Re: GSoC 2019: Git's application submitted

From: Johannes Schindelin <johannes.schindelin@gmx.de>
Date: Feb 7, 2019, 19:39 UTC
Message-ID: <nycvar.QRO.7.76.6.1902072023250.41@tvgsbejvaqbjf.bet>
In-Reply-To: <20190206220942.GE6085@hank.intra.tgummerer.com>

Hi Thomas,

On Wed, 6 Feb 2019, Thomas Gummerer wrote:

> I think the idea here could definitely be split into a couple different
> phases, that could be individually useful, and can be merged
> individually, though I don't know if they would necessarily be.

Good idea.

Show 7 quoted lines

> Of the top of my head:
> 
>  - write test_expect_failure tests for the expected new behaviour
> 
>    This may not be worth including in git.git yet, but it can be a
>    very useful starting point for somebody else continuing the feature
>    if the student finds they don't have time for it.

I like this approach.

Show 8 quoted lines

>  - implement pushing the index state, without dealing with conflicts
>  - implement poping the index state, without dealing with conflicts
> 
>    This can already be individually useful, and I think this is
>    something people asked for on the mailing list, though I didn't try
>    digging up old threads for now.  After these two steps stashing and
>    restoring a merge conflict would still not work, but we have a good
>    first step that could be merged.

We already have `git stash --keep-index`. Is this what you mean here?

>  - implement pushing/poping conflicted state
> 
>    This would obviously be the end goal.

On second thought, this might actually be super trivial. Right now, we support two modes (not counting the `--untracked` stuff): --keep-index and --no-keep-index. In both cases, we seem to create a merge commit whose tree reflects the working directory and whose first parent is HEAD and whose second parent is a single commit on top of HEAD (which contains either no changes in the case of --no-keep-index, or whose tree reflects the index in case of --keep-index).

To extend that to the conflict case, we could introduce a new flag --with-conflicts, and have the commit structure

	Worktree
	 |    \
	 |    index stage 0
	 |      /     |     \
	 | stage 1  stage 2  stage 3
	 |    /     /       /
	HEAD ---------------

The only tricky thing I can see is to maintain backwards compatibility if possible, so that old `git stash` will do something at least semi-sensible with those commit structures.

It might be too small a project, after all.

Ciao, Dscho

Show 25 quoted lines

> > Another potential issue is that a new feature might be prone to naming
> > or user interface discussions which could last for a long time or
> > could not result in clear decisions.
> 
> Yes, this is definitely a potential pitfall.  I haven't thought in
> depth about the interface yet, but I think the discussion around that
> would be something we as mentors could and should guide the student
> through.  We also wouldn't make the feature the default from the
> beginning, but introduce it behind a new flag/maybe a config option,
> to make sure we don't introduce any backwards compatible changes.
> 
> It's probably also something the student should include in their
> proposal, so we can get eyes on it early in the process.
> 
> > So I think we should be very careful if we propose a project that
> > implements a new feature to a student. We should at least consider the
> > above potential issues and see if they can be mitigated before the
> > project starts.
> 
> Thanks for bringing these issues up, they are definitely useful to
> work through.  
> 
> > Thank you anyway for proposing this idea,
> > Christian.
>