Fix git rebase --stat -i <unrelated-history>
cover Fix git rebase --stat -i <unrelated-history>
We're really entering obscure territory here, I would say.
To trigger the bug, two things have to come together: the user must have asked for a diffstat afterwards, and the commits need to have been rebased onto a completely unrelated commit history (i.e. there must exist no merge base between the pre-rebase HEAD and the post-rebase HEAD).
Please note that this bug existed already in the scripted rebase, but it was never detected because the scripted version would not even perform any error checking.
It will make Junio very happy that I squashed the regression test in to the patch that fixes it. The reason, however, was not to make Junio happy (I hope to make him happy by fixing bugs), but simply that an earlier iteration of test would only fail with the built-in rebase, but not with the scripted version. The current version would fail with the scripted version, but I'll save the time to split the patch again now.
Johannes Schindelin (1):
rebase --stat: fix when rebasing to an unrelated history
builtin/rebase.c | 8 +++++--- git-legacy-rebase.sh | 6 ++++-- t/t3406-rebase-message.sh | 10 ++++++++++ 3 files changed, 19 insertions(+), 5 deletions(-)
base-commit: a1598010f775d82b5adf12c29d0f5bc9b41434c6
Published-As: https://github.com/gitgitgadget/git/releases/tags/pr-88%2Fdscho%2Frebase-stat-fix-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-88/dscho/rebase-stat-fix-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/88
Signature · 1 line
gitgitgadget
PATCH 1/1 rebase --stat: fix when rebasing to an unrelated history
From: Johannes Schindelin <[email protected]>
When rebasing to a commit history that has no common commits with the current branch, there is no merge base. The scripted version of the git rebase command was not prepared for that and spewed out
fatal: ambiguous argument '': unknown revision or path not in the working tree.
but then continued (due to lack of error checking).
The built-in version of the git rebase command blindly translated that shell script code, assuming that there is no need to test whether there *was* a merge base, and due to its better error checking, exited with a fatal error (because it tried to read the object with hash 00000000... as a tree).
Fix both scripted and built-in versions to output something sensibly, and add a regression test to keep this working in all eternity.
--- builtin/rebase.c | 8 +++++--- git-legacy-rebase.sh | 6 ++++-- t/t3406-rebase-message.sh | 10 ++++++++++ 3 files changed, 19 insertions(+), 5 deletions(-)
builtin/rebase.c +5 -3
@@ -1483,7 +1483,8 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)if (options.flags &REBASE_VERBOSE )printf (_ ("Changes from %s to %s:\n" ), -oid_to_hex (&merge_base), +is_null_oid (&merge_base) ? +"(empty)" :oid_to_hex (&merge_base),oid_to_hex (&options.onto->object.oid));/* We want color (if set), but no pager */
@@ -1494,8 +1495,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)DIFF_FORMAT_SUMMARY |DIFF_FORMAT_DIFFSTAT ; opts.detect_rename =DIFF_DETECT_RENAME ;diff_setup_done (&opts); -diff_tree_oid (&merge_base, &options.onto->object.oid, -"" , &opts); +diff_tree_oid (is_null_oid (&merge_base) ? + the_hash_algo->empty_tree : &merge_base, + &options.onto->object.oid,"" , &opts);diffcore_std (&opts);diff_flush (&opts); }
git-legacy-rebase.sh +4 -2
@@ -718,10 +718,12 @@ if test -n "$diffstat" then if test -n "$verbose" then - echo "$(eval_gettext "Changes from \$mb to \$onto:")" + mb_display="${mb:-(empty)}" + echo "$(eval_gettext "Changes from \$mb_display to \$onto:")" fi + mb_tree="${mb:-$(git hash-object -t tree /dev/null)}" # We want color (if set), but no pager - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto" fi test -n "$interactive_rebase" && run_specific_rebase
t/t3406-rebase-message.sh +10 -0
@@ -91,4 +91,14 @@ test_expect_success 'error out early upon -C<n> or --whitespace=<bad>' ' test_i18ngrep "Invalid whitespace option" err ' +test_expect_success 'rebase -i onto unrelated history' ' + git init unrelated && + test_commit -C unrelated 1 && + git -C unrelated remote add -f origin "$PWD" && + git -C unrelated branch --set-upstream-to=origin/master && + git -C unrelated -c core.editor=true rebase -i -v --stat >actual && + test_i18ngrep "Changes from (empty)" actual && + test_i18ngrep "5 files changed" actual +' + test_done
Signature · 1 line
gitgitgadget
Re: [PATCH 1/1] rebase --stat: fix when rebasing to an unrelated history
"Johannes Schindelin via GitGitGadget" <[email protected]>
Quoted 5 lines · writes:
The built-in version of the
git rebasecommand blindly translated that shell script code, assuming that there is no need to test whether there *was* a merge base, and due to its better error checking, exited with a fatal error (because it tried to read the object with hash 00000000... as a tree).
