Re: Discussion for interactive --patch commands to get --unified support

From: Phillip Wood <phillip.wood123@gmail.com>
Date: May 2, 2025, 14:33 UTC
Message-ID: <3e38b7b4-eb6e-4af5-a5f6-f0d7ca4d32d6@gmail.com>
In-Reply-To: <20250429220933.GC36727@coredump.intra.peff.net>

Hi Leon and Jeff

On 29/04/2025 23:09, Jeff King wrote:

Show 14 quoted lines

> On Tue, Apr 29, 2025 at 10:16:15AM +0100, Leon Michalak wrote:
> 
>> - make `diff.context` setting extend to the interactive patch commands
>> (not sure how a change like this would be welcomed considering it
>> could change users command outputs seemingly out of nowhere)
>> - add an `interactive.context` setting that would work like the
>> existing `diff.context` setting but apply only to the interactive
>> patch commands
> 
> In my opinion it would be fine to respect diff.context (and probably
> diff.interhunkcontext[1]) by default. Though it does change the command
> output, the interactive output is by definition user-facing, so we
> shouldn't be breaking scripts. And we already respect other porcelain
> level config like colorizing.

I think that would be a good place to start and would be a useful improvement on the status quo. To implement it one can copy what we do to respect diff.algorithm. I'm not sure it is worth the effort to support `git add -p -U <context>`. Being able to interactively change the context as suggested elsewhere sounds more interesting to me but it would be more work to implement as we'd have to regenerate the diff each time.

Best Wishes

Phillip

Show 15 quoted lines

> I think the only reason we don't already do so is that the interactive
> code is built around the plumbing commands, which conservatively avoid
> various config options. So the calling code has to explicitly check the
> config itself.
> 
> -Peff
> 
> [1] Looking at git_diff_ui_config(), which are all the options read by
>      porcelain git-diff but not by plumbing git-diff-files, etc, there
>      may be other config in the same boat. E.g., I'd guess that people
>      with diff.colormoved set would appreciate seeing that effect in the
>      colorized versions we show. But I think it is OK to just consider
>      diff.context for now, and see if anybody ever cares enough about
>      other options to look into them.
>