Skip to contentSkip to conversation

git-prompt: fix sequencer/todo detection

PATCH In review Danny Lin 3 participants 5 messages Mar 25, 2022 – Apr 7, 2022

lore
Share & tools
lore

Danny Lin

author
#0 lore raw

PATCH 1/1 git-prompt: fix sequencer/todo detection

Previous case does not correctly check the "p ..." pattern.

---
 contrib/completion/git-prompt.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/completion/git-prompt.sh +1 -1
index db7c0068fb..8ae341a306 100644
@@ -315,7 +315,7 @@ __git_sequencer_status ()
 	elif __git_eread "$g/sequencer/todo" todo
 	then
 		case "$todo" in
-		p[\ \	]|pick[\ \	]*)
+		p[\ \	]*|pick[\ \	]*)
 			r="|CHERRY-PICKING"
 			return 0
 		;;
Signature · 1 line
2.35.1.windows.2
#1 lore raw
Quoted 12 lines · Danny Lin

Previous case does not correctly check the "p ..." pattern.

---
 contrib/completion/git-prompt.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/completion/git-prompt.sh +0 -0
index db7c0068fb..8ae341a306 100644

contrib/completion/git-prompt.sh __git_sequencer_status () · line 321

Quoted 8 lines
contrib/completion/git-prompt.sh +1 -1
@@ -315,7 +315,7 @@ __git_sequencer_status ()
 	elif __git_eread "$g/sequencer/todo" todo
 	then
 		case "$todo" in
-		p[\ \	]|pick[\ \	]*)
+		p[\ \	]*|pick[\ \	]*)
 			r="|CHERRY-PICKING"
 			return 0
 		;;

The original obviously is broken ;-) Will queue. Thanks.

#2 lore raw
In reply to Danny Lin
Quoted 12 lines · Danny Lin

Previous case does not correctly check the "p ..." pattern.

---
 contrib/completion/git-prompt.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/completion/git-prompt.sh +0 -0
index db7c0068fb..8ae341a306 100644

contrib/completion/git-prompt.sh __git_sequencer_status () · line 321

Quoted 8 lines
contrib/completion/git-prompt.sh +1 -1
@@ -315,7 +315,7 @@ __git_sequencer_status ()
 	elif __git_eread "$g/sequencer/todo" todo
 	then
 		case "$todo" in
-		p[\ \	]|pick[\ \	]*)
+		p[\ \	]*|pick[\ \	]*)
 			r="|CHERRY-PICKING"
 			return 0
 		;;

It is obvious that the original code is *not* prepared to see 'p' followed by whitespace followed by other things, but I am not sure how the code in sequencer.c::todo_list_write_to_file() can choose to pass flags & TODO_LIST_ABBREVIATE_CMDS to todo_list_to_strbuf().

Danny, do you have a reproduction recipe, preferrably one you can turn into a new test in t9903-bash-prompt.sh? Or was this found merely by inspecting the code?

Dscho, as far as I can tell, builtin/rebase.c can set the bit in the flags word when rebase.abbreviatecommands configuration is set, but that configuration variable is about rebase and it shouldn't affect how multi-step cherry-pick would work, should it? I am wondering if an uninitialized "flags" word, whose TODO_LIST_ABBREVIATE_CMDS bit randomly was turned on, caused todo_list_to_strbuf() to write an abbreviated insn in the todo file. If so, the insn word being abbreviated or fully spelled out would not affect the correctness, but the flags word affects other things that are more crucial to correctness, so...

Thanks.

#3 lore raw

Hi Junio,

Quoted 19 lines · Junio C Hamano
Quoted 17 lines · Danny Lin

Previous case does not correctly check the "p ..." pattern.

