Skip to contentSkip to conversation

cherry-pick / pre-commit hook?

DISCUSSION Dave Abrahams 4 participants 11 messages Jun 15, 2016

lore
Share & tools
lore
#0 lore raw

Is there a good reason that git cherry-pick (without --no-commit) doesn't run my pre-commit hook? Is there a hook that cherry-pick /will/ run instead?

Thanks!

Signature · 3 lines
Dave Abrahams
BoostPro Computing
http://www.boostpro.com
#1 lore raw
Quoted 2 lines · Dave Abrahams

Is there a good reason that git cherry-pick (without --no-commit) doesn't run my pre-commit hook?

Interesting question.

 $ git grep -F -e '"cherry-pick"'
 [...]
 git.c:          { "cherry-pick", cmd_cherry_pick, RUN_SETUP | NEED_WORK_TREE },
 $ git grep -F -e cmd_cherry_pick
 builtin.h:extern int cmd_cherry_pick(int argc, const char **argv, const char *prefix);
 builtin/revert.c:int cmd_cherry_pick(int argc, const char **argv, const char *prefix)
 git.c:          { "cherry-pick", cmd_cherry_pick, RUN_SETUP | NEED_WORK_TREE },

cherry-pick is implemented in builtin/revert.c. How does it invoke the "git commit" machinery? Explicitly, as luck would have it.

 $ git grep --show-function -F -h -C5 -e '"commit"' -- builtin/revert.c
 static int run_git_commit(const char *defmsg)
 {
         /* 6 is max possible length of our args array including NULL */
         const char *args[6];
         int i = 0;
         args[i++] = "commit";
         args[i++] = "-n";
         if (signoff)
                 args[i++] = "-s";

[...]

So cherry-pick deliberately uses -n (= --no-verify) when it calls "git commit". Why, though?

 $ git log --oneline --follow -S'"-n"' -- builtin/revert.c
 cfd9c27 Allow cherry-pick (and revert) to add signoff line
 f810379 Make builtin-revert.c use parse_options.
 9509af6 Make git-revert & git-cherry-pick a builtin

The '-n' was copied from the old git-revert.sh script when cherry-pick was made builtin. Not to let the trail grow cold:

 $ git log --oneline --follow -S-n -- git-revert.sh
 9509af6 Make git-revert & git-cherry-pick a builtin
 abd6970 cherry-pick: make -r the default
 674b280 Add documentation for git-revert and git-cherry-pick.
 8bf14d6 Document the --(no-)edit switch of git-revert and git-cherry-pick
 b788498 git-revert: make --edit default.
 e2f5f6e Do not require clean tree when reverting and cherry-picking.
 9fa4db5 Do not verify reverted/cherry-picked/rebased patches.
 [...]
 $ git show -s 9fa4db5
 commit 9fa4db544e2e4d6c931f6adabc5270daec041536
 Author: Junio C Hamano <[email protected]>
 Date:   Mon Aug 29 21:19:04 2005 -0700

Do not verify reverted/cherry-picked/rebased patches.

The original committer may have used validation criteria that is less stricter than yours. You do not want to lose the changes even if they are done in substandard way from your 'commit -v' verifier's point of view.

$

At last, an answer. The main purpose of the pre-commit hook (and builtin checks that preceded it) is to avoid introducing regressions in whitespace style, encoding, and so forth; but it would make cherry-picking unnecessarily difficult, without preventing regressions, to apply the same standards to existing code.

Quoted 2 lines

Is there a hook that cherry-pick
/will/ run instead?

"git log --grep=pre-commit" seems to suggest that the commit-msg and post-commit hooks will be run. But first, what are you trying to accomplish? Maybe there is a simpler way, or maybe with that use case in mind we can make changes to support it better.