And the scripted version produced a wrong result because it did not check the lack of merge base and fed an empty string (presumably, as that is what you would get from mb=$(merge-base A B) when A and B are unrelated) to later steps? If that is the case, then it *is* a very significant thing to record here. As it was not merely "failed to stop due to lack of error checking", but a lot worse. It was producing a wrong result.
A more faithful reimplementation would not have exited with fatal error, but would have produced the same wrong result, so "a better error checking caused the reimplementation die where the original did not die" may be correct but is much less important than the fact that "the logic used in the original to produce diffstat forgot that there may not be a merge base and would not have worked correctly in such a case, and that is what we are correcting" is more important.
Please descibe the issue as such, if that is the case (I cannot read from the description in the proposed log message if that is the case---but I came to the conclusion that it is what you wanted to fix reading the code).
builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1487
Quoted 5 lines
builtin/rebase.c +2
-1if (options.flags &REBASE_VERBOSE ) printf(_("Changes from %s to %s:\n"), - oid_to_hex(&merge_base), + is_null_oid(&merge_base) ? + "(empty)" : oid_to_hex(&merge_base),
I am not sure (empty) is a good idea for two reasons. It is going to be inserted into an translated string without translation. Also it is too similar to the fallback string used by some printf implementations when NULL was given to %s by mistake.
If there weren't i18n issues, I would suggest to use "empty merge base" or "empty tree" instead, but the old one would have shown 0{40}, so perhaps empty_tree_oid_hex() is a good substitute?
builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1500
Quoted 9 lines
builtin/rebase.c +3
-2@@ -1494,8 +1495,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)DIFF_FORMAT_SUMMARY |DIFF_FORMAT_DIFFSTAT ; opts.detect_rename =DIFF_DETECT_RENAME ;diff_setup_done (&opts); -diff_tree_oid (&merge_base, &options.onto->object.oid, -"" , &opts); +diff_tree_oid (is_null_oid (&merge_base) ? + the_hash_algo->empty_tree : &merge_base, + &options.onto->object.oid,"" , &opts);
The original would have run "git diff '' $onto", and died with an error message, so both the original and the reimplementation were failing. Just in different ways ;-)
Quoted 6 lines
git-legacy-rebase.sh +0
-0@@ -718,10 +718,12 @@ if test -n "$diffstat" then
git-legacy-rebase.sh if test -n "$diffstat" · line 726
Quoted 10 lines
git-legacy-rebase.sh +4
-2if test -n "$verbose" then - echo "$(eval_gettext "Changes from \$mb to \$onto:")" + mb_display="${mb:-(empty)}" + echo "$(eval_gettext "Changes from \$mb_display to \$onto:")" fi + mb_tree="${mb:-$(git hash-object -t tree /dev/null)}" # We want color (if set), but no pager - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto"
Code fix for diff-tree invocation, both in the builtin/ version and this one, looks correct.
Re: [PATCH 1/1] rebase --stat: fix when rebasing to an unrelated history
Hi Junio,
Quoted 9 lines · Junio C Hamano
"Johannes Schindelin via GitGitGadget" <[email protected]>
Quoted 5 lines · writes:
The built-in version of the
git rebasecommand blindly translated that shell script code, assuming that there is no need to test whether there *was* a merge base, and due to its better error checking, exited with a fatal error (because it tried to read the object with hash 00000000... as a tree).
Quoted 7 lines
And the scripted version produced a wrong result because it did not check the lack of merge base and fed an empty string (presumably, as that is what you would get from mb=$(merge-base A B) when A and B are unrelated) to later steps? If that is the case, then it *is* a very significant thing to record here. As it was not merely "failed to stop due to lack of error checking", but a lot worse. It was producing a wrong result.
Indeed. But it was only in the --stat reporting, so it did not produce an incorrect rebased history.
