I've isolated what seems to be a loophole in interactive rebase by rebasing a branch for the sole purpose of rewording some commits:
(I just finished performing a rebase for the sole purpose of squashing several commits into 2 commits) (Now I want to reword those 2 commits)
$ git --version = 1.7.1 $ git rebase -i --onto BranchA~2 BranchA~2 BranchA (change "pick" to "reword" in rebase-stack editor) reword Abbrsha1 some-commit-message reword Bbbrsha1 some-other-commit-message # Rebase Cbbrsha1..Dbbrsha1 onto Cbbrsha1 (save changes in editor) ".git/rebase-merge/git-rebase-todo" 15L, 580C written (shows output from pre-commit hook and the corresponding non-zero exit) Start of pre-commit hook... checking for something I don't like... ERROR: I found something I don't like! Commit Aborted! Start of pre-commit hook... checking for something I don't like... ERROR: I found something I don't like! Commit Aborted! Successfully rebased and updated refs/heads/BranchA. $ _
In this case, the thing that caused the pre-commit to error-out was that the format of a report that it runs was changed. So in this case, I will rebase -i and "edit" the commits and git-add the new report format so the pre-commit doesn't complain. However, I originally tried to do the reword and squash in a single rebase and it gave the same errors, but performed the squashes. Is this the intended behavior? On the one hand, you shouldn't be changing content in a "reword", but on the other hand if you're going to run the pre-commit hook then shouldn't you abort the rebase if the pre-commit hook exits with non-zero status?
I admit that a workflow in which content changes on a reword may not be a valid scenario that you should care about in your rebase exception handling, but shouldn't you care that the pre-commit hook exited with non-zero status regardless of the reason why it exited with non-zero status? Just a question.
v/r, Neal