Hope that helps,
Jonathan

#2 lore raw

Hi Jonathan,

At Wed, 8 Dec 2010 11:53:24 -0600,

Quoted 13 lines · Jonathan Nieder
 $ git show -s 9fa4db5
 commit 9fa4db544e2e4d6c931f6adabc5270daec041536
 Author: Junio C Hamano <[email protected]>
 Date:   Mon Aug 29 21:19:04 2005 -0700

Do not verify reverted/cherry-picked/rebased patches.

The original committer may have used validation criteria that is less stricter than yours. You do not want to lose the changes even if they are done in substandard way from your 'commit -v' verifier's point of view.

Quoted 8 lines

$

At last, an answer. The main purpose of the pre-commit hook (and builtin checks that preceded it) is to avoid introducing regressions in whitespace style, encoding, and so forth; but it would make cherry-picking unnecessarily difficult, without preventing regressions, to apply the same standards to existing code.

I suspected as much.

Quoted 7 lines
Quoted 3 lines

Is there a hook that cherry-pick
/will/ run instead?

"git log --grep=pre-commit" seems to suggest that the commit-msg and post-commit hooks will be run. But first, what are you trying to accomplish?

You're going to love this: I had sent a pull request upstream and the maintainer of the project rejected my changes because I didn't follow some formatting convention he didn't tell me about ;-). So I set up a commit hook that would prevent me from making the same mistake again, and cherry-picked the changes one-by-one. So it was exactly the same scenario, except I am the author of the original changes.

I wonder whether this would have gone better had I used rebase.

Quoted 2 lines

Maybe there is a simpler way, or maybe with that use case in mind we can make changes to support it better.

Looking forward to hearing more.

Thanks,

Signature · 3 lines
Dave Abrahams
BoostPro Computing
http://www.boostpro.com
#3 lore raw
Quoted 5 lines · Dave Abrahams

You're going to love this: I had sent a pull request upstream and the maintainer of the project rejected my changes because I didn't follow some formatting convention he didn't tell me about ;-). So I set up a commit hook that would prevent me from making the same mistake again, and cherry-picked the changes one-by-one.

Funny. Maybe "cherry-pick --no-commit" followed by ordinary commit would be appropriate? That way, when the checks fail, you are in a position to clean them up.

If the conventions were whitespace related, "git rebase --whitespace=fix" might be even more useful.

Just for kicks, here is the cherry-pick --verify for picky cherry-pickers.

-- 8< --
Subject: cherry-pick/revert: learn --verify to run pre-commit and commit-msg hooks

The main purpose of the pre-commit and commit-msg hooks is to avoid introducing regressions in whitespace style, encoding, and so forth; and it would make cherry-picking unnecessarily difficult, without preventing regressions, to unconditionally apply the same standards to existing code. For this reason, in v0.99.6~51 (2005-08-29), git learned to skip the usual hooks when cherry-picking or reverting an existing commit.

But sometimes the checks are wanted anyway. For example, with this patch applied, you can safely fetch some new contributor's code:

	$ git cherry-pick -s --verify HEAD..FETCH_HEAD

while allowing the pre-commit and commit-msg hooks to run their usual checks so the result can error out if the patches are not clean.

---
Untested. Please feel free to add some documentation and tests and submit it for real if this looks like a good idea. :)

 builtin/revert.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)
builtin/revert.c +4 -2
index bb6e9e8..511b2ea 100644
@@ -36,7 +36,7 @@ static const char * const cherry_pick_usage[] = {
 	NULL
 };
 
