quoted 13 lines
git-legacy-rebase.sh +9
-0+ say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" + GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ + git checkout -q "$onto^0" || die "could not detach HEAD" + git update-ref ORIG_HEAD $orig_head + move_to_original_branch + finish_rebase + exit 0 +fi + test -n "$interactive_rebase" && run_specific_rebase # Detach HEAD and reset the tree
git-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 731
Quoted 2 lines
git-legacy-rebase.sh +0
-0@@ -725,16 +730,6 @@ GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ git checkout -q "$onto^0" || die "could not detach HEAD" git update-ref ORIG_HEAD $orig_head
It is a pity that this hunk header hides the lines that you copied into the following conditional before moving it before the "if interactive, run it" line:
Quoted 4 lines
git-legacy-rebase.sh +0
-3-# If the $onto is a proper descendant of the tip of the branch, then -# we just fast-forwarded. -if test "$mb" = "$orig_head"
git-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 734
Quoted 10 lines
git-legacy-rebase.sh +0
-7-then - say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" - move_to_original_branch - finish_rebase - exit 0 -fi - if test -n "$rebase_root" then revisions="$onto..$orig_head"
What you did is correct, if duplicating code, and definitely the easiest way to do it. Just move the comment, and we're good here.
git-rebase--merge.sh
Quoted 3 lines
git-rebase--merge.sh deleted +0
-0
Nice. Really nice!
Quoted 5 lines
t/t3406-rebase-message.sh +0
-0
t/t3406-rebase-message.sh test_expect_success 'setup' ' · line 25
Quoted 9 lines
t/t3406-rebase-message.sh +0
-6@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF -
As I had mentioned in the previous round: I think we can retain these messages for the --merge mode. And I think we should: users of --merge have most likely become to expect them.
The most elegant way would probably be by adding code to the sequencer that outputs these lines if in "merge" mode (and add a flag to say that we *are* in "merge" mode).
To that end, the make_script() function would not generate # pick but drop lines, I think, so that the sequencer can see those and print the Already applied: <message> lines. And a successful TODO_PICK would write out Committed: <message>.
Quoted 107 lines
t/t3406-rebase-message.sh +1
-0test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actualt/t3420-rebase-autostash.sh +3
-72@@ -53,41 +53,6 @@ create_expected_success_interactive () { EOF } -create_expected_success_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applied autostash. - EOF -} - create_expected_failure_am () { cat >expected <<-EOF $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual)@@ -112,43 +77,6 @@ create_expected_failure_interactive () { EOF } -create_expected_failure_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applying autostash resulted in conflicts. - Your changes are safe in the stash. - You can run "git stash pop" or "git stash drop" at any time. - EOF -} - testrebase () { type=$1 dotest=$2@@ -177,6 +105,9 @@ testrebase () { test_expect_success "rebase$type --autostash: check output" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_success_$suffix && test_i18ncmp expected actual '
t/t3420-rebase-autostash.sh testrebase () { · line 213
Quoted 9 lines
t/t3420-rebase-autostash.sh +3
-0@@ -274,6 +205,9 @@ testrebase () { test_expect_success "rebase$type: check output with conflicting stash" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_failure_$suffix && test_i18ncmp expected actual '
As I stated earlier, I am uncomfortable with this solution. We change behavior rather noticeably, and the regression test tells us the same. We need to either try to deprecate git rebase --merge, or we need to at least attempt to recreate the old behavior with the sequencer.
Quoted 41 lines
t/t3421-rebase-topology-linear.sh +4
-4@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
t/t3421-rebase-topology-linear.sh test_run_rebase () { · line 321
Quoted 8 lines
t/t3421-rebase-topology-linear.sh +1
-1@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase failure -p
Nice.
Quoted 23 lines
t/t3425-rebase-topology-merges.sh +2
-2@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {
t/t3425-rebase-topology-merges.sh test_run_rebase () { · line 112
Quoted 8 lines
t/t3425-rebase-topology-merges.sh +1
-1@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "
This is a bit unfortunate. I wonder if we could do something like this, to retain the current behavior:
-- snip --
sequencer.c +2 -1
@@ -4394,7 +4394,8 @@ int sequencer_make_script(FILE *out, int argc, const char **argv, revs.reverse =1 ; revs.right_only =1 ; revs.sort_order =REV_SORT_IN_GRAPH_ORDER ; - revs.topo_order =1 ; +if (!(flags &TODO_LIST_NO_TOPO_ORDER )) + revs.topo_order =1 ; revs.pretty_given =1 ;git_config_get_string ("rebase.instructionFormat" , &format);
-- snap -
(and then pass TODO_LIST_NO_TOPO_ORDER if in "merge" mode)?
t/t5407-post-rewrite-hook.sh test_expect_success 'git rebase -m --skip' ' · line 126
Quoted 12 lines
t/t5407-post-rewrite-hook.sh +1
-0@@ -120,6 +120,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
This suggests to me that we call commit too many times. I think we really need to address this without changing the test. This is a change in behavior.
t/t9903-bash-prompt.sh test_expect_success 'prompt - interactive rebase' ' · line 183
Quoted 10 lines
t/t9903-bash-prompt.sh +1
-1@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected &&
As I said, this should be easily addressed by having a tell-tale in $state_dir/. Come to think of it, we must have had one, right? Let's see how the bash-prompt does it.
*clicketyclick*
Ah, it is the *absence* of the interactive file... Which makes me wonder whether we should not simply retain that behavior for git rebase --merge?
Quoted 7 lines
t/t9903-bash-prompt.sh +0
-0git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&Signature · 1 line
2.19.1.858.g526e8fe740.dirty
Thank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
Thanks!
Dscho
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
Hi Elijah
It's good to see these patches progressing, I've just got a couple of comments related to Johannes' points below
Quoted 70 lines · Johannes Schindelin
Hi Elijah,
Quoted 8 lines · Elijah Newren
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore.
Delete
git-rebase--merge.shand have the --merge option be implemented by the now built-in interactive machinery.Okay.
Quoted 1 line
Note that this change fixes a few known test failures (see t3421).
Nice!
Quoted 5 lines
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on veryQuoted 8 lines
different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.Ideally, the test would be fixed, then. If it fails for other reasons than a real regression, it is not a very good regression test, is it.
Quoted 4 lines
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.This concerns me.
In bigger repositories (no, I am not talking about Linux kernel sized ones, I consider those small-ish) there are a ton of files, and checkout and commit (and even more so reset) sadly do not have a runtime complexity growing with the number of modified files, but with the number of tracked files (and some commands even with the number of worktree files).
So a larger number of commit/checkout operations makes me expect performance regressions.
In this light, could you elaborate a bit on the differences you see between rebase -i and rebase -m?
Quoted 2 lines
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.We should be able to make that test pass, still, by writing out a special file (e.g. $state_dir/opt_m) and testing for that. Users are oddly upset when their expectations are broken... (and I actually agree with them.)
Quoted 8 lines
Documentation/git-rebase.txt +0
-0@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS --------------------
Quoted 11 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 514
Quoted 8 lines
Documentation/git-rebase.txt +0
-8-git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) -Could we retain this part, with
s/three/two/and/git-merge/d? *Maybe*s/interactive backend/interactive\/merge backend/
I was hoping we could get rid of that, I'm not sure how useful it is to users.
Quoted 13 lines
Quoted 13 lines
Documentation/git-rebase.txt +2
-2-Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options:
Quoted 7 lines
Documentation/git-rebase.txt Flags only understood by the am backend: · line 531
Quoted 7 lines
Documentation/git-rebase.txt +0
-3* --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: -
It's nice to see this being simplified
Quoted 104 lines
Documentation/git-rebase.txt Flags only understood by the interactive backend: · line 534
Quoted 12 lines
Documentation/git-rebase.txt +1
-1* --[no-]autosquash * --rebase-merges * --preserve-merges@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoffThe rest of the diff is funny to read, but the post image makes a lot of sense.
Quoted 25 lines
builtin/rebase.c +1
-5@@ -118,7 +118,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;@@ -475,10 +475,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1161
Quoted 10 lines
builtin/rebase.c +4
-0@@ -1156,6 +1152,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) { +imply_interactive (&options,"--merge" ); + } +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );Okay, this makes sense.
Quoted 50 lines
git-legacy-rebase.sh +5
-5@@ -218,6 +218,7 @@ then state_dir="$apply_dir" elif test -d "$merge_dir" then + type=interactive if test -d "$merge_dir"/rewritten then type=preserve-merges@@ -225,10 +226,7 @@ then preserve_merges=t elif test -f "$merge_dir"/interactive then - type=interactive interactive_rebase=explicit - else - type=merge fi state_dir="$merge_dir" fi@@ -469,6 +467,7 @@ then test -z "$interactive_rebase" && interactive_rebase=implied fi +actually_interactive= if test -n "$interactive_rebase" then if test -z "$preserve_merges"@@ -477,11 +476,12 @@ then else type=preserve-merges fi - + actually_interactive=t state_dir="$merge_dir" elif test -n "$do_merge" then - type=merge + interactive_rebase=implied + type=interactive state_dir="$merge_dir" else type=am@@ -493,21 +493,13 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fi
Quoted 9 lines
git-legacy-rebase.sh then · line 497
Quoted 6 lines
git-legacy-rebase.sh +2
-4-if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')Why are we no longer guarding this behind the condition that the user specified *any* option intended for the
ambackend?
I was confused by this as well, what if the user asks for 'rebase --exec=<cmd> --ignore-whitespace'?
Quoted 11 lines
Quoted 11 lines
git-legacy-rebase.sh +3
-6+if test -n "$incompatible_opts" +then + if test -n "$actually_interactive" || test "$do_merge" then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" - fi - fi - if test -n "$do_merge"; then
Quoted 8 lines
git-legacy-rebase.sh then · line 505
Quoted 8 lines
git-legacy-rebase.sh +1
-4- if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" - fi + die "$(gettext "error: cannot combine am options ($incompatible_opts) with either interactive or merge options")" fi fi
If you want to change the error message here, I think you need to change the corresponding message in builtin/rebase.c
Best Wishes
Phillip
Quoted 379 lines
The rest of this hunk makes sense.
Quoted 10 lines
git-legacy-rebase.sh +1
-1@@ -672,7 +664,7 @@ require_clean_work_tree "rebase" "$(gettext "Please commit or stash them.")" # but this should be done only when upstream and onto are the same # and if this is not an interactive rebase. mb=$(git merge-base "$onto" "$orig_head") -if test -z "$interactive_rebase" && test "$upstream" = "$onto" && +if test -z "$actually_interactive" && test "$upstream" = "$onto" && test "$mb" = "$onto" && test -z "$restrict_revision" && # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/nullgit-legacy-rebase.sh then · line 714
Quoted 7 lines
git-legacy-rebase.sh +4
-0@@ -716,6 +708,19 @@ then GIT_PAGER='' git diff --stat --summary "$mb" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head" +then + # If the $onto is a proper descendant of the tip of the branch, then + # we just fast-forwarded.This comment is misleading if it comes before, rather than after, the actual
checkout -qcommand.Quoted 13 lines
git-legacy-rebase.sh +9
-0+ say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" + GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ + git checkout -q "$onto^0" || die "could not detach HEAD" + git update-ref ORIG_HEAD $orig_head + move_to_original_branch + finish_rebase + exit 0 +fi + test -n "$interactive_rebase" && run_specific_rebase # Detach HEAD and reset the treegit-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 731
Quoted 2 lines
git-legacy-rebase.sh +0
-0@@ -725,16 +730,6 @@ GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ git checkout -q "$onto^0" || die "could not detach HEAD" git update-ref ORIG_HEAD $orig_headIt is a pity that this hunk header hides the lines that you copied into the following conditional before moving it before the "if interactive, run it" line:
Quoted 4 lines
git-legacy-rebase.sh +0
-3-# If the $onto is a proper descendant of the tip of the branch, then -# we just fast-forwarded. -if test "$mb" = "$orig_head"git-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 734
Quoted 10 lines
git-legacy-rebase.sh +0
-7-then - say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" - move_to_original_branch - finish_rebase - exit 0 -fi - if test -n "$rebase_root" then revisions="$onto..$orig_head"What you did is correct, if duplicating code, and definitely the easiest way to do it. Just move the comment, and we're good here.
git-rebase--merge.sh
Quoted 3 lines
git-rebase--merge.sh deleted +0
-0Nice. Really nice!
Quoted 5 lines
t/t3406-rebase-message.sh +0
-0t/t3406-rebase-message.sh test_expect_success 'setup' ' · line 25
Quoted 9 lines
t/t3406-rebase-message.sh +0
-6@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF -As I had mentioned in the previous round: I think we can retain these messages for the
--mergemode. And I think we should: users of --merge have most likely become to expect them.The most elegant way would probably be by adding code to the sequencer that outputs these lines if in "merge" mode (and add a flag to say that we *are* in "merge" mode).
To that end, the
make_script()function would not generate# pickbutdroplines, I think, so that the sequencer can see those and print theAlready applied: <message>lines. And a successfulTODO_PICKwould write outCommitted: <message>.Quoted 107 lines
t/t3406-rebase-message.sh +1
-0test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actualt/t3420-rebase-autostash.sh +3
-72@@ -53,41 +53,6 @@ create_expected_success_interactive () { EOF } -create_expected_success_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applied autostash. - EOF -} - create_expected_failure_am () { cat >expected <<-EOF $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual)@@ -112,43 +77,6 @@ create_expected_failure_interactive () { EOF } -create_expected_failure_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applying autostash resulted in conflicts. - Your changes are safe in the stash. - You can run "git stash pop" or "git stash drop" at any time. - EOF -} - testrebase () { type=$1 dotest=$2@@ -177,6 +105,9 @@ testrebase () { test_expect_success "rebase$type --autostash: check output" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_success_$suffix && test_i18ncmp expected actual 't/t3420-rebase-autostash.sh testrebase () { · line 213
Quoted 9 lines
t/t3420-rebase-autostash.sh +3
-0@@ -274,6 +205,9 @@ testrebase () { test_expect_success "rebase$type: check output with conflicting stash" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_failure_$suffix && test_i18ncmp expected actual 'As I stated earlier, I am uncomfortable with this solution. We change behavior rather noticeably, and the regression test tells us the same. We need to either try to deprecate
git rebase --merge, or we need to at least attempt to recreate the old behavior with the sequencer.Quoted 41 lines
t/t3421-rebase-topology-linear.sh +4
-4@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -pt/t3421-rebase-topology-linear.sh test_run_rebase () { · line 321
Quoted 8 lines
t/t3421-rebase-topology-linear.sh +1
-1@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase failure -pNice.
Quoted 23 lines
t/t3425-rebase-topology-merges.sh +2
-2@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {t/t3425-rebase-topology-merges.sh test_run_rebase () { · line 112
Quoted 8 lines
t/t3425-rebase-topology-merges.sh +1
-1@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "This is a bit unfortunate. I wonder if we could do something like this, to retain the current behavior:
-- snip --
sequencer.c +2
-2@@ -4394,7 +4394,8 @@ int sequencer_make_script(FILE *out, int argc, const char **argv, revs.reverse =1 ; revs.right_only =1 ; revs.sort_order =REV_SORT_IN_GRAPH_ORDER ; - revs.topo_order =1 ; +if (!(flags &TODO_LIST_NO_TOPO_ORDER )) + revs.topo_order =1 ; revs.pretty_given =1 ;git_config_get_string ("rebase.instructionFormat" , &format); -- snap -(and then pass
TODO_LIST_NO_TOPO_ORDERif in "merge" mode)?t/t5407-post-rewrite-hook.sh test_expect_success 'git rebase -m --skip' ' · line 126
Quoted 12 lines
t/t5407-post-rewrite-hook.sh +1
-0@@ -120,6 +120,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_inputThis suggests to me that we call
committoo many times. I think we really need to address this without changing the test. This is a change in behavior.t/t9903-bash-prompt.sh test_expect_success 'prompt - interactive rebase' ' · line 183
Quoted 10 lines
t/t9903-bash-prompt.sh +1
-1@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected &&As I said, this should be easily addressed by having a tell-tale in $state_dir/. Come to think of it, we must have had one, right? Let's see how the bash-prompt does it.
*clicketyclick*
Ah, it is the *absence* of the
interactivefile... Which makes me wonder whether we should not simply retain that behavior forgit rebase --merge?Quoted 7 lines
t/t9903-bash-prompt.sh +0
-0git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&Signature · 1 line
2.19.1.858.g526e8fe740.dirtyThank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
Thanks!
Dscho
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
Hi Phillip,
Quoted 86 lines · Phillip Wood
It's good to see these patches progressing, I've just got a couple of comments related to Johannes' points below
Quoted 70 lines · Johannes Schindelin
Hi Elijah,
Quoted 8 lines · Elijah Newren
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore.
Delete
git-rebase--merge.shand have the --merge option be implemented by the now built-in interactive machinery.Okay.
Quoted 1 line
Note that this change fixes a few known test failures (see t3421).
Nice!
Quoted 5 lines
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on veryQuoted 8 lines
different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.Ideally, the test would be fixed, then. If it fails for other reasons than a real regression, it is not a very good regression test, is it.
Quoted 4 lines
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.This concerns me.
In bigger repositories (no, I am not talking about Linux kernel sized ones, I consider those small-ish) there are a ton of files, and checkout and commit (and even more so reset) sadly do not have a runtime complexity growing with the number of modified files, but with the number of tracked files (and some commands even with the number of worktree files).
So a larger number of commit/checkout operations makes me expect performance regressions.
In this light, could you elaborate a bit on the differences you see between rebase -i and rebase -m?
Quoted 2 lines
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.We should be able to make that test pass, still, by writing out a special file (e.g. $state_dir/opt_m) and testing for that. Users are oddly upset when their expectations are broken... (and I actually agree with them.)
Quoted 8 lines
Documentation/git-rebase.txt +0
-0@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS --------------------Quoted 11 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 514
Quoted 8 lines
Documentation/git-rebase.txt +0
-8-git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) -Could we retain this part, with
s/three/two/and/git-merge/d? *Maybe*s/interactive backend/interactive\/merge backend/
Quoted 2 lines
I was hoping we could get rid of that, I'm not sure how useful it is to users.
That's a good point. If the commit message makes the case for that (it is an implementation detail that confuses users), I am fine with removing it, too.
Thanks,
Dscho
Quoted 549 lines
Quoted 14 lines
Quoted 14 lines
Documentation/git-rebase.txt +2
-2-Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options:Quoted 7 lines
Documentation/git-rebase.txt Flags only understood by the am backend: · line 531
Quoted 7 lines
Documentation/git-rebase.txt +0
-3* --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: -It's nice to see this being simplified
Quoted 104 lines
Documentation/git-rebase.txt Flags only understood by the interactive backend: · line 534
Quoted 12 lines
Documentation/git-rebase.txt +1
-1* --[no-]autosquash * --rebase-merges * --preserve-merges@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoffThe rest of the diff is funny to read, but the post image makes a lot of sense.
Quoted 25 lines
builtin/rebase.c +1
-5@@ -118,7 +118,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;@@ -475,10 +475,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1161
Quoted 10 lines
builtin/rebase.c +4
-0@@ -1156,6 +1152,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) { +imply_interactive (&options,"--merge" ); + } +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );Okay, this makes sense.
Quoted 50 lines
git-legacy-rebase.sh +5
-5@@ -218,6 +218,7 @@ then state_dir="$apply_dir" elif test -d "$merge_dir" then + type=interactive if test -d "$merge_dir"/rewritten then type=preserve-merges@@ -225,10 +226,7 @@ then preserve_merges=t elif test -f "$merge_dir"/interactive then - type=interactive interactive_rebase=explicit - else - type=merge fi state_dir="$merge_dir" fi@@ -469,6 +467,7 @@ then test -z "$interactive_rebase" && interactive_rebase=implied fi +actually_interactive= if test -n "$interactive_rebase" then if test -z "$preserve_merges"@@ -477,11 +476,12 @@ then else type=preserve-merges fi - + actually_interactive=t state_dir="$merge_dir" elif test -n "$do_merge" then - type=merge + interactive_rebase=implied + type=interactive state_dir="$merge_dir" else type=am@@ -493,21 +493,13 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fiQuoted 9 lines
git-legacy-rebase.sh then · line 497
Quoted 6 lines
git-legacy-rebase.sh +2
-4-if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')Why are we no longer guarding this behind the condition that the user specified *any* option intended for the
ambackend?I was confused by this as well, what if the user asks for 'rebase --exec=<cmd> --ignore-whitespace'?
Quoted 11 lines
Quoted 11 lines
git-legacy-rebase.sh +3
-6+if test -n "$incompatible_opts" +then + if test -n "$actually_interactive" || test "$do_merge" then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" - fi - fi - if test -n "$do_merge"; thenQuoted 8 lines
git-legacy-rebase.sh then · line 505
Quoted 8 lines
git-legacy-rebase.sh +1
-4- if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" - fi + die "$(gettext "error: cannot combine am options ($incompatible_opts) with either interactive or merge options")" fi fiIf you want to change the error message here, I think you need to change the corresponding message in
builtin/rebase.cBest Wishes
Phillip
Quoted 379 lines
The rest of this hunk makes sense.