Quoted 7 lines
A more faithful reimplementation would not have exited with fatal error, but would have produced the same wrong result, so "a better error checking caused the reimplementation die where the original did not die" may be correct but is much less important than the fact that "the logic used in the original to produce diffstat forgot that there may not be a merge base and would not have worked correctly in such a case, and that is what we are correcting" is more important.
True.
Quoted 4 lines
Please descibe the issue as such, if that is the case (I cannot read from the description in the proposed log message if that is the case---but I came to the conclusion that it is what you wanted to fix reading the code).
Indeed, my commit message is a bit too close to what I fixed, and not enough about what needed to be fixed.
Quoted 8 lines
builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1487
Quoted 5 lines
builtin/rebase.c +2
-1if (options.flags &REBASE_VERBOSE ) printf(_("Changes from %s to %s:\n"), - oid_to_hex(&merge_base), + is_null_oid(&merge_base) ? + "(empty)" : oid_to_hex(&merge_base),I am not sure (empty) is a good idea for two reasons. It is going to be inserted into an translated string without translation.
Oooops.
Quoted 2 lines
Also it is too similar to the fallback string used by some printf implementations when NULL was given to %s by mistake.
You mean (null)? That was actually intentional, I just thought that (empty) would be different enough...
Quoted 3 lines
If there weren't i18n issues, I would suggest to use "empty merge base" or "empty tree" instead, but the old one would have shown 0{40}, so perhaps
empty_tree_oid_hex()is a good substitute?
As this is a user-facing message, I'd rather avoid something like empty_tree_oid_hex(), which to every Git user who does not happen to be a Git developer, too, must sound very cryptic.
But I guess that I should not be so lazy and really use two different messages here:
Changes from <merge-base> to <onto>
and if there is no merge base,
Changes in <onto>
Will fix.
Quoted 5 lines
Quoted 5 lines
builtin/rebase.c +0
-1@@ -1494,8 +1495,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)DIFF_FORMAT_SUMMARY |DIFF_FORMAT_DIFFSTAT ; opts.detect_rename =DIFF_DETECT_RENAME ;diff_setup_done (&opts); - diff_tree_oid(&merge_base, &options.onto->object.oid,
Quoted 8 lines
builtin/rebase.c int cmd_rebase(int argc, const char **argv, const char *prefix) · line 1500
Quoted 4 lines
builtin/rebase.c +3
-1- "", &opts); +diff_tree_oid (is_null_oid (&merge_base) ? + the_hash_algo->empty_tree : &merge_base, + &options.onto->object.oid,"" , &opts);The original would have run "git diff '' $onto", and died with an error message, so both the original and the reimplementation were failing. Just in different ways ;-)
Right.
Quoted 11 lines
Quoted 11 lines
git-legacy-rebase.sh +2
-1@@ -718,10 +718,12 @@ if test -n "$diffstat" then if test -n "$verbose" then - echo "$(eval_gettext "Changes from \$mb to \$onto:")" + mb_display="${mb:-(empty)}" + echo "$(eval_gettext "Changes from \$mb_display to \$onto:")"
Quoted 8 lines
git-legacy-rebase.sh if test -n "$diffstat" · line 726
Quoted 5 lines
git-legacy-rebase.sh +2
-1fi + mb_tree="${mb:-$(git hash-object -t tree /dev/null)}" # We want color (if set), but no pager - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto"Code fix for diff-tree invocation, both in the builtin/ version and this one, looks correct.
Okay. I fixed the two things you pointed out, just waiting for the Linux phase to finish (I don't think there is anything OS-specific in this patch, so I trust macOS and Windows to pass if Linux passes):
https://git-for-windows.visualstudio.com/git/_build/results?buildId=26116
Ciao,
Dscho
v2 cover Fix git rebase --stat -i <unrelated-history>
In reply to Johannes Schindelin via GitGitGadgetWe're really entering obscure territory here, I would say.
To trigger the bug, two things have to come together: the user must have asked for a diffstat afterwards, and the commits need to have been rebased onto a completely unrelated commit history (i.e. there must exist no merge base between the pre-rebase HEAD and the post-rebase HEAD).
Please note that this bug existed already in the scripted rebase, but it was never detected because the scripted version would not even perform any error checking.
It will make Junio very happy that I squashed the regression test in to the patch that fixes it. The reason, however, was not to make Junio happy (I hope to make him happy by fixing bugs), but simply that an earlier iteration of test would only fail with the built-in rebase, but not with the scripted version. The current version would fail with the scripted version, but I'll save the time to split the patch again now.