-static int edit, no_replay, no_commit, mainline, signoff, allow_ff;
+static int edit, no_replay, no_commit, verify, mainline, signoff, allow_ff;
 static enum { REVERT, CHERRY_PICK } action;
 static struct commit *commit;
 static int commit_argc;
@@ -67,6 +67,7 @@ static void parse_args(int argc, const char **argv)
 		OPT_INTEGER('m', "mainline", &mainline, "parent number"),
 		OPT_RERERE_AUTOUPDATE(&allow_rerere_auto),
 		OPT_STRING(0, "strategy", &strategy, "strategy", "merge strategy"),
+		OPT_BOOLEAN(0, "verify", &verify, "let hooks intervene before commiting"),
 		OPT_END(),
 		OPT_END(),
 		OPT_END(),
@@ -375,7 +376,8 @@ static int run_git_commit(const char *defmsg)
 	int i = 0;
 
 	args[i++] = "commit";
-	args[i++] = "-n";
+	if (!verify)
+		args[i++] = "-n";
 	if (signoff)
 		args[i++] = "-s";
 	if (!edit) {
Signature · 1 line
1.7.2.4
#4 lore raw

At Wed, 8 Dec 2010 16:05:14 -0600,

Quoted 14 lines · Jonathan Nieder

The main purpose of the pre-commit and commit-msg hooks is to avoid introducing regressions in whitespace style, encoding, and so forth; and it would make cherry-picking unnecessarily difficult, without preventing regressions, to unconditionally apply the same standards to existing code. For this reason, in v0.99.6~51 (2005-08-29), git learned to skip the usual hooks when cherry-picking or reverting an existing commit.

But sometimes the checks are wanted anyway. For example, with this patch applied, you can safely fetch some new contributor's code:

	$ git cherry-pick -s --verify HEAD..FETCH_HEAD
Quoted 7 lines

while allowing the pre-commit and commit-msg hooks to run their usual checks so the result can error out if the patches are not clean.

---
Untested. Please feel free to add some documentation and tests and submit it for real if this looks like a good idea. :)

Well, thanks, but sadly I can only invest enough time to file this bug report right now: if you're going to have a "pre-commit hook" concept, but not run that hook for some kinds of commits, then that fact needs to be documented.

Signature · 3 lines
Dave Abrahams
BoostPro Computing
http://www.boostpro.com
#5 lore raw
Quoted 3 lines · Dave Abrahams

if you're going to have a "pre-commit hook" concept,
but not run that hook for some kinds of commits, then that fact needs to be documented.

True, and thanks for a reminder. Suggested wording?

The current githooks(5) says

pre-commit
This hook is invoked by git commit, and can be bypassed with --no-verify option.

and leaves the question of whether it is invoked by git cherry-pick unanswered.

#6 lore raw
Quoted 8 lines · Jonathan Nieder
Quoted 3 lines · Dave Abrahams

if you're going to have a "pre-commit hook" concept,
but not run that hook for some kinds of commits, then that fact needs to be documented.

True, and thanks for a reminder. Suggested wording?

Quoted 8 lines

The current githooks(5) says

pre-commit
This hook is invoked by git commit, and can be bypassed with --no-verify option.

and leaves the question of whether it is invoked by git cherry-pick unanswered.

Huh? Isn't it very clear that "git commit" calls it and "git status" or anything that is not "git commit" doesn't?

#7 lore raw
Quoted 16 lines · Junio C Hamano
Quoted 14 lines · Jonathan Nieder
Quoted 3 lines · Dave Abrahams

if you're going to have a "pre-commit hook" concept,
but not run that hook for some kinds of commits, then that fact needs to be documented.

True, and thanks for a reminder.  Suggested wording?

The current githooks(5) says

 pre-commit
      This hook is invoked by git commit, and can be bypassed with --no-verify option.

Quoted 5 lines
Quoted 2 lines

and leaves the question of whether it is invoked by git cherry-pick unanswered.

Huh?  Isn't it very clear that "git commit" calls it and "git status" or anything that is not "git commit" doesn't?

Not to me. And is it really true? What about rebase? What about merge?

Signature · 3 lines
Dave Abrahams
BoostPro Computing
http://www.boostpro.com
#8 lore raw
In reply to Jonathan Nieder
Quoted 8 lines · Jonathan Nieder
Quoted 3 lines · Dave Abrahams

if you're going to have a "pre-commit hook" concept,
but not run that hook for some kinds of commits, then that fact needs to be documented.