Quoted 10 lines
git-legacy-rebase.sh +1
-1@@ -672,7 +664,7 @@ require_clean_work_tree "rebase" "$(gettext "Please commit or stash them.")" # but this should be done only when upstream and onto are the same # and if this is not an interactive rebase. mb=$(git merge-base "$onto" "$orig_head") -if test -z "$interactive_rebase" && test "$upstream" = "$onto" && +if test -z "$actually_interactive" && test "$upstream" = "$onto" && test "$mb" = "$onto" && test -z "$restrict_revision" && # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/nullgit-legacy-rebase.sh then · line 714
Quoted 7 lines
git-legacy-rebase.sh +4
-0@@ -716,6 +708,19 @@ then GIT_PAGER='' git diff --stat --summary "$mb" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head" +then + # If the $onto is a proper descendant of the tip of the branch, then + # we just fast-forwarded.This comment is misleading if it comes before, rather than after, the actual
checkout -qcommand.Quoted 13 lines
git-legacy-rebase.sh +9
-0+ say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" + GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ + git checkout -q "$onto^0" || die "could not detach HEAD" + git update-ref ORIG_HEAD $orig_head + move_to_original_branch + finish_rebase + exit 0 +fi + test -n "$interactive_rebase" && run_specific_rebase # Detach HEAD and reset the treegit-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 731
Quoted 2 lines
git-legacy-rebase.sh +0
-0@@ -725,16 +730,6 @@ GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ git checkout -q "$onto^0" || die "could not detach HEAD" git update-ref ORIG_HEAD $orig_headIt is a pity that this hunk header hides the lines that you copied into the following conditional before moving it before the "if interactive, run it" line:
Quoted 4 lines
git-legacy-rebase.sh +0
-3-# If the $onto is a proper descendant of the tip of the branch, then -# we just fast-forwarded. -if test "$mb" = "$orig_head"git-legacy-rebase.sh GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 734
Quoted 10 lines
git-legacy-rebase.sh +0
-7-then - say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" - move_to_original_branch - finish_rebase - exit 0 -fi - if test -n "$rebase_root" then revisions="$onto..$orig_head"What you did is correct, if duplicating code, and definitely the easiest way to do it. Just move the comment, and we're good here.
git-rebase--merge.sh
Quoted 3 lines
git-rebase--merge.sh deleted +0
-0Nice. Really nice!
Quoted 5 lines
t/t3406-rebase-message.sh +0
-0t/t3406-rebase-message.sh test_expect_success 'setup' ' · line 25
Quoted 9 lines
t/t3406-rebase-message.sh +0
-6@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF -As I had mentioned in the previous round: I think we can retain these messages for the
--mergemode. And I think we should: users of --merge have most likely become to expect them.The most elegant way would probably be by adding code to the sequencer that outputs these lines if in "merge" mode (and add a flag to say that we *are* in "merge" mode).
To that end, the
make_script()function would not generate# pickbutdroplines, I think, so that the sequencer can see those and print theAlready applied: <message>lines. And a successfulTODO_PICKwould write outCommitted: <message>.Quoted 107 lines
t/t3406-rebase-message.sh +1
-0test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actualt/t3420-rebase-autostash.sh +3
-72@@ -53,41 +53,6 @@ create_expected_success_interactive () { EOF } -create_expected_success_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applied autostash. - EOF -} - create_expected_failure_am () { cat >expected <<-EOF $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual)@@ -112,43 +77,6 @@ create_expected_failure_interactive () { EOF } -create_expected_failure_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applying autostash resulted in conflicts. - Your changes are safe in the stash. - You can run "git stash pop" or "git stash drop" at any time. - EOF -} - testrebase () { type=$1 dotest=$2@@ -177,6 +105,9 @@ testrebase () { test_expect_success "rebase$type --autostash: check output" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_success_$suffix && test_i18ncmp expected actual 't/t3420-rebase-autostash.sh testrebase () { · line 213
Quoted 9 lines
t/t3420-rebase-autostash.sh +3
-0@@ -274,6 +205,9 @@ testrebase () { test_expect_success "rebase$type: check output with conflicting stash" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_failure_$suffix && test_i18ncmp expected actual 'As I stated earlier, I am uncomfortable with this solution. We change behavior rather noticeably, and the regression test tells us the same. We need to either try to deprecate
git rebase --merge, or we need to at least attempt to recreate the old behavior with the sequencer.Quoted 41 lines
t/t3421-rebase-topology-linear.sh +4
-4@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -pt/t3421-rebase-topology-linear.sh test_run_rebase () { · line 321
Quoted 8 lines
t/t3421-rebase-topology-linear.sh +1
-1@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase failure -pNice.
Quoted 23 lines
t/t3425-rebase-topology-merges.sh +2
-2@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {t/t3425-rebase-topology-merges.sh test_run_rebase () { · line 112
Quoted 8 lines
t/t3425-rebase-topology-merges.sh +1
-1@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "This is a bit unfortunate. I wonder if we could do something like this, to retain the current behavior:
-- snip --
sequencer.c +2
-2@@ -4394,7 +4394,8 @@ int sequencer_make_script(FILE *out, int argc, const char **argv, revs.reverse =1 ; revs.right_only =1 ; revs.sort_order =REV_SORT_IN_GRAPH_ORDER ; - revs.topo_order =1 ; +if (!(flags &TODO_LIST_NO_TOPO_ORDER )) + revs.topo_order =1 ; revs.pretty_given =1 ;git_config_get_string ("rebase.instructionFormat" , &format); -- snap -(and then pass
TODO_LIST_NO_TOPO_ORDERif in "merge" mode)?t/t5407-post-rewrite-hook.sh test_expect_success 'git rebase -m --skip' ' · line 126
Quoted 12 lines
t/t5407-post-rewrite-hook.sh +1
-0@@ -120,6 +120,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_inputThis suggests to me that we call
committoo many times. I think we really need to address this without changing the test. This is a change in behavior.t/t9903-bash-prompt.sh test_expect_success 'prompt - interactive rebase' ' · line 183
Quoted 10 lines
t/t9903-bash-prompt.sh +1
-1@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected &&As I said, this should be easily addressed by having a tell-tale in $state_dir/. Come to think of it, we must have had one, right? Let's see how the bash-prompt does it.
*clicketyclick*
Ah, it is the *absence* of the
interactivefile... Which makes me wonder whether we should not simply retain that behavior forgit rebase --merge?Quoted 7 lines
t/t9903-bash-prompt.sh +0
-0git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&Signature · 1 line
2.19.1.858.g526e8fe740.dirtyThank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
Thanks!
Dscho
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
In reply to Phillip WoodHi Phillip,
Quoted 12 lines · Phillip Wood
Quoted 12 lines
Quoted 12 lines
(quoted diff) +2
-2-Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options:
Quoted 9 lines
Quoted 7 lines
Review Flags only understood by the am backend: · line 531
Quoted 7 lines
(quoted diff) +0
-3* --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: -It's nice to see this being simplified
:-)
Quoted 10 lines
Quoted 9 lines
Review Flags only understood by the am backend:
Quoted 6 lines
(quoted diff) +2
-4-if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')Why are we no longer guarding this behind the condition that the user specified *any* option intended for the
ambackend?
Quoted 2 lines
I was confused by this as well, what if the user asks for 'rebase --exec=<cmd> --ignore-whitespace'?
They'd still get an error message about incompatible options; see my email to Dscho. However, since it tripped you both up, I'll make the clean up here a separate commit with some comments.
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
(quoted diff) +3
-9+if test -n "$incompatible_opts" +then + if test -n "$actually_interactive" || test "$do_merge" then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" - fi - fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")"
Quoted 8 lines
Quoted 5 lines
Review
Quoted 5 lines
(quoted diff) +1
-1- fi + die "$(gettext "error: cannot combine am options ($incompatible_opts) with either interactive or merge options")" fi fiIf you want to change the error message here, I think you need to change the corresponding message in
builtin/rebase.c
Indeed, will fix.
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
In reply to Johannes SchindelinHi Dscho,
Thanks for the detailed review! I'll try to get back to all your comments, but for now I feel bad that I didn't see and respond to at least one sooner...
Quoted 6 lines · Johannes Schindelin
Thank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
https://github.com/newren/git/tree/rebase-new-default, but ignore the last two commits; they are separate and haven't been brought up to date despite having been minimally rebased.
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
In reply to Johannes SchindelinHi Dscho,
Quoted 7 lines · Johannes Schindelin
Quoted 4 lines
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.Ideally, the test would be fixed, then. If it fails for other reasons than a real regression, it is not a very good regression test, is it.
I agree, it's not the best regression test. It'd be better if it defined and tested for "the correct behavior", but I suspect it has some value in that it's is guaranteeing that the rebase flavors at least give a "somewhat reasonable" result. Sadly, It just allowed all three rebase types to have slightly different "somewhat reasonable" answers.
Quoted 14 lines
Quoted 5 lines
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.This concerns me.
In bigger repositories (no, I am not talking about Linux kernel sized ones, I consider those small-ish) there are a ton of files, and checkout and commit (and even more so reset) sadly do not have a runtime complexity growing with the number of modified files, but with the number of tracked files (and some commands even with the number of worktree files).
Quoted 5 lines
So a larger number of commit/checkout operations makes me expect performance regressions.
In this light, could you elaborate a bit on the differences you see between rebase -i and rebase -m?
I wrote this comment months ago and don't remember the full details. From the wording and as best I remember, I suspect I was at least partially annoyed by the fact that these and other regression tests for rebase seemed to not be testing for "correct" behavior, but "existing" behavior. That wouldn't be so bad, except that existing behavior differed on the exact same test setup for different rebase backends and the differences in behavior were checked and enforced in the regression tests. So, it became a matter of taste as to how much things should be made identical and to which backend, or whether it was more important to at least just consolidate the code first and keep the results at least reasonable. At the time, I figured getting fewer backends, all with reasonable behavior was a bit more important, but since this difference is worrisome to you, I will try to dig further into it.
Quoted 6 lines
Quoted 2 lines
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.We should be able to make that test pass, still, by writing out a special file (e.g. $state_dir/opt_m) and testing for that. Users are oddly upset when their expectations are broken... (and I actually agree with them.)
I agree users are upset when expectations are broken, but why would they expect REBASE-m? In fact, I thought the prompt was a reflection of what backend was in use, so that if someone reported an issue to the git mailing list or a power user wanted to know where the control files were located, they could look for them. As such, I thought that switching the prompt to REBASE-i was the right thing to do so that it would match user expectations. Yes, the backend changed; that's part of the release notes.
Is there documentation somewhere that specifies the meaning of this prompt differently than the expectations I had somehow built up?
Quoted 8 lines
Quoted 8 lines
Documentation/git-rebase.txt +0
-0@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS --------------------
Quoted 11 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 514
Quoted 8 lines
Documentation/git-rebase.txt +0
-8-git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) -Could we retain this part, with
s/three/two/and/git-merge/d? *Maybe*s/interactive backend/interactive\/merge backend/
Removing this was actually a suggestion by Phillip back in June at the end of https://public-inbox.org/git/[email protected]/,
for the purpose of removing an implementation details that users don't need to know about. As noted elsewhere in the thread, between you and Phillip, I'll add some comments to the commit message about this.
Quoted 9 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 508
Quoted 6 lines
Documentation/git-rebase.txt +2
-4-if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')Why are we no longer guarding this behind the condition that the user specified *any* option intended for the
ambackend?
The code is still correctly guarding, the diff is just confusing. Sorry about that. To explain, the code previously was basically:
if git_am_opt:
if interactive:
if incompatible_opts:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
if incompatible_opts
show_error_about_merge_and_am_incompatibilities
which was a triply nested if, and the first (git_am_opt) and third (incompatible_opts) were slightly redundant; the latter being a subset of the former. Perhaps I should have done a separate cleanup patch before this that changed it to:
if incomptable_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
show_error_about_merge_and_am_incompatibilities
Before then having this patch coalesce that down to:
if incomptable_opts:
if interactive:
show_error_about_incompatible_options
Since it tripped up both you and Phillip, I'll add this preliminary cleanup as a separate commit with accompanying explanation in my next re-roll.
Quoted 42 lines
Quoted 10 lines
Documentation/git-rebase.txt +1
-1@@ -672,7 +664,7 @@ require_clean_work_tree "rebase" "$(gettext "Please commit or stash them.")" # but this should be done only when upstream and onto are the same # and if this is not an interactive rebase. mb=$(git merge-base "$onto" "$orig_head") -if test -z "$interactive_rebase" && test "$upstream" = "$onto" && +if test -z "$actually_interactive" && test "$upstream" = "$onto" && test "$mb" = "$onto" && test -z "$restrict_revision" && # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/nullDocumentation/git-rebase.txt then · line 714
Quoted 7 lines
Documentation/git-rebase.txt +4
-0@@ -716,6 +708,19 @@ then GIT_PAGER='' git diff --stat --summary "$mb" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head" +then + # If the $onto is a proper descendant of the tip of the branch, then + # we just fast-forwarded.This comment is misleading if it comes before, rather than after, the actual
checkout -qcommand.Quoted 13 lines
Documentation/git-rebase.txt +9
-0+ say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" + GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ + git checkout -q "$onto^0" || die "could not detach HEAD" + git update-ref ORIG_HEAD $orig_head + move_to_original_branch + finish_rebase + exit 0 +fi + test -n "$interactive_rebase" && run_specific_rebase # Detach HEAD and reset the treeDocumentation/git-rebase.txt GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ · line 731
Quoted 2 lines
Documentation/git-rebase.txt +0
-0@@ -725,16 +730,6 @@ GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \ git checkout -q "$onto^0" || die "could not detach HEAD" git update-ref ORIG_HEAD $orig_headIt is a pity that this hunk header hides the lines that you copied into the following conditional before moving it before the "if interactive, run it" line:
Quoted 1 line
Quoted 16 lines
Documentation/git-rebase.txt
Quoted 13 lines
Documentation/git-rebase.txt +0
-10-# If the $onto is a proper descendant of the tip of the branch, then -# we just fast-forwarded. -if test "$mb" = "$orig_head" -then - say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")" - move_to_original_branch - finish_rebase - exit 0 -fi - if test -n "$rebase_root" then revisions="$onto..$orig_head"What you did is correct, if duplicating code, and definitely the easiest way to do it. Just move the comment, and we're good here.
Will do.
Quoted 6 lines
git-rebase--merge.sh
Quoted 4 lines
git-rebase--merge.sh deleted +0
-0Nice. Really nice!
:-)
Quoted 19 lines
Quoted 5 lines
t/t3406-rebase-message.sh +0
-0t/t3406-rebase-message.sh test_expect_success 'setup' ' · line 25
Quoted 9 lines
t/t3406-rebase-message.sh +0
-6@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF -As I had mentioned in the previous round: I think we can retain these messages for the
--mergemode. And I think we should: users of --merge have most likely become to expect them.
Quoted 8 lines
The most elegant way would probably be by adding code to the sequencer that outputs these lines if in "merge" mode (and add a flag to say that we *are* in "merge" mode).
To that end, the
make_script()function would not generate# pickbutdroplines, I think, so that the sequencer can see those and print theAlready applied: <message>lines. And a successfulTODO_PICKwould write outCommitted: <message>.
I'll take a look, but is this a case where you want it only when rebase previously showed them, or should it also affect other cases such as --rebase-merges or --exec or general --interactive?
I'm hoping the latter, or that there's a good UI-level explanation for why certain rebase flags would actually mean that users should expect different output. rebase seems to be loaded with cases where slight variations on options yield very different side output and perhaps even results mostly based on which backend it historically was implemented on top of. Those gratuitous inconsistencies drive this particular user crazy.
Quoted 113 lines
Quoted 113 lines
t/t3406-rebase-message.sh +1
-0test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actualt/t3420-rebase-autostash.sh +6
-72@@ -53,41 +53,6 @@ create_expected_success_interactive () { EOF } -create_expected_success_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applied autostash. - EOF -} - create_expected_failure_am () { cat >expected <<-EOF $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual)@@ -112,43 +77,6 @@ create_expected_failure_interactive () { EOF } -create_expected_failure_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applying autostash resulted in conflicts. - Your changes are safe in the stash. - You can run "git stash pop" or "git stash drop" at any time. - EOF -} - testrebase () { type=$1 dotest=$2@@ -177,6 +105,9 @@ testrebase () { test_expect_success "rebase$type --autostash: check output" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_success_$suffix && test_i18ncmp expected actual '@@ -274,6 +205,9 @@ testrebase () { test_expect_success "rebase$type: check output with conflicting stash" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi &&
Quoted 8 lines
t/t3420-rebase-autostash.sh testrebase () { · line 213
Quoted 3 lines
t/t3420-rebase-autostash.sh +0
-0create_expected_failure_$suffix && test_i18ncmp expected actual 'As I stated earlier, I am uncomfortable with this solution. We change behavior rather noticeably, and the regression test tells us the same. We need to either try to deprecate
git rebase --merge, or we need to at least attempt to recreate the old behavior with the sequencer.
Yes, I understand not wanting to confuse users. But, serious question, are you (unintentionally) enforcing that we continue confusing users? I may be totally misunderstanding you or the problem space -- you know a lot more here than I do -- but to me your comments sound like enforcing bug compatibility over correctness.
As stated in the git-2.19 release notes (relative to my previous series),
"git rebase" behaved slightly differently depending on which one of
the three backends gets used; this has been documented and an effort to make them more uniform has begun.
If we aren't trying to make the behavior more uniform and don't have good UI rationale for things being different, then my motivation to work on the affected aspects of rebase plummets.
Perhaps this is attributable to a difference of worldview we have, or perhaps I'm just missing something totally obvious. I'm hoping it's the latter. But if it helps, here's the angle I viewed it from: These regression tests showed us that even when identical operations were being performed from the user perspective, we got very different results with the only difference being what backend happened to be selected (and the backends had significant overlap in the area they could all handle, as evidenced by the same testcase being run with each). That seems to me like pretty obvious proof that one or more backends was buggy. Said another way, it looks to me like this is another example of regression tests enforcing "one semi-reasonable behavior" instead of "the correct behavior".
I fully admit I don't have good UI rationale for why the output from the interactive backend is correct. It may not be. I also don't have good UI rationale for why the output from the traditional merge backend would be correct. (And it could be that both are wrong.) It'd be better to figure out what is correct and make the code and the tests implement and check for that. However, while I don't know what correct output to show the user is in these cases, I don't see how it's possible that both backends could have previously been correct. If someone could clearly enunciate a reason for different options to yield different output, I'd happily go and implement it. One obvious explanation could be "--interactive implies user-interactivity, which should thus have different output" -- but that doesn't explain the past because we have several interactive_rebase=implied cases which then should have matched --merge's output rather than --interactive's. Absent any such UI rationale, I'd rather at least make all the backends implement the same semi-reasonable behavior so it's at least consistent. Then when someone figures out what correct is, they can fix it all in one place.
I'm curious to hear if someone has a good reason for which messages should be preferred and when. I'm also curious if perhaps people think that I'm focusing too much on consistency and not enough on "backward compatibility".
Quoted 40 lines
Quoted 40 lines
t/t3421-rebase-topology-linear.sh +4
-4@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase success -p
Quoted 11 lines
t/t3421-rebase-topology-linear.sh test_run_rebase () { · line 321
Quoted 9 lines
t/t3421-rebase-topology-linear.sh +1
-1@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_run_rebase failure -pNice.
:-)
Quoted 52 lines
Quoted 23 lines
t/t3425-rebase-topology-merges.sh +2
-2@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {t/t3425-rebase-topology-merges.sh test_run_rebase () { · line 112
Quoted 8 lines
t/t3425-rebase-topology-merges.sh +1
-1@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "This is a bit unfortunate. I wonder if we could do something like this, to retain the current behavior:
-- snip --
sequencer.c +2
-2@@ -4394,7 +4394,8 @@ int sequencer_make_script(FILE *out, int argc, const char **argv, revs.reverse =1 ; revs.right_only =1 ; revs.sort_order =REV_SORT_IN_GRAPH_ORDER ; - revs.topo_order =1 ; +if (!(flags &TODO_LIST_NO_TOPO_ORDER )) + revs.topo_order =1 ; revs.pretty_given =1 ;git_config_get_string ("rebase.instructionFormat" , &format); -- snap -
Quoted 1 line
(and then pass
TODO_LIST_NO_TOPO_ORDERif in "merge" mode)?
Oh, sweet, we can make these consistent? I tried to look at this a little bit didn't dig too far. I love it; I'll definitely give it a try.
...but...why only in merge mode?? am, interactive, and merge backends gave different results, and the regression test even pointed out that whoever wrote these tests thought it was a bug (the several TODO comments). You knew where to fix it, but want to match previous behavior instead of fixing the inconsistency?
<Snipped a couple things that we already discussed since they were brought up in the commit message>
Quoted 2 lines
Thank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Thanks for the detailed review and the pointers. Much appreciated. We may still have a disconnect of some sort (backwards/bug compatibility vs. correctness/consistency), or perhaps I'm just misunderstanding something for part of the review. However, I can definitely get to work on the other parts of your review comments, and hopefully we can talk through the other bits.
Quoted 2 lines
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
So, yesterday I pointed you at my rebase-new-default branch, but since it has other cruft as well, I just pushed this to a new branch with just the stuff in this patch series. I'll keep it up at
https://github.com/newren/git/tree/rebase-merge-on-sequencer
Thanks!
