> Eric Sunshine <sunshine@xxxxxxxxxxxxxx> wrote: > > The idea you proposed in a different thread[2] of exposing > cleanup_message() functionality as a user-facing utility which a hook > can call on an as-needed basis may make more sense(?). > > [2]: https://lore.kernel.org/git/m034onpng4.fsf@xxxxxxxxxxxxxxxxxx/ I don't think that would suffice for our use case. Specifically, the commit-msg hook has no way to know if it's looking at a message that will be washed, or will not. For example, let's say a commit-msg hook aims to enforce a 72-character hard-wrap policy. If a line starts with `#`, what is the hook supposed to do? - If the commit was created using `git commit` and the user's normal editor is invoked, the hook should *not* evaluate any line starting with `#` because that line will be removed prior to creating the commit as part of the message washing process. - If the commit was created using `git commit -m`, the hook *should* evaluate any line starting with `#` because that line will not be removed by washing. This challenge also exists for the patch scissors and any content following it. That could have been added by git because the user called `git commit -v` (most likely), but it also could in theory have been added by the user, in which case the patch (or patch-like content) ends up in the final commit message as well. The hook has no way to know whether the commit was initiated via `git commit -v` or not. The real-world use case here could be a commit-msg hook that adds a trailer: if the patch will be removed during washing, then using `git interpret-trailers` to add the trailer (which places it *before* the patch) is fine. If the patch will not be removed during washing, then `git interpret-trailers` ends up putting the trailer in the middle of the final message between the subject/body and the patch, then the patch isn't removed and thus it's not a valid trailer. Ultimately, the hook doesn't know what will happen to the message after it runs, and thus doesn't know if a given line is guaranteed to be part of the final commit message. This presents these edge case challenges when attempting to implement the hook. If the hook either were always called with the final proposed commit message, or if it knew that the message would be washed after being executed, then we'd have options for handling these edge cases. -- Thanks, Brian Lyles