True, and thanks for a reminder. Suggested wording?

Quoted 8 lines

The current githooks(5) says

pre-commit
This hook is invoked by git commit, and can be bypassed with --no-verify option.

and leaves the question of whether it is invoked by git cherry-pick unanswered.

Yeah, but do we want to answer it in this section, or in git-cherry-pick's manual page?

After all "pre-commit" was _not_ about "before doing anything that creates a commit", but was about "before git-commit creates a commit". Changing that definition is fine as long as we have a good way to explain it to the users and more importantly a simple rule that the users can understand, and that rule _could_ be "anything that creates a new commit".

In reality "anything that creates a new commit" cannot be that simple rule, however. For example, "git pull" does attempt to create a new merge commit, but failing it because the work done by the other person you are pulling from does not conform _your_ standard defined by your hook is not a sane default.

I think the basic direction could be (I haven't thought things through, just a strawman):

- Allow --verify/--no-verify to all commands that possibly create a new commit, and run pre-commit hook where an updated index is about to be made into a commit (for some commands this may not be very easy);

- The guideline of picking the default would probably look like this:

(1) for existing commands, keep the current behaviour;

(2) for a new command, --verify should be the default if the command is primarily about letting the user do what s/he would/could/should have done as "git commit" in the first place (e.g. cherry-picking one's own commit from a separate branch or rebasing one's own unpublished branch on top of updated upstream), and --no-verify otherwise (i.e. taking other's work and using it in a context different from the original).

#9 lore raw
Quoted 11 lines · Junio C Hamano

I think the basic direction could be (I haven't thought things through, just a strawman):

- Allow --verify/--no-verify to all commands that possibly create a new commit, and run pre-commit hook where an updated index is about to be made into a commit (for some commands this may not be very easy);

- The guideline of picking the default would probably look like this:

(1) for existing commands, keep the current behaviour;

Quoted 7 lines

(2) for a new command, --verify should be the default if the command is primarily about letting the user do what s/he would/could/should have done as "git commit" in the first place (e.g. cherry-picking one's own commit from a separate branch or rebasing one's own unpublished branch on top of updated upstream), and --no-verify otherwise (i.e. taking other's work and using it in a context different from the original).

Does it mean that for now (and perhaps also for later) it means that "git commit" by default runs pre-commit hook, unless one use --no-verify, and that all comands that create a new commit (rebase, cherry-pick, revert, merge/pull) can request for pre-commit hook to be run (if they create commit) with --verify?

I think it is a very good idea, though I don't know how difficult it would be to make all commands that can create commit accept --verify..

Signature · 3 lines
Jakub Narebski
Poland
ShadeHawk on #git
#10 lore raw

At Tue, 28 Dec 2010 14:38:20 -0800 (PST),

Quoted 22 lines · Jakub Narebski
Quoted 11 lines · Junio C Hamano

I think the basic direction could be (I haven't thought things through, just a strawman):

- Allow --verify/--no-verify to all commands that possibly create a new commit, and run pre-commit hook where an updated index is about to be made into a commit (for some commands this may not be very easy);

- The guideline of picking the default would probably look like this:

(1) for existing commands, keep the current behaviour;

Quoted 7 lines

(2) for a new command, --verify should be the default if the command is primarily about letting the user do what s/he would/could/should have done as "git commit" in the first place (e.g. cherry-picking one's own commit from a separate branch or rebasing one's own unpublished branch on top of updated upstream), and --no-verify otherwise (i.e. taking other's work and using it in a context different from the original).

Quoted 7 lines

Does it mean that for now (and perhaps also for later) it means that "git commit" by default runs pre-commit hook, unless one use --no-verify, and that all comands that create a new commit (rebase, cherry-pick, revert, merge/pull) can request for pre-commit hook to be run (if they create commit) with --verify?

I think it is a very good idea

+1!

Signature · 3 lines
Dave Abrahams
BoostPro Computing
http://www.boostpro.com