Elijah
Re: [PATCH v2 2/2] rebase: Implement --merge via git-rebase--interactive
Hi Elijah,
Quoted 10 lines · Elijah Newren
Quoted 7 lines · Johannes Schindelin
Quoted 4 lines
t3425: topology linearization was inconsistent across flavors of rebase,
as already noted in a TODO comment in the testcase. This was not considered a bug before, so getting a different linearization due to switching out backends should not be considered a bug now.Ideally, the test would be fixed, then. If it fails for other reasons than a real regression, it is not a very good regression test, is it.
Quoted 6 lines
I agree, it's not the best regression test. It'd be better if it defined and tested for "the correct behavior", but I suspect it has some value in that it's is guaranteeing that the rebase flavors at least give a "somewhat reasonable" result. Sadly, It just allowed all three rebase types to have slightly different "somewhat reasonable" answers.
True. I hope, though, that we can address this relatively easily (by toggling the topo_order flag).
Quoted 19 lines
Quoted 13 lines
Quoted 4 lines
t5407: different rebase types varied slightly in how many times checkout
or commit or equivalents were called based on a quick comparison of this tests and previous ones which covered different rebase flavors. I think this is just attributable to this difference.This concerns me.
In bigger repositories (no, I am not talking about Linux kernel sized ones, I consider those small-ish) there are a ton of files, and checkout and commit (and even more so reset) sadly do not have a runtime complexity growing with the number of modified files, but with the number of tracked files (and some commands even with the number of worktree files).
Quoted 5 lines
So a larger number of commit/checkout operations makes me expect performance regressions.
In this light, could you elaborate a bit on the differences you see between rebase -i and rebase -m?
Quoted 14 lines
I wrote this comment months ago and don't remember the full details. From the wording and as best I remember, I suspect I was at least partially annoyed by the fact that these and other regression tests for rebase seemed to not be testing for "correct" behavior, but "existing" behavior. That wouldn't be so bad, except that existing behavior differed on the exact same test setup for different rebase backends and the differences in behavior were checked and enforced in the regression tests. So, it became a matter of taste as to how much things should be made identical and to which backend, or whether it was more important to at least just consolidate the code first and keep the results at least reasonable. At the time, I figured getting fewer backends, all with reasonable behavior was a bit more important, but since this difference is worrisome to you, I will try to dig further into it.
I agree that there is a ton of inconsistency going on, both in the behavior of the different backends as well as in the tests and their quality.
The best explanation I have for this is: it grew historically, and we always have to strike a balance between strict rule enforcement and fun.
Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.We should be able to make that test pass, still, by writing out a special file (e.g. $state_dir/opt_m) and testing for that. Users are oddly upset when their expectations are broken... (and I actually agree with them.)
Quoted 8 lines
I agree users are upset when expectations are broken, but why would they expect REBASE-m? In fact, I thought the prompt was a reflection of what backend was in use, so that if someone reported an issue to the git mailing list or a power user wanted to know where the control files were located, they could look for them. As such, I thought that switching the prompt to REBASE-i was the right thing to do so that it would match user expectations. Yes, the backend changed; that's part of the release notes.
When you put it that way, I have to agree with you.
Quoted 2 lines
Is there documentation somewhere that specifies the meaning of this prompt differently than the expectations I had somehow built up?
I was just looking from my perspective, with my experience: if I was a heavy user of git rebase -m (I am not), I would stumble over this prompt. That's all.
With your explanation, I agree that this is the best course, though.
Quoted 20 lines
Quoted 8 lines
Quoted 8 lines
Documentation/git-rebase.txt +0
-0@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS --------------------Quoted 11 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 514
Quoted 8 lines
Documentation/git-rebase.txt +0
-8-git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) -Could we retain this part, with
s/three/two/and/git-merge/d? *Maybe*s/interactive backend/interactive\/merge backend/
Quoted 5 lines
Removing this was actually a suggestion by Phillip back in June at the end of https://public-inbox.org/git/[email protected]/,
for the purpose of removing an implementation details that users don't need to know about. As noted elsewhere in the thread, between you and Phillip, I'll add some comments to the commit message about this.
Right. It is not exactly a democracy over here, but I do agree that both of your opinions combined weigh more than my single one. Besides, you have a good point: the documentation is not so much about being technically correct, but about being helpful to the users. And the new version is probably more helpful than the old version.
Quoted 21 lines
Quoted 9 lines
Documentation/git-rebase.txt See also INCOMPATIBLE OPTIONS below. · line 508
Quoted 6 lines
Documentation/git-rebase.txt +2
-4-if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') - if test -n "$interactive_rebase" +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')Why are we no longer guarding this behind the condition that the user specified *any* option intended for the
ambackend?The code is still correctly guarding, the diff is just confusing. Sorry about that. To explain, the code previously was basically:
if
git_am_opt:
if interactive:
ifincompatible_opts:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
ifincompatible_opts
show_error_about_merge_and_am_incompatibilities
Quoted 3 lines
which was a triply nested if, and the first (
git_am_opt) and third (incompatible_opts) were slightly redundant; the latter being a subset of the former.
Ah, that's what I missed! Thank you for your explanation.
Quoted 11 lines
Perhaps I should have done a separate cleanup patch before this that changed it to:
if
incomptable_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
show_error_about_merge_and_am_incompatibilitiesBefore then having this patch coalesce that down to:
Quoted 7 lines
if
incomptable_opts:
if interactive:
show_error_about_incompatible_optionsSince it tripped up both you and Phillip, I'll add this preliminary cleanup as a separate commit with accompanying explanation in my next re-roll.
Thank you, much appreciated.
Quoted 23 lines
Quoted 23 lines
Quoted 5 lines
t/t3406-rebase-message.sh +0
-0t/t3406-rebase-message.sh test_expect_success 'setup' ' · line 25
Quoted 9 lines
t/t3406-rebase-message.sh +0
-6@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF -As I had mentioned in the previous round: I think we can retain these messages for the
--mergemode. And I think we should: users of --merge have most likely become to expect them.The most elegant way would probably be by adding code to the sequencer that outputs these lines if in "merge" mode (and add a flag to say that we *are* in "merge" mode).
Quoted 8 lines
Quoted 4 lines
To that end, the
make_script()function would not generate# pickbutdroplines, I think, so that the sequencer can see those and print theAlready applied: <message>lines. And a successfulTODO_PICKwould write outCommitted: <message>.I'll take a look, but is this a case where you want it only when rebase previously showed them, or should it also affect other cases such as --rebase-merges or --exec or general --interactive?
Oh, I was just interested in retaining the previous behavior. So only for --merge, not for -r, --exec or -i.
Quoted 3 lines
I'm hoping the latter, or that there's a good UI-level explanation for why certain rebase flags would actually mean that users should expect different output.
Hmm. You are right. This *is* an inconsistency, and insisting on perpetuating it is maybe not the best idea.
Quoted 4 lines
rebase seems to be loaded with cases where slight variations on options yield very different side output and perhaps even results mostly based on which backend it historically was implemented on top of. Those gratuitous inconsistencies drive this particular user crazy.
If you put it that way, we can sell this change of behavior as an
improvement: we are *aligning* the output of git rebase --merge with the output of git rebase --interactive.
I am starting to like it.
Quoted 122 lines
Quoted 113 lines
Quoted 113 lines
t/t3406-rebase-message.sh +1
-0test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actualt/t3420-rebase-autostash.sh +6
-72@@ -53,41 +53,6 @@ create_expected_success_interactive () { EOF } -create_expected_success_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applied autostash. - EOF -} - create_expected_failure_am () { cat >expected <<-EOF $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual)@@ -112,43 +77,6 @@ create_expected_failure_interactive () { EOF } -create_expected_failure_merge () { - cat >expected <<-EOF - $(grep "^Created autostash: [0-9a-f][0-9a-f]*\$" actual) - HEAD is now at $(git rev-parse --short feature-branch) third commit - First, rewinding head to replay your work on top of it... - Merging unrelated-onto-branch with HEAD~1 - Merging: - $(git rev-parse --short unrelated-onto-branch) unrelated commit - $(git rev-parse --short feature-branch^) second commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~2) initial commit - [detached HEAD $(git rev-parse --short rebased-feature-branch~1)] second commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:14:13 2005 -0700 - 2 files changed, 2 insertions(+) - create mode 100644 file1 - create mode 100644 file2 - Committed: 0001 second commit - Merging unrelated-onto-branch with HEAD~0 - Merging: - $(git rev-parse --short rebased-feature-branch~1) second commit - $(git rev-parse --short feature-branch) third commit - found 1 common ancestor: - $(git rev-parse --short feature-branch~1) second commit - [detached HEAD $(git rev-parse --short rebased-feature-branch)] third commit - Author: A U Thor <[email protected]> - Date: Thu Apr 7 15:15:13 2005 -0700 - 1 file changed, 1 insertion(+) - create mode 100644 file3 - Committed: 0002 third commit - All done. - Applying autostash resulted in conflicts. - Your changes are safe in the stash. - You can run "git stash pop" or "git stash drop" at any time. - EOF -} - testrebase () { type=$1 dotest=$2@@ -177,6 +105,9 @@ testrebase () { test_expect_success "rebase$type --autostash: check output" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi && create_expected_success_$suffix && test_i18ncmp expected actual '@@ -274,6 +205,9 @@ testrebase () { test_expect_success "rebase$type: check output with conflicting stash" ' test_when_finished git branch -D rebased-feature-branch && suffix=${type#\ --} && suffix=${suffix:-am} && + if test ${suffix} = "merge"; then + suffix=interactive + fi &&Quoted 8 lines
t/t3420-rebase-autostash.sh testrebase () { · line 213
Quoted 3 lines
t/t3420-rebase-autostash.sh +0
-0create_expected_failure_$suffix && test_i18ncmp expected actual 'As I stated earlier, I am uncomfortable with this solution. We change behavior rather noticeably, and the regression test tells us the same. We need to either try to deprecate
git rebase --merge, or we need to at least attempt to recreate the old behavior with the sequencer.
Quoted 5 lines
Yes, I understand not wanting to confuse users. But, serious question, are you (unintentionally) enforcing that we continue confusing users? I may be totally misunderstanding you or the problem space -- you know a lot more here than I do -- but to me your comments sound like enforcing bug compatibility over correctness.
I am actually a really inappropriate person to make judgement calls on git rebase --merge, as I am neither a user of it nor do I know of any. I don't think I ever read any account of a Git for Windows user who called git rebase --merge...
Quoted 41 lines
As stated in the git-2.19 release notes (relative to my previous series),
"git rebase" behaved slightly differently depending on which one of
the three backends gets used; this has been documented and an effort to make them more uniform has begun.
If we aren't trying to make the behavior more uniform and don't have good UI rationale for things being different, then my motivation to work on the affected aspects of rebase plummets.Perhaps this is attributable to a difference of worldview we have, or perhaps I'm just missing something totally obvious. I'm hoping it's the latter. But if it helps, here's the angle I viewed it from: These regression tests showed us that even when identical operations were being performed from the user perspective, we got very different results with the only difference being what backend happened to be selected (and the backends had significant overlap in the area they could all handle, as evidenced by the same testcase being run with each). That seems to me like pretty obvious proof that one or more backends was buggy. Said another way, it looks to me like this is another example of regression tests enforcing "one semi-reasonable behavior" instead of "the correct behavior".
I fully admit I don't have good UI rationale for why the output from the interactive backend is correct. It may not be. I also don't have good UI rationale for why the output from the traditional merge backend would be correct. (And it could be that both are wrong.) It'd be better to figure out what is correct and make the code and the tests implement and check for that. However, while I don't know what correct output to show the user is in these cases, I don't see how it's possible that both backends could have previously been correct. If someone could clearly enunciate a reason for different options to yield different output, I'd happily go and implement it. One obvious explanation could be "--interactive implies user-interactivity, which should thus have different output" -- but that doesn't explain the past because we have several interactive_rebase=implied cases which then should have matched --merge's output rather than --interactive's. Absent any such UI rationale, I'd rather at least make all the backends implement the same semi-reasonable behavior so it's at least consistent. Then when someone figures out what correct is, they can fix it all in one place.
Quoted 4 lines
I'm curious to hear if someone has a good reason for which messages should be preferred and when. I'm also curious if perhaps people think that I'm focusing too much on consistency and not enough on "backward compatibility".
It is good to have your perspective in addition to mine. As na open source maintainer, I gravitate toward aiming for maximal backwards-compatibility out of purely selfish reasons: I don't want to deal with so many people complaining about changes in the software I maintain.
But you raised a good point: relying on behavior no matter how undesirable said behavior is not doing users a big favor. It is better to face the challenge of fixing that behavior.
Quoted 58 lines
Quoted 52 lines
Quoted 23 lines
t/t3425-rebase-topology-merges.sh +2
-2@@ -72,7 +72,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {@@ -89,7 +89,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {t/t3425-rebase-topology-merges.sh test_run_rebase () { · line 112
Quoted 8 lines
t/t3425-rebase-topology-merges.sh +1
-1@@ -106,7 +106,7 @@ test_run_rebase () { } #TODO: make order consistent across all flavors of rebase test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_expect_success "rebase -p is no-op in non-linear history" "This is a bit unfortunate. I wonder if we could do something like this, to retain the current behavior:
-- snip --
sequencer.c +2
-2@@ -4394,7 +4394,8 @@ int sequencer_make_script(FILE *out, int argc, const char **argv, revs.reverse =1 ; revs.right_only =1 ; revs.sort_order =REV_SORT_IN_GRAPH_ORDER ; - revs.topo_order =1 ; +if (!(flags &TODO_LIST_NO_TOPO_ORDER )) + revs.topo_order =1 ; revs.pretty_given =1 ;git_config_get_string ("rebase.instructionFormat" , &format); -- snap -Quoted 1 line
(and then pass
TODO_LIST_NO_TOPO_ORDERif in "merge" mode)?Oh, sweet, we can make these consistent? I tried to look at this a little bit didn't dig too far. I love it; I'll definitely give it a try.
Quoted 5 lines
...but...why only in merge mode?? am, interactive, and merge backends gave different results, and the regression test even pointed out that whoever wrote these tests thought it was a bug (the several TODO comments). You knew where to fix it, but want to match previous behavior instead of fixing the inconsistency?
I am afraid that we *have* to use topo_order for -r, but you know, I could be convinced that it is a good idea to switch away from topo order for regular -i.
Quoted 6 lines
<Snipped a couple things that we already discussed since they were brought up in the commit message>
Quoted 2 lines
Thank you for this pleasant read. I think there is still quite a bit of work to do, but you already did most of it.
Quoted 6 lines
Thanks for the detailed review and the pointers. Much appreciated. We may still have a disconnect of some sort (backwards/bug compatibility vs. correctness/consistency), or perhaps I'm just misunderstanding something for part of the review. However, I can definitely get to work on the other parts of your review comments, and hopefully we can talk through the other bits.
I think it will get easier when other people chime in, so that we do not go back and forth between our two opinions but instead get a broader context in which to make decisions, so that we will hopefully end up with a fine balance between consistency/correctness and backwards-compatibility.
Quoted 8 lines
Quoted 2 lines
Out of curiosity, do you have a public repository with these patches in a branch? (I might have an hour to play with it tonight...)
So, yesterday I pointed you at my rebase-new-default branch, but since it has other cruft as well, I just pushed this to a new branch with just the stuff in this patch series. I'll keep it up at
https://github.com/newren/git/tree/rebase-merge-on-sequencer
Thank you!
I had originally planned on seeing how difficult it would be to reinstate those messages of rebase -m, but in the meantime I get the impression that I may not even want those to survive the merge of the --merge mode into --interactive.
Ciao,
Dscho
Quoted 10 lines · Elijah Newren
Now that the rewrite-interactive-rebases-in-C series have finally merged to master, this series deletes
git-rebase--merge.shand reimplements the --merge behavior on top of the interactive machinery.Differences since v1:
- Updated code to hook into builtin/rebase.C instead ofgit-rebase.sh(No range-diff provided, because it has been months since v1, and v1 was only RFC and was only discussed at a high level.)
Actually, that's not correct; it's been so long that I forgot. Dscho and Phillip both reviewed it and I updated my series at the time with their suggestions, but didn't re-submit because it depended on so many other series and conflicted with the rebase-in-C work. So other differences include the changes I made to address their feedback, but...even if I dug up the old v1 and created a range-diff against it I'm not so sure it'd be helpful. If anyone thinks it would be, holler and I'll generate it.
Original series here:
https://public-inbox.org/git/[email protected]/
v3 cover Reimplement rebase --merge via interactive machinery
In reply to Elijah Newren[Important: Patch 1 fixes a (minor) regression in 2.20 relative to 2.19...but it also modifies a translated string. I'm not sure what the right step we want to take there is. If you want me to submit it separately and also resubmit the rest of this series to depend on the separated first patch, let me know.]
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though two differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Differences since v2 (full range-diff below):
- Addressed feedback from both Phillip and Dscho
- Added five new patches; the biggest change is still the final patch but it should be a little easier to review now.
- Added a patch fixing a very recent regression in the incompatible options error message
- Documented and fixed the inconsistency in how different rebase backends handle --skip relative to the post-rewrite hook
- Added a patch defining the linearization order and enforcing it, getting rid of an age-old TODO
- Added a patch which took a slightly confusing diff hunk from the previous final patch, and give it its own commit message explaining why the drop from a triply-nested if block to a doubly-nested if-block still keeps all necessary checks in place.
- Rebased to latest master
Elijah Newren (7):
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery
.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/am.c | 9 ++ builtin/rebase.c | 24 ++--- git-legacy-rebase.sh | 57 +++++------ git-rebase--am.sh | 2 +- git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 15 ++- t/t5407-post-rewrite-hook.sh | 34 +++++++ t/t9903-bash-prompt.sh | 2 +- 17 files changed, 114 insertions(+), 333 deletions(-) delete mode 100644 git-rebase--merge.sh
-: ---------- > 1: 2f4bdd1980 rebase: fix incompatible options error message -: ---------- > 2: cc33a8ccc1 t5407: add a test demonstrating how interactive handles --skip differently -: ---------- > 3: f5838ef763 am, rebase--merge: do not overlook --skip'ed commits with post-rewrite 1: bf0acd9b27 ! 4: 50dc863d9f git-rebase, sequencer: extend --quiet option for the interactive machinery
@@ -13,7 +13,7 @@
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few
- informational messages -- "Rebasing (n/m)" and "Succesfully rebased..."
+ informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a
-: ---------- > 5: 35cf552f27 git-legacy-rebase: simplify unnecessary triply-nested if -: ---------- > 6: 2a3d8ff1c1 rebase: define linearization ordering and enforce it 2: cd0ccab680 ! 7: 58371d377a rebase: Implement --merge via git-rebase--interactive
@@ -1,35 +1,37 @@
Author: Elijah Newren <[email protected]>
- rebase: Implement --merge via git-rebase--interactive
+ rebase: Implement --merge via the interactive machinery
- Interactive rebases are implemented in terms of cherry-pick rather than
- the merge-recursive builtin, but cherry-pick also calls into the recursive
- merge machinery by default and can accept special merge strategies and/or
- special strategy options. As such, there really is not any need for
- having both git-rebase--merge and git-rebase--interactive anymore.
+ As part of an ongoing effort to make rebase have more uniform behavior,
+ modify the merge backend to behave like the interactive one, by
+ re-implementing it on top of the latter.
- Delete git-rebase--merge.sh and have the --merge option be implemented
- by the now built-in interactive machinery.
+ Interactive rebases are implemented in terms of cherry-pick rather than
+ the merge-recursive builtin, but cherry-pick also calls into the
+ recursive merge machinery by default and can accept special merge
+ strategies and/or special strategy options. As such, there really is
+ not any need for having both git-rebase--merge and
+ git-rebase--interactive anymore. Delete git-rebase--merge.sh and
+ instead implement it in builtin/rebase.c.
- Note that this change fixes a few known test failures (see t3421).
+ This results in a few deliberate but small user-visible changes:
+ * The progress output is modified (see t3406 and t3420 for examples)
+ * A few known test failures are now fixed (see t3421)
+ * bash-prompt during a rebase --merge is now REBASE-i instead of
+ REBASE-m. Reason: The prompt is a reflection of the backend in use;
+ this allows users to report an issue to the git mailing list with
+ the appropriate backend information, and allows advanced users to
+ know where to search for relevant control files. (see t9903)
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
- while running; adjust a test to match
+ while running; adjust a test to match the new expectation
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
- t3425: topology linearization was inconsistent across flavors of rebase,
- as already noted in a TODO comment in the testcase. This was not
- considered a bug before, so getting a different linearization due
- to switching out backends should not be considered a bug now.
- t5407: different rebase types varied slightly in how many times checkout
- or commit or equivalents were called based on a quick comparison
- of this tests and previous ones which covered different rebase
- flavors. I think this is just attributable to this difference.
