Re: Proposal: tell git a file has been renamed

From: Chris Torek <chris.torek@gmail.com>
Date: Apr 24, 2023, 01:43 UTC
Message-ID: <CAPx1Gvdc6bqzt2PpqD1Z4e5w+b=8gZhUSyfUQC1n8QazdBacEw@mail.gmail.com>
In-Reply-To: <8fe188a9-c01f-9fb5-5877-8ff508094b22@game-point.net>

(By the way, apologies for the stuttering "gitt" typos that seem to keep creeping into my messages here. I'm not sure what's causing them -- I make a lot of typos but this is abnormal!)

On Sat, Apr 22, 2023 at 12:22 PM Jeremy Morton <admin@game-point.net> wrote:

> [renames are a problem]

I have long wondered if there was a way to improve this experience myself.

(I also note that most of the followup messages up to this point have missed the point. It's true that you can run `git mv` and `git commit`, but you've already said that this becomes particularly painful when you realize that it's appropriate *after the fact*, when you've already made intermediate commits and/or staged changes or whatever.)

Show 13 quoted lines

> The standard answer for this is to rename the file in one commit, then
> make the changes.  That's fine if you know ahead of time you'll want
> to do this.  However it's a total PITA if you have a bunch of changes
> and you realize that a rename has caused this problem.  You now have
> to back out your changes to the renamed file, add the rename, commit
> it, then re-apply the changes.
>
> Could a command be added to git that means you tell Git that counts as
> a file rename?  Git would add a marker to the staging area that the
> file has been renamed, and upon commit, would first generate an
> additional commit for each rename before generating the main commit,
> ensuring the rename operation counts as an actual rename, and the
> content's history is maintained.

The index *currently* has no room to store anything like this: it is, in effect, just the proposed next commit, stored as a flattened tree. There are, however, extra marker records that can be added. So:

If `git mv` (or a new command) had a flag to say "make a special index entry so that the next `git commit` does a double commit", we could in fact make this work. Alternatively, we could have a command -- similar to `git commit --only` in effect -- that uses the current (HEAD) commit to construct a renames-only commit, in which 100%-identical-file matching would (in general) find the desired renames -- and make it, perhaps also co-ordinating with `git mv` of existing files in the index. (I'd also like to have `git mv --after`, in the same vein as `hg mv --after`; I long ago wrote a cheesy script to achieve this, but it would be nice to have a proper command.)

On top of this, it might be nice to have a standardized commit message and/or other marker (in the commit header?) for a "rename- only" commit, which this kind of extra-rename-commit operation would use. Then `git log` and `git blame` and other commands could easily detect such commits and default to an automatic `--follow` style following, and `git log` might be allowed to omit the *display* of such a commit by default, by showing all the renames as renames in the subsequent commit (though this would presumably require an internal verification step to check for spoofed renames that are not in fact rename-only operations).

In any case, this *idea* is easy, like many ideas. It really comes down to implementation. If someone thinks this is a great idea, someone (perhaps me) should work on *implementing* it. :-)

Chris