---
 contrib/completion/git-prompt.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/completion/git-prompt.sh +1 -1
index db7c0068fb..8ae341a306 100644
@@ -315,7 +315,7 @@ __git_sequencer_status ()
 	elif __git_eread "$g/sequencer/todo" todo
 	then
 		case "$todo" in
-		p[\ \	]|pick[\ \	]*)
+		p[\ \	]*|pick[\ \	]*)
Quoted 8 lines

contrib/completion/git-prompt.sh __git_sequencer_status () · line 321

Quoted 3 lines
contrib/completion/git-prompt.sh +0 -0
 			r="|CHERRY-PICKING"
 			return 0
 		;;

It is obvious that the original code is *not* prepared to see 'p' followed by whitespace followed by other things, but I am not sure how the code in sequencer.c::todo_list_write_to_file() can choose to pass flags & TODO_LIST_ABBREVIATE_CMDS to todo_list_to_strbuf().

At first, I thought that if rebase.abbreviateCommands is set to true, we would write out todo lists with one-letter commands.

But that's not true, it's only while we edit the file that that config setting affects how the todo list is written.

I _guess_ it is during that time window that the prompt does not work as expected?

Quoted 3 lines

Danny, do you have a reproduction recipe, preferrably one you can turn into a new test in t9903-bash-prompt.sh? Or was this found merely by inspecting the code?

I _guess_ there is a script at play which rewrites the todo file and which runs when a cherry-pick fails due to merge conflicts. Here is a patch that will exercise the code and verify the fix:

-- snip --

t/t9903-bash-prompt.sh +4 -1
index bbd513bab0f..c5c3e3c83de 100755
@@ -221,7 +221,10 @@ test_expect_success 'prompt - cherry-pick' '
 	git reset --merge &&
 	test_must_fail git rev-parse CHERRY_PICK_HEAD &&
 	__git_ps1 >"$actual" &&
-	test_cmp expected "$actual"
+	test_cmp expected "$actual" &&
+	echo "p HEAD" >.git/sequencer/todo &&
+	__git_ps1 >"$actual".2 &&
+	test_cmp expected "$actual".2
 '
 
 test_expect_success 'prompt - revert' '

-- snap --

Quoted 4 lines

Dscho, as far as I can tell, builtin/rebase.c can set the bit in the flags word when rebase.abbreviatecommands configuration is set, but that configuration variable is about rebase and it shouldn't affect how multi-step cherry-pick would work, should it?

Indeed. In a multi-commit cherry-pick, we call walk_revs_populate_todo() which uses the long form of the pick command, always (look for command_string).

Quoted 3 lines

I am wondering if an uninitialized "flags" word, whose TODO_LIST_ABBREVIATE_CMDS bit randomly was turned on, caused todo_list_to_strbuf() to write an abbreviated insn in the todo file.

I don't think that todo_list_to_strbuf() is called during a cherry-pick. Instead, walk_revs_populate_todo() is called in git cherry-pick --continue, and it does not modify the todo commands if run in non-rebase-i mode.

Quoted 3 lines

If so, the insn word being abbreviated or fully spelled out would not affect the correctness, but the flags word affects other things that are more crucial to correctness, so...

It looks to me as if the abbreviated commands cannot be generated by Git (the replay_opts in builtin/revert.c are all initialized to REPLAY_OPTS_INIT, so there is not even any chance of uninitialized data there).

Ciao,
Dscho

#4 lore raw
Quoted 15 lines · Johannes Schindelin
Quoted 6 lines
Quoted 6 lines
contrib/completion/git-prompt.sh +0 -0
index db7c0068fb..8ae341a306 100644
@@ -315,7 +315,7 @@ __git_sequencer_status ()
 	elif __git_eread "$g/sequencer/todo" todo
Quoted 8 lines
Quoted 7 lines
contrib/completion/git-prompt.sh +1 -1
 	then
 		case "$todo" in
-		p[\ \	]|pick[\ \	]*)
+		p[\ \	]*|pick[\ \	]*)
 			r="|CHERRY-PICKING"
 			return 0
 		;;

...

Quoted 4 lines

It looks to me as if the abbreviated commands cannot be generated by Git (the replay_opts in builtin/revert.c are all initialized to REPLAY_OPTS_INIT, so there is not even any chance of uninitialized data there).

Good. Then the right "fix" would be to drop the misleading "We would also accept the abbreviated" side of the case arm, instead of fixing "if we were generating the abbreviated one, here is how we might support it" code, I guess.

Thanks for sanity-checking my digging in the sequencer.c code.