Fix amending commits whose commit message is empty

[aidancz]

• PR Description

When the commit message is empty, the --amend command will produce an error.
Solution: Add the --allow-empty-message option.

• Please check if the PR fulfills these requirements

• Cheatsheets are up-to-date (run go generate ./...)
• Code has been formatted (see here)
• Tests have been added/updated (see here for the integration test guide)
• Text is internationalised (see here)
• If a new UserConfig entry was added, make sure it can be hot-reloaded (see here)
• Docs have been updated if necessary
• You've read through your own file changes for silly mistakes etc

[stefanhaller]

This makes sense to me for AmendHeadCmdObj, ResetAuthor, and SetAuthor. If you have a commit whose message is already empty, it should be possible to add file changes to it or reset its author without getting an error. This can be seen as a bug fix.

I am a little less sure about RewordLastCommit and RewordLastCommitInEditor; this would now allow rewording a commit to make its message empty, which wasn't possible before. To me this would be more of a behavior change than a bug fix.

What's your use case? Why do you want to work with commits with empty messages at all? You didn't give a rationale for this in #4643 either.

[aidancz]

I sometimes make commits with an empty message in the following situation:

When a stage of work is completed, but the overall task isn’t finished yet.

The reason is:

After committing, I no longer need to worry about losing changes. An empty commit message stands out clearly in the commit history, reminding me that this is just a partial milestone. Later, I can squash these into a single commit with a proper message.

Regarding the behavior change in LazyGit:

For #4643, the c enter key sequence is very easy to press, but it’s currently underutilized. Having it create a commit with an empty message seems like the most reasonable behavior in that UI context ,after all, Enter means “confirm.” If the user wants to cancel the commit, pressing Esc would be more appropriate.

As for this PR, I believe allowing an empty message during rewording is a reasonable change. Here’s why:
1. For RewordLastCommit, the same logic applies: allowing an empty message makes sense in this UI context
2. For RewordLastCommitInEditor, it’s possible that some users might delete all content in the editor to cancel the reword. However, I think that’s not a very healthy habit. If many users do operate this way, perhaps a confirmation dialog could be added?
3. Even if a user makes a mistake, they can easily understand what happened from the command history and just reword again.
4. Finally, if #4643 is updated, this change should be made here as well for consistency.

[stefanhaller]

An empty commit message stands out clearly in the commit history, reminding me that this is just a partial milestone.

I don't understand why it needs to be empty for this. A message such as - would also stand out clearly, and wouldn't have the tooling problems you are trying to solve here.

Having it create a commit with an empty message seems like the most reasonable behavior in that UI context

I disagree with this, as I said in #4643; this has the potential of accidentally creating commits with empty messages, which is something that most people probably don't want to do.

For rewording, it seems less like a problem to me, as you would have to actively delete all text before pressing enter, so this is not something that would happen accidentally. However:

For RewordLastCommitInEditor, it’s possible that some users might delete all content in the editor to cancel the reword.

Yes, this is actually what I do. It sometimes happens that I accidentally hit shift-r instead of r to rename (confusing this with the branches panel, where it's shift-r for renaming a branch), and then deleting all text and saving seems like the only way to cleanly abort the rename. How else would you do it? Most editors don't have a way to say "quit with an error code".

For this reason I'm strongly against allowing empty messages for RewordLastCommitInEditor, and then I would keep RewordLastCommit unchanged as well for consistency.

I'm fine with changing the remaining three commands, if you still find that valuable enough. If so, please fix the failing unit tests.

[aidancz]

I understand your point of view, this PR has been updated

[stefanhaller]

Thanks!