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
@@ -133,13 +135,24 @@
}
}
-+ if (options.type == REBASE_MERGE) {
++ if (options.type == REBASE_MERGE)
+ imply_interactive(&options, "--merge");
-+ }
+
if (options.root && !options.onto_name)
imply_interactive(&options, "--root without --onto");
+@@
+
+ if (is_interactive(&options) && i >= 0)
+ die(_("error: cannot combine am options "
+- "with interactive options"));
+- if (options.type == REBASE_MERGE && i >= 0)
+- die(_("error: cannot combine am options "
+- "with merge options "));
++ "with either interactive or merge options"));
+ }
+
+ if (options.signoff) {
diff --git a/git-legacy-rebase.sh b/git-legacy-rebase.sh --- a/git-legacy-rebase.sh
@@ -187,30 +200,18 @@
else
type=am
@@
- git_format_patch_opt="$git_format_patch_opt --progress"
- fi
-
--if test -n "$git_am_opt"; then
-- incompatible_opts=$(echo " $git_am_opt " | \
-- sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')
+ sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')
+ if test -n "$incompatible_opts"
+ then
- if test -n "$interactive_rebase"
-+incompatible_opts=$(echo " $git_am_opt " | \
-+ sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/')
-+if test -n "$incompatible_opts"
-+then
+- then
+- die "$(gettext "error: cannot combine am options with interactive options")"
+- fi
+- if test -n "$do_merge"
+ if test -n "$actually_interactive" || test "$do_merge"
then
-- if test -n "$incompatible_opts"
-- then
-- die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")"
-- fi
-- fi
-- if test -n "$do_merge"; then
-- if test -n "$incompatible_opts"
-- then
-- die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")"
-- fi
-+ die "$(gettext "error: cannot combine am options ($incompatible_opts) with either interactive or merge options")"
+- die "$(gettext "error: cannot combine am options with merge options")"
++ die "$(gettext "error: cannot combine am options with either interactive or merge options")"
fi
fi
@@ -229,11 +230,11 @@
+if test -z "$actually_interactive" && test "$mb" = "$orig_head" +then
-+ # If the $onto is a proper descendant of the tip of the branch, then
-+ # we just fast-forwarded.
+ say "$(eval_gettext "Fast-forwarded \$branch_name to \$onto_name.")"
+ GIT_REFLOG_ACTION="$GIT_REFLOG_ACTION: checkout $onto_name" \
+ git checkout -q "$onto^0" || die "could not detach HEAD"
++ # If the $onto is a proper descendant of the tip of the branch, then
++ # we just fast-forwarded.
+ git update-ref ORIG_HEAD $orig_head
+ move_to_original_branch
+ finish_rebase
@@ -389,6 +390,8 @@
-skip)
- read_state
- git rerere clear
+- cmt="$(cat "$state_dir/cmt.$msgnum")"
+- echo "$cmt $(git rev-parse HEAD^0)" >> "$state_dir/rewritten"
- msgnum=$(($msgnum + 1))
- while test "$msgnum" -le "$end"
- do
@@ -409,7 +412,7 @@
-rm -f "$(git rev-parse --git-path REBASE_HEAD)"
-
-msgnum=0
--for cmt in $(git rev-list --reverse --no-merges "$revisions") +-for cmt in $(git rev-list --topo-order --reverse --no-merges "$revisions")
-do
- msgnum=$(($msgnum + 1))
- echo "$cmt" > "$state_dir/cmt.$msgnum"
@@ -571,7 +574,7 @@
-test_run_rebase failure -m
+test_run_rebase success -m
test_run_rebase success -i
- test_run_rebase success -p
+ test_have_prereq !REBASE_P || test_run_rebase success -p
@@
"
@@ -580,7 +583,7 @@
-test_run_rebase failure -m
+test_run_rebase success -m
test_run_rebase success -i
- test_run_rebase success -p
+ test_have_prereq !REBASE_P || test_run_rebase success -p
@@
"
@@ -589,7 +592,7 @@
-test_run_rebase failure -m
+test_run_rebase success -m
test_run_rebase success -i
- test_run_rebase success -p
+ test_have_prereq !REBASE_P || test_run_rebase success -p
@@
"
@@ -598,7 +601,7 @@
-test_run_rebase failure -m
+test_run_rebase success -m
test_run_rebase success -i
- test_run_rebase success -p
+ test_have_prereq !REBASE_P || test_run_rebase success -p
@@
"
@@ -607,52 +610,9 @@
-test_run_rebase failure -m
+test_run_rebase success -m
test_run_rebase success -i
- test_run_rebase failure -p
+ test_have_prereq !REBASE_P || test_run_rebase failure -p
- diff --git a/t/t3425-rebase-topology-merges.sh b/t/t3425-rebase-topology-merges.sh
- --- a/t/t3425-rebase-topology-merges.sh
- +++ b/t/t3425-rebase-topology-merges.sh
-@@
- }
- #TODO: make order consistent across all flavors of rebase
- test_run_rebase success 'e n o' ''
--test_run_rebase success 'e n o' -m
-+test_run_rebase success 'n o e' -m
- test_run_rebase success 'n o e' -i
-
- test_run_rebase () {
-@@
- }
- #TODO: make order consistent across all flavors of rebase
- test_run_rebase success 'd e n o' ''
--test_run_rebase success 'd e n o' -m
-+test_run_rebase success 'd n o e' -m
- test_run_rebase success 'd n o e' -i
-
- test_run_rebase () {
-@@
- }
- #TODO: make order consistent across all flavors of rebase
- test_run_rebase success 'd e n o' ''
--test_run_rebase success 'd e n o' -m
-+test_run_rebase success 'd n o e' -m
- test_run_rebase success 'd n o e' -i
-
- test_expect_success "rebase -p is no-op in non-linear history" "
-
- diff --git a/t/t5407-post-rewrite-hook.sh b/t/t5407-post-rewrite-hook.sh
- --- a/t/t5407-post-rewrite-hook.sh
- +++ b/t/t5407-post-rewrite-hook.sh
-@@
- git rebase --continue &&
- echo rebase >expected.args &&
- cat >expected.data <<-EOF &&
-+ $(git rev-parse C) $(git rev-parse HEAD^)
- $(git rev-parse D) $(git rev-parse HEAD)
- EOF
- verify_hook_input
-
diff --git a/t/t9903-bash-prompt.sh b/t/t9903-bash-prompt.sh
--- a/t/t9903-bash-prompt.sh
+++ b/t/t9903-bash-prompt.sh
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 1/7 rebase: fix incompatible options error message
In commit f57696802c30 ("rebase: really just passthru the git am options", 2018-11-14), the handling of git am options was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:
$ git rebase --keep --ignore-whitespace fatal: error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)
$ git rebase --merge --ignore-whitespace fatal: error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)
Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
--- builtin/rebase.c | 11 ++++------- git-legacy-rebase.sh | 4 ++-- 2 files changed, 6 insertions(+), 9 deletions(-)
builtin/rebase.c +4 -7
@@ -1202,14 +1202,11 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("error: cannot combine interactive options " -"(--interactive, --exec, --rebase-merges, " -"--preserve-merges, --keep-empty, --root + " -"--onto) with am options (%s)" ), buf.buf); +die (_ ("error: cannot combine am options " +"with interactive options" ));if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("error: cannot combine merge options (--merge, " -"--strategy, --strategy-option) with am options " -"(%s)" ), buf.buf); +die (_ ("error: cannot combine am options " +"with merge options " )); }if (options.signoff) {
git-legacy-rebase.sh +2 -2
@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "error: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "error: cannot combine am options with merge options")" fi fi fi
Signature · 1 line
2.20.0.rc1.7.g58371d377a
Re: [PATCH v3 1/7] rebase: fix incompatible options error message
Hi Elijah,
Quoted 22 lines · Elijah Newren
In commit
f57696802c30("rebase: really just passthru thegit amoptions", 2018-11-14), the handling ofgit amoptions was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:$ git rebase --keep --ignore-whitespace fatal: error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)$ git rebase --merge --ignore-whitespace fatal: error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
Quoted 2 lines
Signed-off-by Elijah Newren <[email protected]>---
This patch is obviously good.
Given that you embedded it in the patch series that makes the sequencer the work horse also for the merge backend of git rebase in addition to the interactive one, may I assume that you intend this patch for post v2.20.0?
Ciao,
Dscho
Quoted 51 lines
builtin/rebase.c | 11 ++++------- git-legacy-rebase.sh | 4 ++-- 2 files changed, 6 insertions(+), 9 deletions(-)builtin/rebase.c +4
-7@@ -1202,14 +1202,11 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("error: cannot combine interactive options " -"(--interactive, --exec, --rebase-merges, " -"--preserve-merges, --keep-empty, --root + " -"--onto) with am options (%s)" ), buf.buf); +die (_ ("error: cannot combine am options " +"with interactive options" ));if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("error: cannot combine merge options (--merge, " -"--strategy, --strategy-option) with am options " -"(%s)" ), buf.buf); +die (_ ("error: cannot combine am options " +"with merge options " )); }if (options.signoff) {git-legacy-rebase.sh +2
-2@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "error: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "error: cannot combine am options with merge options")" fi fi fiSignature · 1 line
2.20.0.rc1.7.g58371d377a
Re: [PATCH v3 1/7] rebase: fix incompatible options error message
Quoted 32 lines · Johannes Schindelin
Hi Elijah,
Quoted 22 lines · Elijah Newren
In commit
f57696802c30("rebase: really just passthru thegit amoptions", 2018-11-14), the handling ofgit amoptions was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:$ git rebase --keep --ignore-whitespace fatal: error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)$ git rebase --merge --ignore-whitespace fatal: error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
Quoted 2 lines
Signed-off-by Elijah Newren <[email protected]>---
This patch is obviously good.
Quoted 7 lines
Given that you embedded it in the patch series that makes the sequencer the work horse also for the
mergebackend ofgit rebasein addition to theinteractiveone, may I assume that you intend this patch for post v2.20.0?Ciao,
Dscho
I think post v2.20.0 probably makes the most sense. I was unsure what the policy was around changing strings late in the cycle, but figured that the worst case with this regression is someone basically understands what the message is trying to say but thinks it might be saying more than they understand and reach out with questions. In contrast, if we decide to change the string and some translators don't have enough time to translate it before 2.20.0 goes out, then someone who doesn't understand English gets an English error message, which seems a little worse.
I at least wanted it to be discussed, though, which is why I highlighted it in the cover letter.
Re: [PATCH v3 1/7] rebase: fix incompatible options error message
In reply to Elijah NewrenQuoted 16 lines · Elijah Newren
In commit
f57696802c30("rebase: really just passthru thegit amoptions", 2018-11-14), the handling ofgit amoptions was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:$ git rebase --keep --ignore-whitespace fatal: error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)$ git rebase --merge --ignore-whitespace fatal: error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)
Quoted 6 lines
Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
Can we simplify it further and remove the "error: " prefix? "fatal:
error: " looks redundant.
Signature · 1 line
Duy
Re: [PATCH v3 1/7] rebase: fix incompatible options error message
Quoted 25 lines · Duy Nguyen
Quoted 16 lines · Elijah Newren
In commit
f57696802c30("rebase: really just passthru thegit amoptions", 2018-11-14), the handling ofgit amoptions was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:$ git rebase --keep --ignore-whitespace fatal: error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)$ git rebase --merge --ignore-whitespace fatal: error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)Quoted 6 lines
Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
Quoted 2 lines
Can we simplify it further and remove the "error: " prefix? "fatal:
error: " looks redundant.
Sure, I can do that. Looks like there are a few other cases that need fixing as well:
$ git grep error: builtin/rebase.c
builtin/rebase.c: die(_("error: cannot combine
interactive options "
builtin/rebase.c: die(_("error: cannot combine
merge options (--merge, "
builtin/rebase.c: die(_("error: cannot combine
'--preserve-merges' with "
builtin/rebase.c: die(_("error: cannot combine
'--rebase-merges' with "
builtin/rebase.c: die(_("error: cannot combine
'--rebase-merges' with "Perhaps, for consistency, I should also change the error message in the git-legacy-rebase.sh script to use 'fatal' instead of 'error'?:
$ git grep error: *rebase*.sh git-legacy-rebase.sh: die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" git-legacy-rebase.sh: die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" git-legacy-rebase.sh: die "$(gettext "error: cannot combine '--signoff' with '--preserve-merges'")" git-legacy-rebase.sh: die "$(gettext "error: cannot combine '--preserve-merges' with '--rebase-merges'")" git-legacy-rebase.sh: die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy-option'")" git-legacy-rebase.sh: die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy'")"
PATCH v3 2/7 t5407: add a test demonstrating how interactive handles --skip differently
In reply to Elijah NewrenThe post-rewrite hook is documented as being invoked by commands that rewrite commits such as commit --amend and rebase, and that it will be called for each rewritten commit.
Apparently, the three backends handled --skip'ed commits differently:
am: treat the skipped commit as though it weren't rewritten
merge: same as 'am' backend
interactive: treat skipped commits as having been rewritten to empty
(view them as an empty fixup to their parent)
For now, just add a testcase documenting the different behavior (use --keep to force usage of the interactive machinery even though we have no empty commits). A subsequent commit will remove the inconsistency in --skip handling.
--- t/t5407-post-rewrite-hook.sh | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
t/t5407-post-rewrite-hook.sh +31 -0
@@ -125,6 +125,37 @@ test_expect_success 'git rebase -m --skip' ' verify_hook_input ' +test_expect_success 'git rebase with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + echo C > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + +test_expect_success 'git rebase --skip with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + test_must_fail git rebase --skip && + echo D > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + . "$TEST_DIRECTORY"/lib-rebase.sh set_fake_editor
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 3/7 am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
In reply to Elijah NewrenThe post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)
builtin/am.c +9 -0
@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" )); +if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit)); +fprintf (fp,"%s\n" ,oid_to_hex (&head)); +fclose (fp); + } +am_next (state);am_load (state);am_run (state,0 );
git-rebase--merge.sh +2 -0
@@ -121,6 +121,8 @@ continue) skip) read_state git rerere clear + cmt="$(cat "$state_dir/cmt.$msgnum")" + echo "$cmt $(git rev-parse HEAD^0)" >> "$state_dir/rewritten" msgnum=$(($msgnum + 1)) while test "$msgnum" -le "$end" do
t/t5407-post-rewrite-hook.sh +3 -0
@@ -78,6 +78,7 @@ test_expect_success 'git rebase --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
@@ -91,6 +92,7 @@ test_expect_success 'git rebase --skip the last one' ' echo rebase >expected.args && cat >expected.data <<-EOF && $(git rev-parse E) $(git rev-parse HEAD) + $(git rev-parse F) $(git rev-parse HEAD) EOF verify_hook_input '
@@ -120,6 +122,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 4/7 git-rebase, sequencer: extend --quiet option for the interactive machinery
In reply to Elijah NewrenWhile 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)
builtin/rebase.c +1 -4
@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts))) opts->flags &= ~REBASE_NO_QUIET ;else opts->flags |=REBASE_NO_QUIET ;
git-legacy-rebase.sh +1 -1
@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&
git-rebase--common.sh +1 -1
@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \
sequencer.c +13 -10
@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")
@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {
@@ -2357,6 +2357,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;
@@ -2424,9 +2427,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)
@@ -3503,10 +3503,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());
@@ -3738,8 +3739,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);
sequencer.h +1 -0
@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 5/7 git-legacy-rebase: simplify unnecessary triply-nested if
In reply to Elijah NewrenThe git-legacy-rebase.sh script previously had code of the form:
if git_am_opt:
if interactive:
if incompatible_opts:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
if incompatible_opts
show_error_about_merge_and_am_incompatibilities
which was a triply nested if. However, the first conditional (git_am_opt) and third (incompatible_opts) were somewhat redundant: the latter condition was a strict subset of the former. Simplify this by moving the innermost conditional to the outside, allowing us to remove the test on git_am_opt entirely and giving us the following form:
if incomptable_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
show_error_about_merge_and_am_incompatibilities
--- git-legacy-rebase.sh | 20 ++++++++------------ 1 file changed, 8 insertions(+), 12 deletions(-)
git-legacy-rebase.sh +8 -12
@@ -501,21 +501,17 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fi -if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +if test -n "$incompatible_opts" +then if test -n "$interactive_rebase" then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine am options with interactive options")" - fi + die "$(gettext "error: cannot combine am options with interactive options")" fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then - die "$(gettext "error: cannot combine am options with merge options")" - fi + if test -n "$do_merge" + then + die "$(gettext "error: cannot combine am options with merge options")" fi fi
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 6/7 rebase: define linearization ordering and enforce it
In reply to Elijah NewrenEver since commit 3f213981e44a ("add tests for rebasing merged history", 2013-06-06), t3425 has had tests which included the rebasing of merged history and whose order of applied commits was checked. Unfortunately, the tests expected different behavior depending on which backend was in use. Implementing these checks was the following four lines (including the TODO message) which were repeated verbatim three times in t3425:
#TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' test_run_rebase success 'e n o' -m test_run_rebase success 'n o e' -i
As part of the effort to reduce differences between the rebase backends so that users get more uniform behavior, let's define the correct behavior and modify the different backends so they all get the right answer. It turns out that the difference in behavior here is entirely due to topological sorting; since some backends require topological sorting (particularly when --rebase-merges is specified), require it for all modes. Modify the am and merge backends to implement this.
Performance Considerations:
I was unable to measure any appreciable performance difference with this change. Trying to control the run-to-run variation was difficult; I eventually found a headless beefy box that I could ssh into, which seemed to help. Using git.git, I ran the following testcase:
$ git reset --hard v2.20.0-rc1~2
$ time git rebase --quiet v2.20.0-rc0~16I first ran once to warm any disk caches, then ran five subsequent runs and recorded the times of those five. I observed the following results for the average time:
Before this change:
"real" timing: 1.340s (standard deviation: 0.040s)
"user" timing: 1.050s (standard deviation: 0.041s)
"sys" timing: 0.270s (standard deviation: 0.011s)
After this change:
"real" timing: 1.327s (standard deviation: 0.065s)
"user" timing: 1.031s (standard deviation: 0.061s)
"sys" timing: 0.280s (standard deviation: 0.014s)Measurements aside, I would expect the timing for walking revisions to be dwarfed by the work involved in creating and applying patches, so this isn't too surprising. Further, while somewhat counter-intuitive, it is possible that turning on topological sorting is actually a performance improvement: by way of comparison, turning on --topo-order made fast-export faster (see
https://public-inbox.org/git/[email protected]/).
--- git-rebase--am.sh | 2 +- git-rebase--merge.sh | 2 +- t/t3425-rebase-topology-merges.sh | 15 ++++++--------- 3 files changed, 8 insertions(+), 11 deletions(-)
git-rebase--am.sh +1 -1
@@ -36,7 +36,7 @@ rm -f "$GIT_DIR/rebased-patches" git format-patch -k --stdout --full-index --cherry-pick --right-only \ --src-prefix=a/ --dst-prefix=b/ --no-renames --no-cover-letter \ - --pretty=mboxrd \ + --pretty=mboxrd --topo-order \ $git_format_patch_opt \ "$revisions" ${restrict_revision+^$restrict_revision} \ >"$GIT_DIR/rebased-patches"
git-rebase--merge.sh +1 -1
@@ -143,7 +143,7 @@ write_basic_state rm -f "$(git rev-parse --git-path REBASE_HEAD)" msgnum=0 -for cmt in $(git rev-list --reverse --no-merges "$revisions") +for cmt in $(git rev-list --topo-order --reverse --no-merges "$revisions") do msgnum=$(($msgnum + 1)) echo "$cmt" > "$state_dir/cmt.$msgnum"
t/t3425-rebase-topology-merges.sh +6 -9
@@ -70,9 +70,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" d.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' '' +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {
@@ -87,9 +86,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {
@@ -104,9 +102,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i if ! test_have_prereq REBASE_P; then
Signature · 1 line
2.20.0.rc1.7.g58371d377a
PATCH v3 7/7 rebase: Implement --merge via the interactive machinery
In reply to Elijah NewrenAs part of an ongoing effort to make rebase have more uniform behavior, modify the merge backend to behave like the interactive one, by re-implementing it on top of the latter.
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore. Delete git-rebase--merge.sh and instead implement it in builtin/rebase.c.
