On 09/06/2012 10:16 AM, Michael J Gruber wrote: > Michael Haggerty venit, vidit, dixit 05.09.2012 17:30: >> On 09/05/2012 03:39 PM, Michael J Gruber wrote: >>> git-merge does not honor the pre-commit hook when doing automatic merge >>> commits, and for compatibility reasons this is going to stay. >>> >>> Introduce a pre-merge hook which is called for an automatic merge commit >>> just like pre-commit is called for a non-automatic merge commit (or any >>> other commit). >> >> What exactly is an "automatic merge commit"? Is it any merge that >> doesn't have a conflict? A merge that doesn't invoke the editor? A >> merge done as part of another operation (e.g., pull)? I don't see the >> term mentioned in the git-merge or githooks man pages. >> >> I think it would be good if you would define this term in the >> documentation files that your patch touched, and perhaps in the githooks >> section about "pre-commit" as well. > > "git merge" can go three ways: > > F: fast forward: no commit is created, only a ref is changed > A: automatic: true merge (non-ff) without conflicts (i.e. chosen > strategy can perform the merge); a new commit is created > C: merge with conflicts: no commit is created but the index is prepared > (partially) for a merge commit > > In case F, no commit hook is run (talking only about pre-commit/pre-merge). > > In case A, no commit is run so far but my patch proposes pre-merge to be > run. > > In case C, pre-commit (!) is run so far and after my patch.
Thanks for the explanation. I hope you will explain this briefly in the patch to the docs. >> Secondly, though it is impossible (for backwards compatibility reasons) >> for the pre-commit hook to be invoked for automatic merges, no such >> considerations prohibit the pre-merge commit from being invoked for >> non-automatic merges. In other words, both hooks, pre-commit *and* >> pre-merge, could be invoked for non-automatic merges. Would this be >> preferable? >> >> It depends on what pre-merge scripts are likely to be used for. If they >> will tend to be used for merge-specific actions, then it might be more >> convenient for *all* merges to be vetted by them. On the other hand, if >> they tend to do the same actions as pre-commit hooks, then having >> non-automatic merge commits go through both hooks would tend to be more >> annoying than helpful. Specifically, one of the scripts would probably >> have to check whether the merge is a non-automatic merge, and if so do >> nothing (i.e., letting the other script take care of it). This would >> also require an easy way for a script to determine whether a commit is a >> non-automatic merge commit. >> >> Have you considered this? > > Your second paragraph explains why I did it the way I did. One can > easily have pre-merge call pre-commit, or have them be different. One > can not easily have only pre-merge called for a non-automatic merge > commit, but that is because of backward compatibility. The way *I* would > like it is: > > - call pre-merge for any non-ff merge commit (automatic or not) > - call pre-commit for any non-merge commit (#parents <=1) > > But that would break compatibility. > > So I hope my patch is the best approximation to the above which keeps > compatibility and is simple to handle in most situations. I can understand your reasoning and won't object. But before I shut up, I will point out a third alternative that is arguably closer to your "ideal": - For non-merge commits, call pre-commit - For automatic merge commits, call pre-merge - For non-automatic merge commits: if pre-merge exists, call pre-merge (only) else if pre-commit exists, call pre-commit (for backwards comptibility) Michael -- Michael Haggerty mhag...@alum.mit.edu http://softwareswirl.blogspot.com/ -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html