Changes since v1:
* The commit message now talks more about what we should do in case that there is no merge base, rather than stressing the differences between the way scripted vs built-in rebase handled it (both buggy, and fixed by this patch).
* In case that there is no merge base, it no longer reports "Changes from (empty) to ..." but "Changes to ...", which should be a lot less controversial.
Johannes Schindelin (1):
rebase --stat: fix when rebasing to an unrelated history
builtin/rebase.c | 18 ++++++++++++------ git-legacy-rebase.sh | 10 ++++++++-- t/t3406-rebase-message.sh | 10 ++++++++++ 3 files changed, 30 insertions(+), 8 deletions(-)
base-commit: a1598010f775d82b5adf12c29d0f5bc9b41434c6
Published-As: https://github.com/gitgitgadget/git/releases/tags/pr-88%2Fdscho%2Frebase-stat-fix-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-88/dscho/rebase-stat-fix-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/88
Range-diff vs v1:
1: 680385e4bd ! 1: 190c7856ad rebase --stat: fix when rebasing to an unrelated history
@@ -3,22 +3,21 @@
rebase --stat: fix when rebasing to an unrelated history
When rebasing to a commit history that has no common commits with the
- current branch, there is no merge base. The scripted version of the `git
- rebase` command was not prepared for that and spewed out
+ current branch, there is no merge base. In diffstat mode, this means
+ that we cannot compare to the merge base, but we have to compare to the
+ empty tree instead.
- fatal: ambiguous argument '': unknown revision or path not in
- the working tree.
+ Also, if running in verbose diffstat mode, we should not output
- but then continued (due to lack of error checking).
+ Changes from <merge-base> to <onto>
- The built-in version of the `git rebase` command blindly translated that
- shell script code, assuming that there is no need to test whether there
- *was* a merge base, and due to its better error checking, exited with a
- fatal error (because it tried to read the object with hash 00000000...
- as a tree).
+ as that does not make sense without any merge base.
- Fix both scripted and built-in versions to output something sensibly,
- and add a regression test to keep this working in all eternity.
+ Note: neither scripted nor built-in versoin of `git rebase` were
+ prepared for this situation well. We use this opportunity not only to
+ fix the bug(s), but also to make both versions' output consistent in
+ this instance. And add a regression test to keep this working in all
+ eternity.
Reported-by: Ævar Arnfjörð Bjarmason <[email protected]>
Signed-off-by: Johannes Schindelin <[email protected]>
@@ -27,15 +26,25 @@
--- a/builtin/rebase.c
+++ b/builtin/rebase.c
@@
+ if (options.flags & REBASE_DIFFSTAT) {
+ struct diff_options opts;
- if (options.flags & REBASE_VERBOSE)
- printf(_("Changes from %s to %s:\n"),
+- if (options.flags & REBASE_VERBOSE)
+- printf(_("Changes from %s to %s:\n"),
- oid_to_hex(&merge_base),
-+ is_null_oid(&merge_base) ?
-+ "(empty)" : oid_to_hex(&merge_base),
- oid_to_hex(&options.onto->object.oid));
+- oid_to_hex(&options.onto->object.oid));
++ if (options.flags & REBASE_VERBOSE) {
++ if (is_null_oid(&merge_base))
++ printf(_("Changes to %s:\n"),
++ oid_to_hex(&options.onto->object.oid));
++ else
++ printf(_("Changes from %s to %s:\n"),
++ oid_to_hex(&merge_base),
++ oid_to_hex(&options.onto->object.oid));
++ }
/* We want color (if set), but no pager */
+ diff_setup(&opts);
@@
DIFF_FORMAT_SUMMARY | DIFF_FORMAT_DIFFSTAT;
opts.detect_rename = DIFF_DETECT_RENAME;
@@ -57,8 +66,12 @@
if test -n "$verbose"
then
- echo "$(eval_gettext "Changes from \$mb to \$onto:")"
-+ mb_display="${mb:-(empty)}"
-+ echo "$(eval_gettext "Changes from \$mb_display to \$onto:")"
++ if test -z "$mb"
++ then
++ echo "$(eval_gettext "Changes to \$onto:")"
++ else
++ echo "$(eval_gettext "Changes from \$mb to \$onto:")"
++ fi
fi
+ mb_tree="${mb:-$(git hash-object -t tree /dev/null)}"
# We want color (if set), but no pager
@@ -81,7 +94,7 @@
+ git -C unrelated remote add -f origin "$PWD" &&
+ git -C unrelated branch --set-upstream-to=origin/master &&
+ git -C unrelated -c core.editor=true rebase -i -v --stat >actual &&
-+ test_i18ngrep "Changes from (empty)" actual &&
++ test_i18ngrep "Changes to " actual &&
+ test_i18ngrep "5 files changed" actual
+'
+Signature · 1 line
gitgitgadget
PATCH v2 1/1 rebase --stat: fix when rebasing to an unrelated history
From: Johannes Schindelin <[email protected]>
When rebasing to a commit history that has no common commits with the current branch, there is no merge base. In diffstat mode, this means that we cannot compare to the merge base, but we have to compare to the empty tree instead.