This results in a few deliberate but small user-visible changes:
* The progress output is modified (see t3406 and t3420 for examples)
* A few known test failures are now fixed (see t3421)
* bash-prompt during a rebase --merge is now REBASE-i instead of REBASE-m. Reason: The prompt is a reflection of the backend in use; this allows users to report an issue to the git mailing list with the appropriate backend information, and allows advanced users to know where to search for relevant control files. (see t9903)
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match the new expectation
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
--- .gitignore | 1 - Documentation/git-rebase.txt | 17 +-- Makefile | 1 - builtin/rebase.c | 14 +-- git-legacy-rebase.sh | 43 ++++---- git-rebase--merge.sh | 166 ------------------------------ t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t9903-bash-prompt.sh | 2 +- 10 files changed, 42 insertions(+), 297 deletions(-) delete mode 100644 git-rebase--merge.sh
.gitignore +0 -1
@@ -124,7 +124,6 @@ /git-rebase--am /git-rebase--common /git-rebase--interactive -/git-rebase--merge /git-rebase--preserve-merges /git-receive-pack /git-reflog
Documentation/git-rebase.txt +3 -14
@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS -------------------- -git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) - -Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date
@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options: * --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: - * --[no-]autosquash * --rebase-merges * --preserve-merges
@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoff
Makefile +0 -1
@@ -628,7 +628,6 @@ SCRIPT_LIB += git-parse-remote SCRIPT_LIB += git-rebase--am SCRIPT_LIB += git-rebase--common SCRIPT_LIB += git-rebase--preserve-merges -SCRIPT_LIB += git-rebase--merge SCRIPT_LIB += git-sh-setup SCRIPT_LIB += git-sh-i18n
builtin/rebase.c +5 -9
@@ -122,7 +122,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;
@@ -481,10 +481,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;
@@ -1170,6 +1166,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) +imply_interactive (&options,"--merge" ); +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );
@@ -1200,10 +1199,7 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)if (is_interactive (&options) && i >=0 )die (_ ("error: cannot combine am options " -"with interactive options" )); -if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("error: cannot combine am options " -"with merge options " )); +"with either interactive or merge options" )); }if (options.signoff) {
git-legacy-rebase.sh +21 -22
96 lines not on this page · raw message
git-rebase--merge.sh deleted +0 -166
166 lines not on this page · raw message
t/t3406-rebase-message.sh +1 -6
@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF - test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actual
t/t3420-rebase-autostash.sh +6 -72
102 lines not on this page · raw message
t/t3421-rebase-topology-linear.sh +5 -5
@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase failure -p
t/t9903-bash-prompt.sh +1 -1
@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected && git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&
Signature · 1 line
2.20.0.rc1.7.g58371d377a
v4 cover Reimplement rebase --merge via interactive machinery
In reply to Elijah NewrenThis series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0
Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery
.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/am.c | 9 ++ builtin/rebase.c | 30 ++---- git-legacy-rebase.sh | 65 ++++++------ git-rebase--am.sh | 2 +- git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 15 ++- t/t5407-post-rewrite-hook.sh | 34 +++++++ t/t9903-bash-prompt.sh | 2 +- 17 files changed, 121 insertions(+), 340 deletions(-) delete mode 100644 git-rebase--merge.sh
Range-diff:
-: ---------- > 1: 2e8b1bcb8b rebase: make builtin and legacy script error messages the same
1: 2f4bdd1980 ! 2: eba87828c6 rebase: fix incompatible options error message
@@ -9,12 +9,12 @@
understood by separate backends were used:
$ git rebase --keep --ignore-whitespace
- fatal: error: cannot combine interactive options (--interactive, --exec,
+ fatal: cannot combine interactive options (--interactive, --exec,
--rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with
am options (.git/rebase-apply/applying)
$ git rebase --merge --ignore-whitespace
- fatal: error: cannot combine merge options (--merge, --strategy,
+ fatal: cannot combine merge options (--merge, --strategy,
--strategy-option) with am options (.git/rebase-apply/applying)
Note that in both cases, the list of "am options" is
@@ -33,18 +33,17 @@
break;
if (is_interactive(&options) && i >= 0)
-- die(_("error: cannot combine interactive options "
+- die(_("cannot combine interactive options "
- "(--interactive, --exec, --rebase-merges, "
- "--preserve-merges, --keep-empty, --root + "
- "--onto) with am options (%s)"), buf.buf);
-+ die(_("error: cannot combine am options "
++ die(_("cannot combine am options "
+ "with interactive options"));
if (options.type == REBASE_MERGE && i >= 0)
-- die(_("error: cannot combine merge options (--merge, "
+- die(_("cannot combine merge options (--merge, "
- "--strategy, --strategy-option) with am options "
- "(%s)"), buf.buf);
-+ die(_("error: cannot combine am options "
-+ "with merge options "));
++ die(_("cannot combine am options with merge options "));
}
if (options.signoff) {
@@ -56,15 +55,15 @@
then
if test -n "$incompatible_opts"
then
-- die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")"
-+ die "$(gettext "error: cannot combine am options with interactive options")"
+- die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")"
++ die "$(gettext "fatal: cannot combine am options with interactive options")"
fi
fi
if test -n "$do_merge"; then
if test -n "$incompatible_opts"
then
-- die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")"
-+ die "$(gettext "error: cannot combine am options with merge options")"
+- die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")"
++ die "$(gettext "fatal: cannot combine am options with merge options")"
fi
fi
fi
2: cc33a8ccc1 = 3: 15d929edb2 t5407: add a test demonstrating how interactive handles --skip differently
3: f5838ef763 = 4: c9d6d5141e am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
4: 50dc863d9f = 5: 0b19ad8e2d git-rebase, sequencer: extend --quiet option for the interactive machinery
5: 35cf552f27 ! 6: 5ded8654ec git-legacy-rebase: simplify unnecessary triply-nested if
@@ -18,7 +18,7 @@
moving the innermost conditional to the outside, allowing us to remove
the test on git_am_opt entirely and giving us the following form:
- if incomptable_opts:
+ if incompatible_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
@@ -44,18 +44,18 @@
then
- if test -n "$incompatible_opts"
- then
-- die "$(gettext "error: cannot combine am options with interactive options")"
+- die "$(gettext "fatal: cannot combine am options with interactive options")"
- fi
-+ die "$(gettext "error: cannot combine am options with interactive options")"
++ die "$(gettext "fatal: cannot combine am options with interactive options")"
fi
- if test -n "$do_merge"; then
- if test -n "$incompatible_opts"
- then
-- die "$(gettext "error: cannot combine am options with merge options")"
+- die "$(gettext "fatal: cannot combine am options with merge options")"
- fi
+ if test -n "$do_merge"
+ then
-+ die "$(gettext "error: cannot combine am options with merge options")"
++ die "$(gettext "fatal: cannot combine am options with merge options")"
fi
fi6: 2a3d8ff1c1 = 7: bb8e5a4527 rebase: define linearization ordering and enforce it
7: 58371d377a ! 8: 5de428d695 rebase: Implement --merge via the interactive machinery
@@ -142,14 +142,15 @@
imply_interactive(&options, "--root without --onto");
@@
+ break;
if (is_interactive(&options) && i >= 0)
- die(_("error: cannot combine am options "
+- die(_("cannot combine am options "
- "with interactive options"));
- if (options.type == REBASE_MERGE && i >= 0)
-- die(_("error: cannot combine am options "
-- "with merge options "));
-+ "with either interactive or merge options"));
+- die(_("cannot combine am options with merge options "));
++ die(_("cannot combine am options with either "
++ "interactive or merge options"));
}
if (options.signoff) {
@@ -205,13 +206,13 @@
then
- if test -n "$interactive_rebase"
- then
-- die "$(gettext "error: cannot combine am options with interactive options")"
+- die "$(gettext "fatal: cannot combine am options with interactive options")"
- fi
- if test -n "$do_merge"
+ if test -n "$actually_interactive" || test "$do_merge"
then
-- die "$(gettext "error: cannot combine am options with merge options")"
-+ die "$(gettext "error: cannot combine am options with either interactive or merge options")"
+- die "$(gettext "fatal: cannot combine am options with merge options")"
++ die "$(gettext "fatal: cannot combine am options with either interactive or merge options")"
fi
fi
@@ -225,7 +226,7 @@
# linear history?
! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/null
@@
- GIT_PAGER='' git diff --stat --summary "$mb" "$onto"
+ GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto"
fi
+if test -z "$actually_interactive" && test "$mb" = "$orig_head"Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 1/8 rebase: make builtin and legacy script error messages the same
The conversion of the script version of rebase took messages that were prefixed with "error:" and passed them along to die(), which adds a "fatal:" prefix, thus resulting in messages of the form:
fatal: error: cannot combine...
which seems redundant. Remove the "error:" prefix from the builtin version of rebase, and change the prefix from "error:" to "fatal:" in the legacy script to match.
--- builtin/rebase.c | 10 +++++----- git-legacy-rebase.sh | 12 ++++++------ 2 files changed, 11 insertions(+), 11 deletions(-)
builtin/rebase.c +5 -5
@@ -1223,12 +1223,12 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("error: cannot combine interactive options " +die (_ ("cannot combine interactive options " "(--interactive, --exec, --rebase-merges, " "--preserve-merges, --keep-empty, --root + " "--onto) with am options (%s)" ), buf.buf);if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("error: cannot combine merge options (--merge, " +die (_ ("cannot combine merge options (--merge, " "--strategy, --strategy-option) with am options " "(%s)" ), buf.buf); }
@@ -1248,15 +1248,15 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)* git-rebase.txt caveats with "unless you know what you are doing" */ if (options.rebase_merges) -die (_ ("error: cannot combine '--preserve-merges' with " +die (_ ("cannot combine '--preserve-merges' with " "'--rebase-merges'" ));if (options.rebase_merges) {if (strategy_options.nr) -die (_ ("error: cannot combine '--rebase-merges' with " +die (_ ("cannot combine '--rebase-merges' with " "'--strategy-option'" ));if (options.strategy) -die (_ ("error: cannot combine '--rebase-merges' with " +die (_ ("cannot combine '--rebase-merges' with " "'--strategy'" )); }
git-legacy-rebase.sh +6 -6
@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" fi fi fi
@@ -522,7 +522,7 @@ fi if test -n "$signoff" then test -n "$preserve_merges" && - die "$(gettext "error: cannot combine '--signoff' with '--preserve-merges'")" + die "$(gettext "fatal: cannot combine '--signoff' with '--preserve-merges'")" git_am_opt="$git_am_opt $signoff" force_rebase=t fi
@@ -533,15 +533,15 @@ then # Note: incompatibility with --interactive is just a strong warning; # git-rebase.txt caveats with "unless you know what you are doing" test -n "$rebase_merges" && - die "$(gettext "error: cannot combine '--preserve-merges' with '--rebase-merges'")" + die "$(gettext "fatal: cannot combine '--preserve-merges' with '--rebase-merges'")" fi if test -n "$rebase_merges" then test -n "$strategy_opts" && - die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy-option'")" + die "$(gettext "fatal: cannot combine '--rebase-merges' with '--strategy-option'")" test -n "$strategy" && - die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy'")" + die "$(gettext "fatal: cannot combine '--rebase-merges' with '--strategy'")" fi if test -z "$rebase_root"
Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 2/8 rebase: fix incompatible options error message
In reply to Elijah NewrenIn commit f57696802c30 ("rebase: really just passthru the git am options", 2018-11-14), the handling of git am options was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:
$ git rebase --keep --ignore-whitespace fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)
$ git rebase --merge --ignore-whitespace fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)
Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
--- builtin/rebase.c | 10 +++------- git-legacy-rebase.sh | 4 ++-- 2 files changed, 5 insertions(+), 9 deletions(-)
builtin/rebase.c +3 -7
@@ -1223,14 +1223,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("cannot combine interactive options " -"(--interactive, --exec, --rebase-merges, " -"--preserve-merges, --keep-empty, --root + " -"--onto) with am options (%s)" ), buf.buf); +die (_ ("cannot combine am options " +"with interactive options" ));if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("cannot combine merge options (--merge, " -"--strategy, --strategy-option) with am options " -"(%s)" ), buf.buf); +die (_ ("cannot combine am options with merge options " )); }if (options.signoff) {
git-legacy-rebase.sh +2 -2
@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine am options with merge options")" fi fi fi
Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 3/8 t5407: add a test demonstrating how interactive handles --skip differently
In reply to Elijah NewrenThe post-rewrite hook is documented as being invoked by commands that rewrite commits such as commit --amend and rebase, and that it will be called for each rewritten commit.
Apparently, the three backends handled --skip'ed commits differently:
am: treat the skipped commit as though it weren't rewritten
merge: same as 'am' backend
interactive: treat skipped commits as having been rewritten to empty
(view them as an empty fixup to their parent)
For now, just add a testcase documenting the different behavior (use --keep to force usage of the interactive machinery even though we have no empty commits). A subsequent commit will remove the inconsistency in --skip handling.
--- t/t5407-post-rewrite-hook.sh | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
t/t5407-post-rewrite-hook.sh +31 -0
@@ -125,6 +125,37 @@ test_expect_success 'git rebase -m --skip' ' verify_hook_input ' +test_expect_success 'git rebase with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + echo C > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + +test_expect_success 'git rebase --skip with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + test_must_fail git rebase --skip && + echo D > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + . "$TEST_DIRECTORY"/lib-rebase.sh set_fake_editor
Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 4/8 am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
In reply to Elijah NewrenThe post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)
builtin/am.c +9 -0
@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" )); +if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit)); +fprintf (fp,"%s\n" ,oid_to_hex (&head)); +fclose (fp); + } +am_next (state);am_load (state);am_run (state,0 );
git-rebase--merge.sh +2 -0
@@ -121,6 +121,8 @@ continue) skip) read_state git rerere clear + cmt="$(cat "$state_dir/cmt.$msgnum")" + echo "$cmt $(git rev-parse HEAD^0)" >> "$state_dir/rewritten" msgnum=$(($msgnum + 1)) while test "$msgnum" -le "$end" do
t/t5407-post-rewrite-hook.sh +3 -0
@@ -78,6 +78,7 @@ test_expect_success 'git rebase --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
@@ -91,6 +92,7 @@ test_expect_success 'git rebase --skip the last one' ' echo rebase >expected.args && cat >expected.data <<-EOF && $(git rev-parse E) $(git rev-parse HEAD) + $(git rev-parse F) $(git rev-parse HEAD) EOF verify_hook_input '
@@ -120,6 +122,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
Signature · 1 line
2.20.0.8.g5de428d695
Re: [PATCH v4 4/8] am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
Hi,
Quoted 11 lines · Elijah Newren
The post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
Signed-off-by Elijah Newren <[email protected]>
This makes sense. I think, though, that we need to be more careful...
Quoted 11 lines
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)builtin/am.c +0
-0
builtin/am.c static void am_skip(struct am_state *state) · line 2007
Quoted 8 lines
builtin/am.c +5
-0@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" )); +if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit));
... here. What if fp == NULL? (Users do all kinds of interesting things...)
Ciao,
Dscho
Quoted 52 lines
builtin/am.c +4
-0+fprintf (fp,"%s\n" ,oid_to_hex (&head)); +fclose (fp); + } +am_next (state);am_load (state);am_run (state,0 );git-rebase--merge.sh +2
-0@@ -121,6 +121,8 @@ continue) skip) read_state git rerere clear + cmt="$(cat "$state_dir/cmt.$msgnum")" + echo "$cmt $(git rev-parse HEAD^0)" >> "$state_dir/rewritten" msgnum=$(($msgnum + 1)) while test "$msgnum" -le "$end" dot/t5407-post-rewrite-hook.sh +3
-0@@ -78,6 +78,7 @@ test_expect_success 'git rebase --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input@@ -91,6 +92,7 @@ test_expect_success 'git rebase --skip the last one' ' echo rebase >expected.args && cat >expected.data <<-EOF && $(git rev-parse E) $(git rev-parse HEAD) + $(git rev-parse F) $(git rev-parse HEAD) EOF verify_hook_input '@@ -120,6 +122,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_inputSignature · 1 line
2.20.0.8.g5de428d695
Re: [PATCH v4 4/8] am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
Hi Dscho,
Quoted 33 lines · Johannes Schindelin
Hi,
Quoted 11 lines · Elijah Newren
The post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
Signed-off-by Elijah Newren <[email protected]>This makes sense. I think, though, that we need to be more careful...
Quoted 14 lines
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)builtin/am.c +0
-0@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" ));
Quoted 8 lines
builtin/am.c static void am_skip(struct am_state *state) · line 2007
Quoted 5 lines
builtin/am.c +5
-0+if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit));... here. What if
fp == NULL? (Users do all kinds of interesting things...)
This if-block is actually copy-pasted from the end of the do_commit() function, since the same logic was needed in both places. The fact that a fp == NULL case never triggered for do_commit() suggests that the check has never been needed in the wild (or perhaps it just indicates a latent bug that no one has triggered yet). However, it does suggest a code cleanup regardless. I thought of it as such a small block that I didn't think to put it in a separate function, but perhaps I should so that someone spotting the possibility of a NULL fp could fix it for both callers in a single place.
Should I insert a preliminary change pulling this block out of do_commit into a separate function, and then modify this patch to just call this function? Or perhaps given the length of time it has already been cooking (and Junio's rerere resolution of our two series that I don't want to mess up), just do it as a simple post-series fixup?
Elijah
Re: [PATCH v4 4/8] am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
Hi Elijah,
Quoted 52 lines · Elijah Newren
Quoted 31 lines · Johannes Schindelin
Quoted 11 lines · Elijah Newren
The post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
Signed-off-by Elijah Newren <[email protected]>This makes sense. I think, though, that we need to be more careful...
Quoted 14 lines
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)builtin/am.c +0
-0@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" ));Quoted 8 lines
builtin/am.c static void am_skip(struct am_state *state) · line 2007
Quoted 5 lines
builtin/am.c +5
-0+if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit));... here. What if
fp == NULL? (Users do all kinds of interesting things...)This if-block is actually copy-pasted from the end of the
do_commit()function, since the same logic was needed in both places. The fact that afp == NULLcase never triggered fordo_commit()suggests that the check has never been needed in the wild (or perhaps it just indicates a latent bug that no one has triggered yet). However, it does suggest a code cleanup regardless. I thought of it as such a small block that I didn't think to put it in a separate function, but perhaps I should so that someone spotting the possibility of a NULL fp could fix it for both callers in a single place.
Quoted 6 lines
Should I insert a preliminary change pulling this block out of
do_commitinto a separate function, and then modify this patch to just call this function? Or perhaps given the length of time it has already been cooking (and Junio's rerere resolution of our two series that I don't want to mess up), just do it as a simple post-series fixup?
Speaking for myself: I am fine with either way.
Ciao,
Dscho
PATCH v4 5/8 git-rebase, sequencer: extend --quiet option for the interactive machinery
In reply to Elijah NewrenWhile 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)
builtin/rebase.c +1 -4
@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts))) opts->flags &= ~REBASE_NO_QUIET ;else opts->flags |=REBASE_NO_QUIET ;
git-legacy-rebase.sh +1 -1
@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&
git-rebase--common.sh +1 -1
@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \
sequencer.c +13 -10
@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")
@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {
@@ -2357,6 +2357,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;
@@ -2424,9 +2427,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)
@@ -3503,10 +3503,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());
@@ -3738,8 +3739,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);
sequencer.h +1 -0
@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;
Signature · 1 line
2.20.0.8.g5de428d695
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Elijah,
Quoted 13 lines · Elijah Newren
While 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Quoted 8 lines
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the
state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.Signed-off-by Elijah Newren <[email protected]>
This is convincing. I would like to point out, though...
Quoted 13 lines
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)builtin/rebase.c +0
-0
builtin/rebase.c static int read_basic_state(struct rebase_options *opts) · line 188
Quoted 8 lines
builtin/rebase.c +1
-4@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts)))
This changes the behavior. AFAIR the quiet file was always written, but contained t in quiet mode. I have to interrupt my review here, and will continue later, but maybe you will beat me to looking into that.
Ciao,
Dscho
Quoted 114 lines
builtin/rebase.c +0
-0opts->flags &= ~REBASE_NO_QUIET ; else opts->flags |= REBASE_NO_QUIET;git-legacy-rebase.sh +1
-1@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&git-rebase--common.sh +1
-1@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \sequencer.c +13
-10@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {@@ -2357,6 +2357,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;@@ -2424,9 +2427,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)@@ -3503,10 +3503,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());@@ -3738,8 +3739,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);sequencer.h +1
-0@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;Signature · 1 line
2.20.0.8.g5de428d695
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Dscho,
Quoted 55 lines · Johannes Schindelin
Hi Elijah,
Quoted 13 lines · Elijah Newren
While 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Quoted 8 lines
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the
state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.Signed-off-by Elijah Newren <[email protected]>This is convincing. I would like to point out, though...
Quoted 13 lines
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)builtin/rebase.c +0
-0builtin/rebase.c static int read_basic_state(struct rebase_options *opts) · line 188
Quoted 8 lines
builtin/rebase.c +1
-4@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts)))This changes the behavior. AFAIR the
quietfile was always written, but containedtin quiet mode. I have to interrupt my review here, and will continue later, but maybe you will beat me to looking into that.
Quoted 2 lines
Ciao,
Dscho
You are certainly consistent; you commented on this exact same issue in both v1 and v2 (you didn't have time to review v3). In v2, your comment was[1]:
"
I am slightly concerned that some creative power user could have written scripts that rely on this behavior.
But only *slightly* concerned.
The patch looks correct.
"
Also, I have a fuzzy memory of discussing a very similar case with some rebase-oriented option and its on-disk representation, where the concern was more about users upgrading git versions during an incomplete rebase rather than power users looking at internal file contents. And I think either Phillip or Junio made some statement about considering these internal details and that they felt the worry about upgrade mid-rebase was overly worrying. But I can't find the emails right now, and it's been so long (at least half a year) that I might be imagining things.
But from first principles, in this case even if you're worried about power users reaching into internal files of rebase or about users upgrade mid-rebase, at *worst* people will get or lose a few fluffy progress-update messages. To me, that's an innocuous level of change that's certainly worth the risk to allow us to get rid of the annoying differences in implementation of handling of different options. But, if you strongly feel that's too big a risk, we can remove the part of the patch making verbose and quiet be handled consistently; it's not critical to the rest of the series. I just thought it was a good cleanup while I was touching the area.
Elijah
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Elijah,
Quoted 55 lines · Elijah Newren
Quoted 53 lines · Johannes Schindelin
Quoted 13 lines · Elijah Newren
While 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Quoted 8 lines
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the
state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.Signed-off-by Elijah Newren <[email protected]>This is convincing. I would like to point out, though...
Quoted 13 lines
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)builtin/rebase.c +0
-0builtin/rebase.c static int read_basic_state(struct rebase_options *opts) · line 188
Quoted 8 lines
builtin/rebase.c +1
-4@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts)))This changes the behavior. AFAIR the
quietfile was always written, but containedtin quiet mode. I have to interrupt my review here, and will continue later, but maybe you will beat me to looking into that.
Quoted 4 lines
Quoted 2 lines
Ciao,
DschoYou are certainly consistent;
You flatter me!
Quoted 12 lines
you commented on this exact same issue in both v1 and v2 (you didn't have time to review v3). In v2, your comment was[1]:
"
I am slightly concerned that some creative power user could have written scripts that rely on this behavior.But only *slightly* concerned.
The patch looks correct.
"
Quoted 9 lines
Also, I have a fuzzy memory of discussing a very similar case with some rebase-oriented option and its on-disk representation, where the concern was more about users upgrading git versions during an incomplete rebase rather than power users looking at internal file contents. And I think either Phillip or Junio made some statement about considering these internal details and that they felt the worry about upgrade mid-rebase was overly worrying. But I can't find the emails right now, and it's been so long (at least half a year) that I might be imagining things.
Indeed, it was one of Phillip's patch series that fixed an incorrect quoting in the author script. And we eventually all agreed that it would not be worth the trouble to take care of upgrade-spanning rebases (both Phillip and I spent some time to make it so, IIRC).
Quoted 10 lines
But from first principles, in this case even if you're worried about power users reaching into internal files of rebase or about users upgrade mid-rebase, at *worst* people will get or lose a few fluffy progress-update messages. To me, that's an innocuous level of change that's certainly worth the risk to allow us to get rid of the annoying differences in implementation of handling of different options. But, if you strongly feel that's too big a risk, we can remove the part of the patch making verbose and quiet be handled consistently; it's not critical to the rest of the series. I just thought it was a good cleanup while I was touching the area.
I think you are correct. It is much easier in C to check whether a file exists than to read it, parse it and determine whether its Boolean value should be false or true.
So color me convinced that your clean-up is worth it. Maybe, just maybe, if you submit another iteration (and only if), could you add a little paragraph for future Dscho to understand that this clean-up was slipped in?
Thanks,
Dscho
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Elijah,
Quoted 3 lines · Johannes Schindelin
Maybe, just maybe, if you submit another iteration (and only if), could you add a little paragraph for future Dscho to understand that this clean-up was slipped in?
Oh wow. I guess I should learn how to read. The last paragraph says exactly that. For some reason, I missed that earlier.
Sorry for the noise,
Dscho
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
In reply to Elijah NewrenQuoted 9 lines · Elijah Newren
Also, I have a fuzzy memory of discussing a very similar case with some rebase-oriented option and its on-disk representation, where the concern was more about users upgrading git versions during an incomplete rebase rather than power users looking at internal file contents. And I think either Phillip or Junio made some statement about considering these internal details and that they felt the worry about upgrade mid-rebase was overly worrying. But I can't find the emails right now, and it's been so long (at least half a year) that I might be imagining things.
I do recall saying that mid-rebase upgrade is probably not worth getting worried about.
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Quoted 12 lines · Junio C Hamano
Quoted 9 lines · Elijah Newren
Also, I have a fuzzy memory of discussing a very similar case with some rebase-oriented option and its on-disk representation, where the concern was more about users upgrading git versions during an incomplete rebase rather than power users looking at internal file contents. And I think either Phillip or Junio made some statement about considering these internal details and that they felt the worry about upgrade mid-rebase was overly worrying. But I can't find the emails right now, and it's been so long (at least half a year) that I might be imagining things.
Quoted 3 lines
I do recall saying that mid-rebase upgrade is probably not worth getting worried about.
In light of yesterday's bug report [1] about those other changes I'm more concerned about this change. We were worrying about whether or not to worry about a mid-rebase upgrade but it seems people can have two different versions of git installed - one bundled with something like tig and another they use on the command line. If they start a rebase with a version containing this patch and try to continue it with a version that does not then the older version will fail with a complaint about a missing quiet file. The other way round they'll potentially get the wrong quiet setting which isn't such a problem. It's probably a bit late in the release cycle now to change this? But we could flag it up in the release notes and bear it in mind when making changes in the future.
Best Wishes
Phillip
[1]
https://public-inbox.org/git/[email protected]/T/#maa521f788bd61f9b65d52c14430a88bf077e6752
Re: [PATCH v4 5/8] git-rebase, sequencer: extend --quiet option for the interactive machinery
Hi Phillip,
Quoted 17 lines · Phillip Wood
Quoted 12 lines · Junio C Hamano
Quoted 9 lines · Elijah Newren
Also, I have a fuzzy memory of discussing a very similar case with some rebase-oriented option and its on-disk representation, where the concern was more about users upgrading git versions during an incomplete rebase rather than power users looking at internal file contents. And I think either Phillip or Junio made some statement about considering these internal details and that they felt the worry about upgrade mid-rebase was overly worrying. But I can't find the emails right now, and it's been so long (at least half a year) that I might be imagining things.
Quoted 3 lines
I do recall saying that mid-rebase upgrade is probably not worth getting worried about.
Quoted 11 lines
In light of yesterday's bug report [1] about those other changes I'm more concerned about this change. We were worrying about whether or not to worry about a mid-rebase upgrade but it seems people can have two different versions of git installed - one bundled with something like tig and another they use on the command line. If they start a rebase with a version containing this patch and try to continue it with a version that does not then the older version will fail with a complaint about a missing quiet file. The other way round they'll potentially get the wrong quiet setting which isn't such a problem. It's probably a bit late in the release cycle now to change this? But we could flag it up in the release notes and bear it in mind when making changes in the future.
Thanks for the heads up; I'll try to keep it in mind for the future.
PATCH v4 6/8 git-legacy-rebase: simplify unnecessary triply-nested if
In reply to Elijah NewrenThe git-legacy-rebase.sh script previously had code of the form:
if git_am_opt:
if interactive:
if incompatible_opts:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
if incompatible_opts
show_error_about_merge_and_am_incompatibilities
which was a triply nested if. However, the first conditional (git_am_opt) and third (incompatible_opts) were somewhat redundant: the latter condition was a strict subset of the former. Simplify this by moving the innermost conditional to the outside, allowing us to remove the test on git_am_opt entirely and giving us the following form:
if incompatible_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
show_error_about_merge_and_am_incompatibilities
--- git-legacy-rebase.sh | 20 ++++++++------------ 1 file changed, 8 insertions(+), 12 deletions(-)
git-legacy-rebase.sh +8 -12
@@ -501,21 +501,17 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fi -if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +if test -n "$incompatible_opts" +then if test -n "$interactive_rebase" then - if test -n "$incompatible_opts" - then - die "$(gettext "fatal: cannot combine am options with interactive options")" - fi + die "$(gettext "fatal: cannot combine am options with interactive options")" fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then - die "$(gettext "fatal: cannot combine am options with merge options")" - fi + if test -n "$do_merge" + then + die "$(gettext "fatal: cannot combine am options with merge options")" fi fi
Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 7/8 rebase: define linearization ordering and enforce it
In reply to Elijah NewrenEver since commit 3f213981e44a ("add tests for rebasing merged history", 2013-06-06), t3425 has had tests which included the rebasing of merged history and whose order of applied commits was checked. Unfortunately, the tests expected different behavior depending on which backend was in use. Implementing these checks was the following four lines (including the TODO message) which were repeated verbatim three times in t3425:
#TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' test_run_rebase success 'e n o' -m test_run_rebase success 'n o e' -i
As part of the effort to reduce differences between the rebase backends so that users get more uniform behavior, let's define the correct behavior and modify the different backends so they all get the right answer. It turns out that the difference in behavior here is entirely due to topological sorting; since some backends require topological sorting (particularly when --rebase-merges is specified), require it for all modes. Modify the am and merge backends to implement this.
Performance Considerations:
I was unable to measure any appreciable performance difference with this change. Trying to control the run-to-run variation was difficult; I eventually found a headless beefy box that I could ssh into, which seemed to help. Using git.git, I ran the following testcase:
$ git reset --hard v2.20.0-rc1~2
$ time git rebase --quiet v2.20.0-rc0~16I first ran once to warm any disk caches, then ran five subsequent runs and recorded the times of those five. I observed the following results for the average time:
Before this change:
"real" timing: 1.340s (standard deviation: 0.040s)
"user" timing: 1.050s (standard deviation: 0.041s)
"sys" timing: 0.270s (standard deviation: 0.011s)
After this change:
"real" timing: 1.327s (standard deviation: 0.065s)
"user" timing: 1.031s (standard deviation: 0.061s)
"sys" timing: 0.280s (standard deviation: 0.014s)Measurements aside, I would expect the timing for walking revisions to be dwarfed by the work involved in creating and applying patches, so this isn't too surprising. Further, while somewhat counter-intuitive, it is possible that turning on topological sorting is actually a performance improvement: by way of comparison, turning on --topo-order made fast-export faster (see
https://public-inbox.org/git/[email protected]/).
--- git-rebase--am.sh | 2 +- git-rebase--merge.sh | 2 +- t/t3425-rebase-topology-merges.sh | 15 ++++++--------- 3 files changed, 8 insertions(+), 11 deletions(-)
git-rebase--am.sh +1 -1
@@ -36,7 +36,7 @@ rm -f "$GIT_DIR/rebased-patches" git format-patch -k --stdout --full-index --cherry-pick --right-only \ --src-prefix=a/ --dst-prefix=b/ --no-renames --no-cover-letter \ - --pretty=mboxrd \ + --pretty=mboxrd --topo-order \ $git_format_patch_opt \ "$revisions" ${restrict_revision+^$restrict_revision} \ >"$GIT_DIR/rebased-patches"
git-rebase--merge.sh +1 -1
@@ -143,7 +143,7 @@ write_basic_state rm -f "$(git rev-parse --git-path REBASE_HEAD)" msgnum=0 -for cmt in $(git rev-list --reverse --no-merges "$revisions") +for cmt in $(git rev-list --topo-order --reverse --no-merges "$revisions") do msgnum=$(($msgnum + 1)) echo "$cmt" > "$state_dir/cmt.$msgnum"
t/t3425-rebase-topology-merges.sh +6 -9
@@ -70,9 +70,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" d.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' '' +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {
@@ -87,9 +86,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {
@@ -104,9 +102,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i if ! test_have_prereq REBASE_P; then
Signature · 1 line
2.20.0.8.g5de428d695
PATCH v4 8/8 rebase: Implement --merge via the interactive machinery
In reply to Elijah NewrenAs part of an ongoing effort to make rebase have more uniform behavior, modify the merge backend to behave like the interactive one, by re-implementing it on top of the latter.
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore. Delete git-rebase--merge.sh and instead implement it in builtin/rebase.c.
This results in a few deliberate but small user-visible changes:
* The progress output is modified (see t3406 and t3420 for examples)
* A few known test failures are now fixed (see t3421)
* bash-prompt during a rebase --merge is now REBASE-i instead of REBASE-m. Reason: The prompt is a reflection of the backend in use; this allows users to report an issue to the git mailing list with the appropriate backend information, and allows advanced users to know where to search for relevant control files. (see t9903)
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match the new expectation
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
--- .gitignore | 1 - Documentation/git-rebase.txt | 17 +-- Makefile | 1 - builtin/rebase.c | 15 ++- git-legacy-rebase.sh | 43 ++++---- git-rebase--merge.sh | 166 ------------------------------ t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t9903-bash-prompt.sh | 2 +- 10 files changed, 43 insertions(+), 297 deletions(-) delete mode 100644 git-rebase--merge.sh
.gitignore +0 -1
@@ -124,7 +124,6 @@ /git-rebase--am /git-rebase--common /git-rebase--interactive -/git-rebase--merge /git-rebase--preserve-merges /git-receive-pack /git-reflog
Documentation/git-rebase.txt +3 -14
@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS -------------------- -git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) - -Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date
@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options: * --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: - * --[no-]autosquash * --rebase-merges * --preserve-merges
@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoff
Makefile +0 -1
@@ -628,7 +628,6 @@ SCRIPT_LIB += git-parse-remote SCRIPT_LIB += git-rebase--am SCRIPT_LIB += git-rebase--common SCRIPT_LIB += git-rebase--preserve-merges -SCRIPT_LIB += git-rebase--merge SCRIPT_LIB += git-sh-setup SCRIPT_LIB += git-sh-i18n
builtin/rebase.c +6 -9
@@ -122,7 +122,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;
@@ -481,10 +481,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;
@@ -1191,6 +1187,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) +imply_interactive (&options,"--merge" ); +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );
@@ -1220,10 +1219,8 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("cannot combine am options " -"with interactive options" )); -if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("cannot combine am options with merge options " )); +die (_ ("cannot combine am options with either " +"interactive or merge options" )); }if (options.signoff) {
git-legacy-rebase.sh +21 -22
96 lines not on this page · raw message
git-rebase--merge.sh deleted +0 -166
166 lines not on this page · raw message
t/t3406-rebase-message.sh +1 -6
@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF - test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actual
t/t3420-rebase-autostash.sh +6 -72
102 lines not on this page · raw message
t/t3421-rebase-topology-linear.sh +5 -5
@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase failure -p
t/t9903-bash-prompt.sh +1 -1
@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected && git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&
Signature · 1 line
2.20.0.8.g5de428d695
Quoted 8 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Friendly ping...let me know if you want me to simply resend v4.
Quoted 17 lines
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery
...
Quoted 10 lines · Elijah Newren
Quoted 8 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Quoted 2 lines
Friendly ping...let me know if you want me to simply resend v4.
If you have anything newer than 90673135 ("rebase: Implement --merge via the interactive machinery", 2018-12-11), then yeah, I haven't seen it.
Thanks.
P.S. even if that one is latest, I would need to downcase Implement before it hits 'next' ;-)
Quoted 18 lines
Quoted 17 lines
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery...
Quoted 15 lines · Junio C Hamano
Quoted 10 lines · Elijah Newren
Quoted 8 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Quoted 2 lines
Friendly ping...let me know if you want me to simply resend v4.
Quoted 8 lines
If you have anything newer than 90673135 ("rebase: Implement --merge via the interactive machinery", 2018-12-11), then yeah, I haven't seen it.
Thanks.
P.S. even if that one is latest, I would need to downcase Implement before it hits 'next' ;-)
Ah, one thing I forgot to mention. Some of the tests updated in this series are unhappy with Dscho's "drive 'am' directly from the built-in code, bypassing git-rebase--am.sh scriptlet" topic.
Hi,
Quoted 22 lines · Junio C Hamano
Quoted 19 lines · Junio C Hamano
Quoted 10 lines · Elijah Newren
Quoted 8 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Quoted 2 lines
Friendly ping...let me know if you want me to simply resend v4.
If you have anything newer than 90673135 ("rebase: Implement --merge via the interactive machinery", 2018-12-11), then yeah, I haven't seen it.
Quoted 8 lines
Quoted 4 lines
Thanks.
P.S. even if that one is latest, I would need to downcase Implement before it hits 'next' ;-)Ah, one thing I forgot to mention. Some of the tests updated in this series are unhappy with Dscho's "drive 'am' directly from the built-in code, bypassing
git-rebase--am.shscriptlet" topic.
2018-12-11 is the newest (and is almost the same as the version from mid November); it's just been waiting for review. I'll fix up the casing of 'Implement' along with any other feedback, if any...maybe including rebasing on Dscho's series depending on how he wants to take it.
Dscho: Looks like our series conflicts slightly. Would you like me to rebase mine on top of yours and squash the following change into commit c91c944a068e ("rebase: define linearization ordering and enforce it", 2018-12-11), or do you want to rebase your series on mine and either make a new commit out of this change or squash it in somewhere?
builtin/rebase.c +2 -1
@@ -578,7 +578,8 @@ static int run_am(struct rebase_options *opts)argv_array_pushl (&format_patch.args,"format-patch" ,"-k" ,"--stdout" ,"--full-index" ,"--cherry-pick" ,"--right-only" ,"--src-prefix=a/" ,"--dst-prefix=b/" ,"--no-renames" , -"--no-cover-letter" ,"--pretty=mboxrd" ,NULL ); +"--no-cover-letter" ,"--pretty=mboxrd" , +"--topo-order" ,NULL );if (opts->git_format_patch_opt.len)argv_array_split (&format_patch.args, opts->git_format_patch_opt.buf);
Elijah
Hi Junio,
A small update...
Quoted 9 lines · Elijah Newren
Quoted 8 lines · Junio C Hamano
Quoted 6 lines · Junio C Hamano
Quoted 4 lines · Elijah Newren
Quoted 3 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In
....
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
P.S. even if that one is latest, I would need to downcase Implement before it hits 'next' ;-)
...
Quoted 3 lines
Quoted 3 lines
Ah, one thing I forgot to mention. Some of the tests updated in this series are unhappy with Dscho's "drive 'am' directly from the built-in code, bypassing
git-rebase--am.shscriptlet" topic.
It looks like you've already done the downcasing in pu; thanks. Also, Dscho told me in private email that:
"FWIW I am fine with your patches going in first, and I would rebase mine
on top. It will take me probably until next week to get to my patch series and Junio's comments, anyway."
So, I'm guessing that means he'll apply my patch below to his series. As such, other than maybe wait for folks to review my updated series, I'm not aware of any further changes I need to make to the en/rebase-merge-on-sequencer topic.
Quoted 24 lines
Dscho: Looks like our series conflicts slightly. Would you like me to rebase mine on top of yours and squash the following change into commit
c91c944a068e("rebase: define linearization ordering and enforce it", 2018-12-11), or do you want to rebase your series on mine and either make a new commit out of this change or squash it in somewhere?builtin/rebase.c +2
-1@@ -578,7 +578,8 @@ static int run_am(struct rebase_options *opts)argv_array_pushl (&format_patch.args,"format-patch" ,"-k" ,"--stdout" ,"--full-index" ,"--cherry-pick" ,"--right-only" ,"--src-prefix=a/" ,"--dst-prefix=b/" ,"--no-renames" , -"--no-cover-letter" ,"--pretty=mboxrd" ,NULL ); +"--no-cover-letter" ,"--pretty=mboxrd" , +"--topo-order" ,NULL );if (opts->git_format_patch_opt.len)argv_array_split (&format_patch.args, opts->git_format_patch_opt.buf);Elijah
Hi Elijah,
Quoted 51 lines · Elijah Newren
Quoted 22 lines · Junio C Hamano
Quoted 19 lines · Junio C Hamano
Quoted 10 lines · Elijah Newren
Quoted 8 lines · Elijah Newren
This series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Quoted 2 lines
Friendly ping...let me know if you want me to simply resend v4.
If you have anything newer than 90673135 ("rebase: Implement --merge via the interactive machinery", 2018-12-11), then yeah, I haven't seen it.
Quoted 8 lines
Quoted 4 lines
Thanks.
P.S. even if that one is latest, I would need to downcase Implement before it hits 'next' ;-)Ah, one thing I forgot to mention. Some of the tests updated in this series are unhappy with Dscho's "drive 'am' directly from the built-in code, bypassing
git-rebase--am.shscriptlet" topic.2018-12-11 is the newest (and is almost the same as the version from mid November); it's just been waiting for review. I'll fix up the casing of 'Implement' along with any other feedback, if any...maybe including rebasing on Dscho's series depending on how he wants to take it.
Dscho: Looks like our series conflicts slightly. Would you like me to rebase mine on top of yours and squash the following change into commit
c91c944a068e("rebase: define linearization ordering and enforce it", 2018-12-11), or do you want to rebase your series on mine and either make a new commit out of this change or squash it in somewhere?builtin/rebase.c +0
-0
builtin/rebase.c static int run_am(struct rebase_options *opts) · line 585
Quoted 9 lines
builtin/rebase.c +2
-1@@ -578,7 +578,8 @@ static int run_am(struct rebase_options *opts)argv_array_pushl (&format_patch.args,"format-patch" ,"-k" ,"--stdout" ,"--full-index" ,"--cherry-pick" ,"--right-only" ,"--src-prefix=a/" ,"--dst-prefix=b/" ,"--no-renames" , -"--no-cover-letter" ,"--pretty=mboxrd" ,NULL ); +"--no-cover-letter" ,"--pretty=mboxrd" , +"--topo-order" ,NULL );if (opts->git_format_patch_opt.len)argv_array_split (&format_patch.args, opts->git_format_patch_opt.buf);
I can easily squash that in. Thank you!
Dscho
Quoted 5 lines
Elijah
Hi Elijah,
Quoted 9 lines · Johannes Schindelin
Quoted 7 lines · Elijah Newren
Dscho: Looks like our series conflicts slightly. Would you like me to rebase mine on top of yours and squash the following change into commit
c91c944a068e("rebase: define linearization ordering and enforce it", 2018-12-11), or do you want to rebase your series on mine and either make a new commit out of this change or squash it in somewhere?
Quoted 16 lines
Quoted 5 lines
builtin/rebase.c +0
-0builtin/rebase.c static int run_am(struct rebase_options *opts) · line 585
Quoted 9 lines
builtin/rebase.c +2
-1@@ -578,7 +578,8 @@ static int run_am(struct rebase_options *opts)argv_array_pushl (&format_patch.args,"format-patch" ,"-k" ,"--stdout" ,"--full-index" ,"--cherry-pick" ,"--right-only" ,"--src-prefix=a/" ,"--dst-prefix=b/" ,"--no-renames" , -"--no-cover-letter" ,"--pretty=mboxrd" ,NULL ); +"--no-cover-letter" ,"--pretty=mboxrd" , +"--topo-order" ,NULL );if (opts->git_format_patch_opt.len)argv_array_split (&format_patch.args, opts->git_format_patch_opt.buf);I can easily squash that in.