Also, if running in verbose diffstat mode, we should not output
Changes from <merge-base> to <onto>
as that does not make sense without any merge base.
Note: neither scripted nor built-in versoin of git rebase were prepared for this situation well. We use this opportunity not only to fix the bug(s), but also to make both versions' output consistent in this instance. And add a regression test to keep this working in all eternity.
--- builtin/rebase.c | 18 ++++++++++++------ git-legacy-rebase.sh | 10 ++++++++-- t/t3406-rebase-message.sh | 10 ++++++++++ 3 files changed, 30 insertions(+), 8 deletions(-)
builtin/rebase.c +12 -6
@@ -1481,10 +1481,15 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)if (options.flags &REBASE_DIFFSTAT ) {struct diff_options opts; -if (options.flags &REBASE_VERBOSE ) -printf (_ ("Changes from %s to %s:\n" ), -oid_to_hex (&merge_base), -oid_to_hex (&options.onto->object.oid)); +if (options.flags &REBASE_VERBOSE ) { +if (is_null_oid (&merge_base)) +printf (_ ("Changes to %s:\n" ), +oid_to_hex (&options.onto->object.oid)); +else +printf (_ ("Changes from %s to %s:\n" ), +oid_to_hex (&merge_base), +oid_to_hex (&options.onto->object.oid)); + }/* We want color (if set), but no pager */ diff_setup (&opts);
@@ -1494,8 +1499,9 @@ int cmd_rebase(int argc, const char **argv, const char *prefix)DIFF_FORMAT_SUMMARY |DIFF_FORMAT_DIFFSTAT ; opts.detect_rename =DIFF_DETECT_RENAME ;diff_setup_done (&opts); -diff_tree_oid (&merge_base, &options.onto->object.oid, -"" , &opts); +diff_tree_oid (is_null_oid (&merge_base) ? + the_hash_algo->empty_tree : &merge_base, + &options.onto->object.oid,"" , &opts);diffcore_std (&opts);diff_flush (&opts); }
git-legacy-rebase.sh +8 -2
@@ -718,10 +718,16 @@ if test -n "$diffstat" then if test -n "$verbose" then - echo "$(eval_gettext "Changes from \$mb to \$onto:")" + if test -z "$mb" + then + echo "$(eval_gettext "Changes to \$onto:")" + else + echo "$(eval_gettext "Changes from \$mb to \$onto:")" + fi fi + mb_tree="${mb:-$(git hash-object -t tree /dev/null)}" # We want color (if set), but no pager - GIT_PAGER='' git diff --stat --summary "$mb" "$onto" + GIT_PAGER='' git diff --stat --summary "$mb_tree" "$onto" fi test -n "$interactive_rebase" && run_specific_rebase
t/t3406-rebase-message.sh +10 -0
@@ -91,4 +91,14 @@ test_expect_success 'error out early upon -C<n> or --whitespace=<bad>' ' test_i18ngrep "Invalid whitespace option" err ' +test_expect_success 'rebase -i onto unrelated history' ' + git init unrelated && + test_commit -C unrelated 1 && + git -C unrelated remote add -f origin "$PWD" && + git -C unrelated branch --set-upstream-to=origin/master && + git -C unrelated -c core.editor=true rebase -i -v --stat >actual && + test_i18ngrep "Changes to " actual && + test_i18ngrep "5 files changed" actual +' + test_done
Signature · 1 line
gitgitgadget
Re: [PATCH 1/1] rebase --stat: fix when rebasing to an unrelated history
In reply to Johannes SchindelinQuoted 8 lines · Johannes Schindelin
But I guess that I should not be so lazy and really use two different messages here:
Changes from <merge-base> to <onto>
and if there is no merge base,
Changes in <onto>
Ah, that's excellent.
Thanks.