Actually, I take that back. This is tested-for in the regression test suite, and I need to keep the built-in and the scripted rebase in sync for that test suite to pass, so that single-liner would incur more changes than I am comfortable with adopting into the builtin-rebase--am patch series...
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Ciao,
Dscho
Quoted 3 lines · Johannes Schindelin
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Or both of you keep the topics as-is and self-consistent, and let the rerere machinery to squash it in when the two topics gets merged.
Quoted 10 lines · Junio C Hamano
Quoted 3 lines · Johannes Schindelin
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Or both of you keep the topics as-is and self-consistent, and let the rerere machinery to squash it in when the two topics gets merged.
Yeah, I see that commit 8d808e12a2cc ("Merge branch 'js/rebase-am' into jch", 2019-01-17) has my fix squashed in; thanks Junio.
I'm happy to do whatever is needed with en/rebase-merge-on-sequencer that makes it easiest for everyone else. I'm assuming right now that there's nothing I should do to improve it, but if anyone wants me to rebase it or fix something up, just let me know.
Hi Junio,
Quoted 9 lines · Junio C Hamano
Quoted 3 lines · Johannes Schindelin
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Or both of you keep the topics as-is and self-consistent, and let the rerere machinery to squash it in when the two topics gets merged.
With all the experience I have with rerere, I don't trust it.
Ciao,
Johannes
Quoted 10 lines · Johannes Schindelin
Hi Junio,
Quoted 6 lines · Junio C Hamano
Quoted 3 lines · Johannes Schindelin
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Quoted 5 lines
Quoted 3 lines
Or both of you keep the topics as-is and self-consistent, and let the rerere machinery to squash it in when the two topics gets merged.
With all the experience I have with rerere, I don't trust it.
FWIW, I trust it once I got right resolution better than randomly rebased resubmission that needs to be re-reviewed afresh.
Hi Junio,
Quoted 10 lines · Junio C Hamano
Quoted 8 lines · Johannes Schindelin
Quoted 6 lines · Junio C Hamano
Quoted 3 lines · Johannes Schindelin
I am about to submit a new iteration of my patch series, would it be too much trouble for you to rebase on top? If it would be, let me know, then I will rebase on top of yours.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
Or both of you keep the topics as-is and self-consistent, and let the rerere machinery to squash it in when the two topics gets merged.
With all the experience I have with rerere, I don't trust it.
FWIW, I trust it once I got right resolution better than randomly rebased resubmission that needs to be re-reviewed afresh.
Once again, our experiences teach us diametrically opposed things. I definitely trust my rebased version more than your rerere-affected resolution.
Ciao,
Dscho
Hi Elijah,
Quoted 4 lines · Elijah Newren
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0
- Fixed the "comptable" tyop
This, and the range-diff, look reasonable to me.
Thanks,
Dscho
Quoted 181 lines
Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/am.c | 9 ++ builtin/rebase.c | 30 ++---- git-legacy-rebase.sh | 65 ++++++------ git-rebase--am.sh | 2 +- git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 15 ++- t/t5407-post-rewrite-hook.sh | 34 +++++++ t/t9903-bash-prompt.sh | 2 +- 17 files changed, 121 insertions(+), 340 deletions(-) delete mode 100644 git-rebase--merge.shRange-diff:
-: ---------- > 1: 2e8b1bcb8b rebase: make builtin and legacy script error messages the same 1: 2f4bdd1980 ! 2: eba87828c6 rebase: fix incompatible options error message @@ -9,12 +9,12 @@ understood by separate backends were used: $ git rebase --keep --ignore-whitespace - fatal: error: cannot combine interactive options (--interactive, --exec, + fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying) $ git rebase --merge --ignore-whitespace - fatal: error: cannot combine merge options (--merge, --strategy, + fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying) Note that in both cases, the list of "am options" is @@ -33,18 +33,17 @@ break; if (is_interactive(&options) && i >= 0) -- die(_("error: cannot combine interactive options " +- die(_("cannot combine interactive options " - "(--interactive, --exec, --rebase-merges, " - "--preserve-merges, --keep-empty, --root + " - "--onto) with am options (%s)"), buf.buf); -+ die(_("error: cannot combine am options " ++ die(_("cannot combine am options " + "with interactive options")); if (options.type == REBASE_MERGE && i >= 0) -- die(_("error: cannot combine merge options (--merge, " +- die(_("cannot combine merge options (--merge, " - "--strategy, --strategy-option) with am options " - "(%s)"), buf.buf); -+ die(_("error: cannot combine am options " -+ "with merge options ")); ++ die(_("cannot combine am options with merge options ")); } if (options.signoff) { @@ -56,15 +55,15 @@ then if test -n "$incompatible_opts" then -- die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" -+ die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" ++ die "$(gettext "fatal: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then -- die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" -+ die "$(gettext "error: cannot combine am options with merge options")" +- die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" ++ die "$(gettext "fatal: cannot combine am options with merge options")" fi fi fi 2: cc33a8ccc1 = 3: 15d929edb2 t5407: add a test demonstrating how interactive handles --skip differently 3: f5838ef763 = 4: c9d6d5141e am, rebase--merge: do not overlook --skip'ed commits with post-rewrite 4: 50dc863d9f = 5: 0b19ad8e2d git-rebase, sequencer: extend --quiet option for the interactive machinery 5: 35cf552f27 ! 6: 5ded8654ec git-legacy-rebase: simplify unnecessary triply-nested if @@ -18,7 +18,7 @@ moving the innermost conditional to the outside, allowing us to remove the test on git_am_opt entirely and giving us the following form: - if incomptable_opts: + if incompatible_opts: if interactive: show_error_about_interactive_and_am_incompatibilities if rebase-merge: @@ -44,18 +44,18 @@ then - if test -n "$incompatible_opts" - then -- die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine am options with interactive options")" - fi -+ die "$(gettext "error: cannot combine am options with interactive options")" ++ die "$(gettext "fatal: cannot combine am options with interactive options")" fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then -- die "$(gettext "error: cannot combine am options with merge options")" +- die "$(gettext "fatal: cannot combine am options with merge options")" - fi + if test -n "$do_merge" + then -+ die "$(gettext "error: cannot combine am options with merge options")" ++ die "$(gettext "fatal: cannot combine am options with merge options")" fi fi6: 2a3d8ff1c1 = 7: bb8e5a4527 rebase: define linearization ordering and enforce it 7: 58371d377a ! 8: 5de428d695 rebase: Implement --merge via the interactive machinery @@ -142,14 +142,15 @@ imply_interactive(&options, "--root without --onto"); @@ + break; if (is_interactive(&options) && i >= 0) - die(_("error: cannot combine am options " +- die(_("cannot combine am options " - "with interactive options")); - if (options.type == REBASE_MERGE && i >= 0) -- die(_("error: cannot combine am options " -- "with merge options ")); -+ "with either interactive or merge options")); +- die(_("cannot combine am options with merge options ")); ++ die(_("cannot combine am options with either " ++ "interactive or merge options")); } if (options.signoff) { @@ -205,13 +206,13 @@ then - if test -n "$interactive_rebase" - then -- die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine am options with interactive options")" - fi - if test -n "$do_merge" + if test -n "$actually_interactive" || test "$do_merge" then -- die "$(gettext "error: cannot combine am options with merge options")" -+ die "$(gettext "error: cannot combine am options with either interactive or merge options")" +- die "$(gettext "fatal: cannot combine am options with merge options")" ++ die "$(gettext "fatal: cannot combine am options with either interactive or merge options")" fi fi @@ -225,7 +226,7 @@ # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/null @@ - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head"Signature · 1 line
2.20.0.8.g5de428d695
Hi Elijah,
Quoted 10 lines · Johannes Schindelin
Quoted 4 lines · Elijah Newren
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0- Fixed the "comptable" tyop
This, and the range-diff, look reasonable to me.
And now that I looked through all 8 patches, those look reasonable to me, too. I think this is ready for next.
Ciao,
Dscho
Quoted 186 lines
Thanks,
DschoQuoted 181 lines
Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: Implement --merge via the interactive machinery.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/am.c | 9 ++ builtin/rebase.c | 30 ++---- git-legacy-rebase.sh | 65 ++++++------ git-rebase--am.sh | 2 +- git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 15 ++- t/t5407-post-rewrite-hook.sh | 34 +++++++ t/t9903-bash-prompt.sh | 2 +- 17 files changed, 121 insertions(+), 340 deletions(-) delete mode 100644 git-rebase--merge.shRange-diff:
-: ---------- > 1: 2e8b1bcb8b rebase: make builtin and legacy script error messages the same 1: 2f4bdd1980 ! 2: eba87828c6 rebase: fix incompatible options error message @@ -9,12 +9,12 @@ understood by separate backends were used: $ git rebase --keep --ignore-whitespace - fatal: error: cannot combine interactive options (--interactive, --exec, + fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying) $ git rebase --merge --ignore-whitespace - fatal: error: cannot combine merge options (--merge, --strategy, + fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying) Note that in both cases, the list of "am options" is @@ -33,18 +33,17 @@ break; if (is_interactive(&options) && i >= 0) -- die(_("error: cannot combine interactive options " +- die(_("cannot combine interactive options " - "(--interactive, --exec, --rebase-merges, " - "--preserve-merges, --keep-empty, --root + " - "--onto) with am options (%s)"), buf.buf); -+ die(_("error: cannot combine am options " ++ die(_("cannot combine am options " + "with interactive options")); if (options.type == REBASE_MERGE && i >= 0) -- die(_("error: cannot combine merge options (--merge, " +- die(_("cannot combine merge options (--merge, " - "--strategy, --strategy-option) with am options " - "(%s)"), buf.buf); -+ die(_("error: cannot combine am options " -+ "with merge options ")); ++ die(_("cannot combine am options with merge options ")); } if (options.signoff) { @@ -56,15 +55,15 @@ then if test -n "$incompatible_opts" then -- die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" -+ die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" ++ die "$(gettext "fatal: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then -- die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" -+ die "$(gettext "error: cannot combine am options with merge options")" +- die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" ++ die "$(gettext "fatal: cannot combine am options with merge options")" fi fi fi 2: cc33a8ccc1 = 3: 15d929edb2 t5407: add a test demonstrating how interactive handles --skip differently 3: f5838ef763 = 4: c9d6d5141e am, rebase--merge: do not overlook --skip'ed commits with post-rewrite 4: 50dc863d9f = 5: 0b19ad8e2d git-rebase, sequencer: extend --quiet option for the interactive machinery 5: 35cf552f27 ! 6: 5ded8654ec git-legacy-rebase: simplify unnecessary triply-nested if @@ -18,7 +18,7 @@ moving the innermost conditional to the outside, allowing us to remove the test on git_am_opt entirely and giving us the following form: - if incomptable_opts: + if incompatible_opts: if interactive: show_error_about_interactive_and_am_incompatibilities if rebase-merge: @@ -44,18 +44,18 @@ then - if test -n "$incompatible_opts" - then -- die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine am options with interactive options")" - fi -+ die "$(gettext "error: cannot combine am options with interactive options")" ++ die "$(gettext "fatal: cannot combine am options with interactive options")" fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then -- die "$(gettext "error: cannot combine am options with merge options")" +- die "$(gettext "fatal: cannot combine am options with merge options")" - fi + if test -n "$do_merge" + then -+ die "$(gettext "error: cannot combine am options with merge options")" ++ die "$(gettext "fatal: cannot combine am options with merge options")" fi fi6: 2a3d8ff1c1 = 7: bb8e5a4527 rebase: define linearization ordering and enforce it 7: 58371d377a ! 8: 5de428d695 rebase: Implement --merge via the interactive machinery @@ -142,14 +142,15 @@ imply_interactive(&options, "--root without --onto"); @@ + break; if (is_interactive(&options) && i >= 0) - die(_("error: cannot combine am options " +- die(_("cannot combine am options " - "with interactive options")); - if (options.type == REBASE_MERGE && i >= 0) -- die(_("error: cannot combine am options " -- "with merge options ")); -+ "with either interactive or merge options")); +- die(_("cannot combine am options with merge options ")); ++ die(_("cannot combine am options with either " ++ "interactive or merge options")); } if (options.signoff) { @@ -205,13 +206,13 @@ then - if test -n "$interactive_rebase" - then -- die "$(gettext "error: cannot combine am options with interactive options")" +- die "$(gettext "fatal: cannot combine am options with interactive options")" - fi - if test -n "$do_merge" + if test -n "$actually_interactive" || test "$do_merge" then -- die "$(gettext "error: cannot combine am options with merge options")" -+ die "$(gettext "error: cannot combine am options with either interactive or merge options")" +- die "$(gettext "fatal: cannot combine am options with merge options")" ++ die "$(gettext "fatal: cannot combine am options with either interactive or merge options")" fi fi @@ -225,7 +226,7 @@ # linear history? ! (git rev-list --parents "$onto".."$orig_head" | sane_grep " .* ") > /dev/null @@ - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto" fi +if test -z "$actually_interactive" && test "$mb" = "$orig_head"Signature · 1 line
2.20.0.8.g5de428d695
Hi Dscho,
Quoted 14 lines · Johannes Schindelin
Hi Elijah,
Quoted 9 lines · Johannes Schindelin
Quoted 4 lines · Elijah Newren
Differences since v3 (full range-diff below):
- Fixed the redundant "fatal: error:" error message prefixes, as pointed out by Duy
- Rebased on 2.20.0- Fixed the "comptable" tyop
Quoted 7 lines
Quoted 1 line
This, and the range-diff, look reasonable to me.
And now that I looked through all 8 patches, those look reasonable to me, too. I think this is ready for
next.Ciao,
Dscho
Thanks for reviewing it!
v5 cover Reimplement rebase --merge via interactive machinery
In reply to Elijah NewrenThis series continues the work of making rebase more self-consistent by removing inconsistencies between different backends. In particular, this series focuses on making the merge machinery behave like the interactive machinery (though a few differences between the am and interactive backends are also fixed along the way), and ultimately removes the merge backend in favor of reimplementing the relevant options on top of the interactive machinery.
Differences since v4:
- Included Dscho's Acked-by (from
https://public-inbox.org/git/[email protected]/)
Elijah Newren (8):
rebase: make builtin and legacy script error messages the same
rebase: fix incompatible options error message
t5407: add a test demonstrating how interactive handles --skip
differently
am, rebase--merge: do not overlook --skip'ed commits with post-rewrite git-rebase, sequencer: extend --quiet option for the interactive
machinery
git-legacy-rebase: simplify unnecessary triply-nested if
rebase: define linearization ordering and enforce it
rebase: implement --merge via the interactive machinery
.gitignore | 1 - Documentation/git-rebase.txt | 17 +--- Makefile | 1 - builtin/am.c | 9 ++ builtin/rebase.c | 30 ++---- git-legacy-rebase.sh | 65 ++++++------ git-rebase--am.sh | 2 +- git-rebase--common.sh | 2 +- git-rebase--merge.sh | 164 ------------------------------ sequencer.c | 23 +++-- sequencer.h | 1 + t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t3425-rebase-topology-merges.sh | 15 ++- t/t5407-post-rewrite-hook.sh | 34 +++++++ t/t9903-bash-prompt.sh | 2 +- 17 files changed, 121 insertions(+), 340 deletions(-) delete mode 100644 git-rebase--merge.sh
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 1/8 rebase: make builtin and legacy script error messages the same
The conversion of the script version of rebase took messages that were prefixed with "error:" and passed them along to die(), which adds a "fatal:" prefix, thus resulting in messages of the form:
fatal: error: cannot combine...
which seems redundant. Remove the "error:" prefix from the builtin version of rebase, and change the prefix from "error:" to "fatal:" in the legacy script to match.
--- builtin/rebase.c | 10 +++++----- git-legacy-rebase.sh | 12 ++++++------ 2 files changed, 11 insertions(+), 11 deletions(-)
builtin/rebase.c +5 -5
@@ -1223,12 +1223,12 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("error: cannot combine interactive options " +die (_ ("cannot combine interactive options " "(--interactive, --exec, --rebase-merges, " "--preserve-merges, --keep-empty, --root + " "--onto) with am options (%s)" ), buf.buf);if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("error: cannot combine merge options (--merge, " +die (_ ("cannot combine merge options (--merge, " "--strategy, --strategy-option) with am options " "(%s)" ), buf.buf); }
@@ -1248,15 +1248,15 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)* git-rebase.txt caveats with "unless you know what you are doing" */ if (options.rebase_merges) -die (_ ("error: cannot combine '--preserve-merges' with " +die (_ ("cannot combine '--preserve-merges' with " "'--rebase-merges'" ));if (options.rebase_merges) {if (strategy_options.nr) -die (_ ("error: cannot combine '--rebase-merges' with " +die (_ ("cannot combine '--rebase-merges' with " "'--strategy-option'" ));if (options.strategy) -die (_ ("error: cannot combine '--rebase-merges' with " +die (_ ("cannot combine '--rebase-merges' with " "'--strategy'" )); }
git-legacy-rebase.sh +6 -6
@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "error: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" fi fi fi
@@ -522,7 +522,7 @@ fi if test -n "$signoff" then test -n "$preserve_merges" && - die "$(gettext "error: cannot combine '--signoff' with '--preserve-merges'")" + die "$(gettext "fatal: cannot combine '--signoff' with '--preserve-merges'")" git_am_opt="$git_am_opt $signoff" force_rebase=t fi
@@ -533,15 +533,15 @@ then # Note: incompatibility with --interactive is just a strong warning; # git-rebase.txt caveats with "unless you know what you are doing" test -n "$rebase_merges" && - die "$(gettext "error: cannot combine '--preserve-merges' with '--rebase-merges'")" + die "$(gettext "fatal: cannot combine '--preserve-merges' with '--rebase-merges'")" fi if test -n "$rebase_merges" then test -n "$strategy_opts" && - die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy-option'")" + die "$(gettext "fatal: cannot combine '--rebase-merges' with '--strategy-option'")" test -n "$strategy" && - die "$(gettext "error: cannot combine '--rebase-merges' with '--strategy'")" + die "$(gettext "fatal: cannot combine '--rebase-merges' with '--strategy'")" fi if test -z "$rebase_root"
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 2/8 rebase: fix incompatible options error message
In reply to Elijah NewrenIn commit f57696802c30 ("rebase: really just passthru the git am options", 2018-11-14), the handling of git am options was simplified dramatically (and an option parsing bug was fixed), but it introduced a small regression in the error message shown when options only understood by separate backends were used:
$ git rebase --keep --ignore-whitespace fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options (.git/rebase-apply/applying)
$ git rebase --merge --ignore-whitespace fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options (.git/rebase-apply/applying)
Note that in both cases, the list of "am options" is ".git/rebase-apply/applying", which makes no sense. Since the lists of backend-specific options is documented pretty thoroughly in the rebase man page (in the "Incompatible Options" section, with multiple links throughout the document), and since I expect this list to change over time, just simplify the error message.
--- builtin/rebase.c | 10 +++------- git-legacy-rebase.sh | 4 ++-- 2 files changed, 5 insertions(+), 9 deletions(-)
builtin/rebase.c +3 -7
@@ -1223,14 +1223,10 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("cannot combine interactive options " -"(--interactive, --exec, --rebase-merges, " -"--preserve-merges, --keep-empty, --root + " -"--onto) with am options (%s)" ), buf.buf); +die (_ ("cannot combine am options " +"with interactive options" ));if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("cannot combine merge options (--merge, " -"--strategy, --strategy-option) with am options " -"(%s)" ), buf.buf); +die (_ ("cannot combine am options with merge options " )); }if (options.signoff) {
git-legacy-rebase.sh +2 -2
@@ -508,13 +508,13 @@ if test -n "$git_am_opt"; then then if test -n "$incompatible_opts" then - die "$(gettext "fatal: cannot combine interactive options (--interactive, --exec, --rebase-merges, --preserve-merges, --keep-empty, --root + --onto) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine am options with interactive options")" fi fi if test -n "$do_merge"; then if test -n "$incompatible_opts" then - die "$(gettext "fatal: cannot combine merge options (--merge, --strategy, --strategy-option) with am options ($incompatible_opts)")" + die "$(gettext "fatal: cannot combine am options with merge options")" fi fi fi
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 3/8 t5407: add a test demonstrating how interactive handles --skip differently
In reply to Elijah NewrenThe post-rewrite hook is documented as being invoked by commands that rewrite commits such as commit --amend and rebase, and that it will be called for each rewritten commit.
Apparently, the three backends handled --skip'ed commits differently:
am: treat the skipped commit as though it weren't rewritten
merge: same as 'am' backend
interactive: treat skipped commits as having been rewritten to empty
(view them as an empty fixup to their parent)
For now, just add a testcase documenting the different behavior (use --keep to force usage of the interactive machinery even though we have no empty commits). A subsequent commit will remove the inconsistency in --skip handling.
--- t/t5407-post-rewrite-hook.sh | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
t/t5407-post-rewrite-hook.sh +31 -0
@@ -125,6 +125,37 @@ test_expect_success 'git rebase -m --skip' ' verify_hook_input ' +test_expect_success 'git rebase with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + echo C > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + +test_expect_success 'git rebase --skip with implicit use of interactive backend' ' + git reset --hard D && + clear_hook_input && + test_must_fail git rebase --keep --onto A B && + test_must_fail git rebase --skip && + echo D > foo && + git add foo && + git rebase --continue && + echo rebase >expected.args && + cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) + $(git rev-parse D) $(git rev-parse HEAD) + EOF + verify_hook_input +' + . "$TEST_DIRECTORY"/lib-rebase.sh set_fake_editor
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 4/8 am, rebase--merge: do not overlook --skip'ed commits with post-rewrite
In reply to Elijah NewrenThe post-rewrite hook is supposed to be invoked for each rewritten commit. The fact that a commit was selected and processed by the rebase operation (even though when we hit an error a user said it had no more useful changes), suggests we should write an entry for it. In particular, let's treat it as an empty commit trivially squashed into its parent.
This brings the rebase--am and rebase--merge backends in sync with the behavior of the interactive rebase backend.
--- builtin/am.c | 9 +++++++++ git-rebase--merge.sh | 2 ++ t/t5407-post-rewrite-hook.sh | 3 +++ 3 files changed, 14 insertions(+)
builtin/am.c +9 -0
@@ -2000,6 +2000,15 @@ static void am_skip(struct am_state *state)if (clean_index (&head, &head))die (_ ("failed to clean index" )); +if (state->rebasing) { +FILE *fp =xfopen (am_path (state,"rewritten" ),"a" ); + +assert (!is_null_oid (&state->orig_commit)); +fprintf (fp,"%s " ,oid_to_hex (&state->orig_commit)); +fprintf (fp,"%s\n" ,oid_to_hex (&head)); +fclose (fp); + } +am_next (state);am_load (state);am_run (state,0 );
git-rebase--merge.sh +2 -0
@@ -121,6 +121,8 @@ continue) skip) read_state git rerere clear + cmt="$(cat "$state_dir/cmt.$msgnum")" + echo "$cmt $(git rev-parse HEAD^0)" >> "$state_dir/rewritten" msgnum=$(($msgnum + 1)) while test "$msgnum" -le "$end" do
t/t5407-post-rewrite-hook.sh +3 -0
@@ -78,6 +78,7 @@ test_expect_success 'git rebase --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
@@ -91,6 +92,7 @@ test_expect_success 'git rebase --skip the last one' ' echo rebase >expected.args && cat >expected.data <<-EOF && $(git rev-parse E) $(git rev-parse HEAD) + $(git rev-parse F) $(git rev-parse HEAD) EOF verify_hook_input '
@@ -120,6 +122,7 @@ test_expect_success 'git rebase -m --skip' ' git rebase --continue && echo rebase >expected.args && cat >expected.data <<-EOF && + $(git rev-parse C) $(git rev-parse HEAD^) $(git rev-parse D) $(git rev-parse HEAD) EOF verify_hook_input
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 5/8 git-rebase, sequencer: extend --quiet option for the interactive machinery
In reply to Elijah NewrenWhile 'quiet' and 'interactive' may sound like antonyms, the interactive machinery actually has logic that implements several interactive_rebase=implied cases (--exec, --keep-empty, --rebase-merges) which won't pop up an editor. The rewrite of interactive rebase in C added a quiet option, though it only turns stats off. Since we want to make the interactive machinery also take over for git-rebase--merge, it should fully implement the --quiet option.
git-rebase--interactive was already somewhat quieter than git-rebase--merge and git-rebase--am, possibly because cherry-pick has just traditionally been quieter. As such, we only drop a few informational messages -- "Rebasing (n/m)" and "Successfully rebased..."
Also, for simplicity, remove the differences in how quiet and verbose options were recorded. Having one be signalled by the presence of a "verbose" file in the state_dir, while the other was signalled by the contents of a "quiet" file was just weirdly inconsistent. (This inconsistency pre-dated the rewrite into C.) Make them consistent by having them both key off the presence of the file.
--- builtin/rebase.c | 5 +---- git-legacy-rebase.sh | 2 +- git-rebase--common.sh | 2 +- sequencer.c | 23 +++++++++++++---------- sequencer.h | 1 + 5 files changed, 17 insertions(+), 16 deletions(-)
builtin/rebase.c +1 -4
@@ -185,10 +185,7 @@ static int read_basic_state(struct rebase_options *opts)if (get_oid (buf.buf, &opts->orig_head))return error (_ ("invalid orig-head: '%s'" ), buf.buf); -strbuf_reset (&buf); -if (read_one (state_dir_path ("quiet" , opts), &buf)) -return -1 ; -if (buf.len) +if (file_exists (state_dir_path ("quiet" , opts))) opts->flags &= ~REBASE_NO_QUIET ;else opts->flags |=REBASE_NO_QUIET ;
git-legacy-rebase.sh +1 -1
@@ -113,7 +113,7 @@ read_basic_state () { else orig_head=$(cat "$state_dir"/head) fi && - GIT_QUIET=$(cat "$state_dir"/quiet) && + test -f "$state_dir"/quiet && GIT_QUIET=t test -f "$state_dir"/verbose && verbose=t test -f "$state_dir"/strategy && strategy="$(cat "$state_dir"/strategy)" test -f "$state_dir"/strategy_opts &&
git-rebase--common.sh +1 -1
@@ -10,7 +10,7 @@ write_basic_state () { echo "$head_name" > "$state_dir"/head-name && echo "$onto" > "$state_dir"/onto && echo "$orig_head" > "$state_dir"/orig-head && - echo "$GIT_QUIET" > "$state_dir"/quiet && + test t = "$GIT_QUIET" && : > "$state_dir"/quiet test t = "$verbose" && : > "$state_dir"/verbose test -n "$strategy" && echo "$strategy" > "$state_dir"/strategy test -n "$strategy_opts" && echo "$strategy_opts" > \
sequencer.c +13 -10
@@ -150,6 +150,7 @@ static GIT_PATH_FUNC(rebase_path_refs_to_delete, "rebase-merge/refs-to-delete") static GIT_PATH_FUNC(rebase_path_gpg_sign_opt, "rebase-merge/gpg_sign_opt") staticGIT_PATH_FUNC (rebase_path_orig_head,"rebase-merge/orig-head" ) static GIT_PATH_FUNC(rebase_path_verbose, "rebase-merge/verbose") +staticGIT_PATH_FUNC (rebase_path_quiet,"rebase-merge/quiet" ) static GIT_PATH_FUNC(rebase_path_signoff, "rebase-merge/signoff") staticGIT_PATH_FUNC (rebase_path_head_name,"rebase-merge/head-name" ) static GIT_PATH_FUNC(rebase_path_onto, "rebase-merge/onto")
@@ -157,7 +158,6 @@ static GIT_PATH_FUNC(rebase_path_autostash, "rebase-merge/autostash") static GIT_PATH_FUNC(rebase_path_strategy, "rebase-merge/strategy") static GIT_PATH_FUNC(rebase_path_strategy_opts, "rebase-merge/strategy_opts") static GIT_PATH_FUNC(rebase_path_allow_rerere_autoupdate, "rebase-merge/allow_rerere_autoupdate") -static GIT_PATH_FUNC(rebase_path_quiet, "rebase-merge/quiet") static int git_sequencer_config(const char *k, const char *v, void *cb) {
@@ -2357,6 +2357,9 @@ static int read_populate_opts(struct replay_opts *opts)if (file_exists (rebase_path_verbose ())) opts->verbose =1 ; +if (file_exists (rebase_path_quiet ())) + opts->quiet =1 ; +if (file_exists (rebase_path_signoff ())) { opts->allow_ff =0 ; opts->signoff =1 ;
@@ -2424,9 +2427,6 @@ int write_basic_state(struct replay_opts *opts, const char *head_name,if (quiet)write_file (rebase_path_quiet (),"%s\n" , quiet); -else -write_file (rebase_path_quiet (),"\n" ); -if (opts->verbose)write_file (rebase_path_verbose (),"%s" ,"" );if (opts->strategy)
@@ -3503,10 +3503,11 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts)fprintf (f,"%d\n" , todo_list->done_nr);fclose (f); } -fprintf (stderr,"Rebasing (%d/%d)%s" , - todo_list->done_nr, - todo_list->total_nr, - opts->verbose ?"\n" :"\r" ); +if (!opts->quiet) +fprintf (stderr,"Rebasing (%d/%d)%s" , + todo_list->done_nr, + todo_list->total_nr, + opts->verbose ?"\n" :"\r" ); }unlink (rebase_path_message ());unlink (rebase_path_author_script ());
@@ -3738,8 +3739,10 @@ static int pick_commits(struct todo_list *todo_list, struct replay_opts *opts) }apply_autostash (opts); -fprintf (stderr,"Successfully rebased and updated %s.\n" , - head_ref.buf); +if (!opts->quiet) +fprintf (stderr, +"Successfully rebased and updated %s.\n" , + head_ref.buf);strbuf_release (&buf);strbuf_release (&head_ref);
sequencer.h +1 -0
@@ -39,6 +39,7 @@ struct replay_opts {int allow_empty_message;int keep_redundant_commits;int verbose; +int quiet;int mainline;
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 6/8 git-legacy-rebase: simplify unnecessary triply-nested if
In reply to Elijah NewrenThe git-legacy-rebase.sh script previously had code of the form:
if git_am_opt:
if interactive:
if incompatible_opts:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
if incompatible_opts
show_error_about_merge_and_am_incompatibilities
which was a triply nested if. However, the first conditional (git_am_opt) and third (incompatible_opts) were somewhat redundant: the latter condition was a strict subset of the former. Simplify this by moving the innermost conditional to the outside, allowing us to remove the test on git_am_opt entirely and giving us the following form:
if incompatible_opts:
if interactive:
show_error_about_interactive_and_am_incompatibilities
if rebase-merge:
show_error_about_merge_and_am_incompatibilities
--- git-legacy-rebase.sh | 20 ++++++++------------ 1 file changed, 8 insertions(+), 12 deletions(-)
git-legacy-rebase.sh +8 -12
@@ -501,21 +501,17 @@ then git_format_patch_opt="$git_format_patch_opt --progress" fi -if test -n "$git_am_opt"; then - incompatible_opts=$(echo " $git_am_opt " | \ - sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +incompatible_opts=$(echo " $git_am_opt " | \ + sed -e 's/ -q / /g' -e 's/^ \(.*\) $/\1/') +if test -n "$incompatible_opts" +then if test -n "$interactive_rebase" then - if test -n "$incompatible_opts" - then - die "$(gettext "fatal: cannot combine am options with interactive options")" - fi + die "$(gettext "fatal: cannot combine am options with interactive options")" fi - if test -n "$do_merge"; then - if test -n "$incompatible_opts" - then - die "$(gettext "fatal: cannot combine am options with merge options")" - fi + if test -n "$do_merge" + then + die "$(gettext "fatal: cannot combine am options with merge options")" fi fi
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 7/8 rebase: define linearization ordering and enforce it
In reply to Elijah NewrenEver since commit 3f213981e44a ("add tests for rebasing merged history", 2013-06-06), t3425 has had tests which included the rebasing of merged history and whose order of applied commits was checked. Unfortunately, the tests expected different behavior depending on which backend was in use. Implementing these checks was the following four lines (including the TODO message) which were repeated verbatim three times in t3425:
#TODO: make order consistent across all flavors of rebase test_run_rebase success 'e n o' '' test_run_rebase success 'e n o' -m test_run_rebase success 'n o e' -i
As part of the effort to reduce differences between the rebase backends so that users get more uniform behavior, let's define the correct behavior and modify the different backends so they all get the right answer. It turns out that the difference in behavior here is entirely due to topological sorting; since some backends require topological sorting (particularly when --rebase-merges is specified), require it for all modes. Modify the am and merge backends to implement this.
Performance Considerations:
I was unable to measure any appreciable performance difference with this change. Trying to control the run-to-run variation was difficult; I eventually found a headless beefy box that I could ssh into, which seemed to help. Using git.git, I ran the following testcase:
$ git reset --hard v2.20.0-rc1~2
$ time git rebase --quiet v2.20.0-rc0~16I first ran once to warm any disk caches, then ran five subsequent runs and recorded the times of those five. I observed the following results for the average time:
Before this change:
"real" timing: 1.340s (standard deviation: 0.040s)
"user" timing: 1.050s (standard deviation: 0.041s)
"sys" timing: 0.270s (standard deviation: 0.011s)
After this change:
"real" timing: 1.327s (standard deviation: 0.065s)
"user" timing: 1.031s (standard deviation: 0.061s)
"sys" timing: 0.280s (standard deviation: 0.014s)Measurements aside, I would expect the timing for walking revisions to be dwarfed by the work involved in creating and applying patches, so this isn't too surprising. Further, while somewhat counter-intuitive, it is possible that turning on topological sorting is actually a performance improvement: by way of comparison, turning on --topo-order made fast-export faster (see
https://public-inbox.org/git/[email protected]/).
--- git-rebase--am.sh | 2 +- git-rebase--merge.sh | 2 +- t/t3425-rebase-topology-merges.sh | 15 ++++++--------- 3 files changed, 8 insertions(+), 11 deletions(-)
git-rebase--am.sh +1 -1
@@ -36,7 +36,7 @@ rm -f "$GIT_DIR/rebased-patches" git format-patch -k --stdout --full-index --cherry-pick --right-only \ --src-prefix=a/ --dst-prefix=b/ --no-renames --no-cover-letter \ - --pretty=mboxrd \ + --pretty=mboxrd --topo-order \ $git_format_patch_opt \ "$revisions" ${restrict_revision+^$restrict_revision} \ >"$GIT_DIR/rebased-patches"
git-rebase--merge.sh +1 -1
@@ -143,7 +143,7 @@ write_basic_state rm -f "$(git rev-parse --git-path REBASE_HEAD)" msgnum=0 -for cmt in $(git rev-list --reverse --no-merges "$revisions") +for cmt in $(git rev-list --topo-order --reverse --no-merges "$revisions") do msgnum=$(($msgnum + 1)) echo "$cmt" > "$state_dir/cmt.$msgnum"
t/t3425-rebase-topology-merges.sh +6 -9
@@ -70,9 +70,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" d.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'e n o' '' -test_run_rebase success 'e n o' -m +test_run_rebase success 'n o e' '' +test_run_rebase success 'n o e' -m test_run_rebase success 'n o e' -i test_run_rebase () {
@@ -87,9 +86,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i test_run_rebase () {
@@ -104,9 +102,8 @@ test_run_rebase () { test_linear_range "\'"$expected"\'" c.. " } -#TODO: make order consistent across all flavors of rebase -test_run_rebase success 'd e n o' '' -test_run_rebase success 'd e n o' -m +test_run_rebase success 'd n o e' '' +test_run_rebase success 'd n o e' -m test_run_rebase success 'd n o e' -i if ! test_have_prereq REBASE_P; then
Signature · 1 line
2.20.1.310.g17ca096f17
PATCH v5 8/8 rebase: implement --merge via the interactive machinery
In reply to Elijah NewrenAs part of an ongoing effort to make rebase have more uniform behavior, modify the merge backend to behave like the interactive one, by re-implementing it on top of the latter.
Interactive rebases are implemented in terms of cherry-pick rather than the merge-recursive builtin, but cherry-pick also calls into the recursive merge machinery by default and can accept special merge strategies and/or special strategy options. As such, there really is not any need for having both git-rebase--merge and git-rebase--interactive anymore. Delete git-rebase--merge.sh and instead implement it in builtin/rebase.c.
This results in a few deliberate but small user-visible changes:
* The progress output is modified (see t3406 and t3420 for examples)
* A few known test failures are now fixed (see t3421)
* bash-prompt during a rebase --merge is now REBASE-i instead of REBASE-m. Reason: The prompt is a reflection of the backend in use; this allows users to report an issue to the git mailing list with the appropriate backend information, and allows advanced users to know where to search for relevant control files. (see t9903)
testcase modification notes:
t3406: --interactive and --merge had slightly different progress output
while running; adjust a test to match the new expectation
t3420: these test precise output while running, but rebase--am,
rebase--merge, and rebase--interactive all were built on very different commands (am, merge-recursive, cherry-pick), so the tests expected different output for each type. Now we expect --merge and --interactive to have the same output.
t3421: --interactive fixes some bugs in --merge! Wahoo!
t9903: --merge uses the interactive backend so the prompt expected is
now REBASE-i.
--- .gitignore | 1 - Documentation/git-rebase.txt | 17 +-- Makefile | 1 - builtin/rebase.c | 15 ++- git-legacy-rebase.sh | 43 ++++---- git-rebase--merge.sh | 166 ------------------------------ t/t3406-rebase-message.sh | 7 +- t/t3420-rebase-autostash.sh | 78 ++------------ t/t3421-rebase-topology-linear.sh | 10 +- t/t9903-bash-prompt.sh | 2 +- 10 files changed, 43 insertions(+), 297 deletions(-) delete mode 100644 git-rebase--merge.sh
.gitignore +0 -1
@@ -124,7 +124,6 @@ /git-rebase--am /git-rebase--common /git-rebase--interactive -/git-rebase--merge /git-rebase--preserve-merges /git-receive-pack /git-reflog
Documentation/git-rebase.txt +3 -14
@@ -504,15 +504,7 @@ See also INCOMPATIBLE OPTIONS below. INCOMPATIBLE OPTIONS -------------------- -git-rebase has many flags that are incompatible with each other, -predominantly due to the fact that it has three different underlying -implementations: - - * one based on linkgit:git-am[1] (the default) - * one based on git-merge-recursive (merge backend) - * one based on linkgit:git-cherry-pick[1] (interactive backend) - -Flags only understood by the am backend: +The following options: * --committer-date-is-author-date * --ignore-date
@@ -520,15 +512,12 @@ Flags only understood by the am backend: * --ignore-whitespace * -C -Flags understood by both merge and interactive backends: +are incompatible with the following options: * --merge * --strategy * --strategy-option * --allow-empty-message - -Flags only understood by the interactive backend: - * --[no-]autosquash * --rebase-merges * --preserve-merges
@@ -539,7 +528,7 @@ Flags only understood by the interactive backend: * --edit-todo * --root when used in combination with --onto -Other incompatible flag pairs: +In addition, the following pairs of options are incompatible: * --preserve-merges and --interactive * --preserve-merges and --signoff
Makefile +0 -1
@@ -628,7 +628,6 @@ SCRIPT_LIB += git-parse-remote SCRIPT_LIB += git-rebase--am SCRIPT_LIB += git-rebase--common SCRIPT_LIB += git-rebase--preserve-merges -SCRIPT_LIB += git-rebase--merge SCRIPT_LIB += git-sh-setup SCRIPT_LIB += git-sh-i18n
builtin/rebase.c +6 -9
@@ -122,7 +122,7 @@ static void imply_interactive(struct rebase_options *opts, const char *option)case REBASE_PRESERVE_MERGES :break ;case REBASE_MERGE : -/* we silently *upgrade* --merge to --interactive if needed */ +/* we now implement --merge via --interactive */ default : opts->type =REBASE_INTERACTIVE ;/* implied */ break ;
@@ -481,10 +481,6 @@ static int run_specific_rebase(struct rebase_options *opts) backend ="git-rebase--am" ; backend_func ="git_rebase__am" ;break ; -case REBASE_MERGE : - backend ="git-rebase--merge" ; - backend_func ="git_rebase__merge" ; -break ;case REBASE_PRESERVE_MERGES : backend ="git-rebase--preserve-merges" ; backend_func ="git_rebase__preserve_merges" ;
@@ -1191,6 +1187,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix) } } +if (options.type ==REBASE_MERGE ) +imply_interactive (&options,"--merge" ); +if (options.root && !options.onto_name)imply_interactive (&options,"--root without --onto" );
@@ -1220,10 +1219,8 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)break ;if (is_interactive (&options) && i >=0 ) -die (_ ("cannot combine am options " -"with interactive options" )); -if (options.type ==REBASE_MERGE && i >=0 ) -die (_ ("cannot combine am options with merge options " )); +die (_ ("cannot combine am options with either " +"interactive or merge options" )); }if (options.signoff) {
git-legacy-rebase.sh +21 -22
96 lines not on this page · raw message
git-rebase--merge.sh deleted +0 -166
166 lines not on this page · raw message
t/t3406-rebase-message.sh +1 -6
@@ -17,14 +17,9 @@ test_expect_success 'setup' ' git tag start ' -cat >expect <<\EOF -Already applied: 0001 A -Already applied: 0002 B -Committed: 0003 Z -EOF - test_expect_success 'rebase -m' ' git rebase -m master >report && + >expect && sed -n -e "/^Already applied: /p" \ -e "/^Committed: /p" report >actual && test_cmp expect actual
t/t3420-rebase-autostash.sh +6 -72
102 lines not on this page · raw message
t/t3421-rebase-topology-linear.sh +5 -5
@@ -111,7 +111,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -126,7 +126,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -141,7 +141,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -284,7 +284,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase success -p
@@ -315,7 +315,7 @@ test_run_rebase () { " } test_run_rebase success '' -test_run_rebase failure -m +test_run_rebase success -m test_run_rebase success -i test_have_prereq !REBASE_P || test_run_rebase failure -p
t/t9903-bash-prompt.sh +1 -1
@@ -180,7 +180,7 @@ test_expect_success 'prompt - interactive rebase' ' ' test_expect_success 'prompt - rebase merge' ' - printf " (b2|REBASE-m 1/3)" >expected && + printf " (b2|REBASE-i 1/3)" >expected && git checkout b2 && test_when_finished "git checkout master" && test_must_fail git rebase --merge b1 b2 &&
Signature · 1 line
2.20.1.310.g17ca096f17