Re: [RFC] bisect: Introduce skip-when to automatically skip commits

From: Olliver Schinagl <oliver@schinagl.nl>
Date: Apr 10, 2024, 19:39 UTC
Message-ID: <0c9ad7f2-55e8-4763-8201-f29f15a39616@schinagl.nl>
In-Reply-To: <xmqqle5lrp46.fsf@gitster.g>

On 10-04-2024 21:31, Junio C Hamano wrote:

Show 9 quoted lines

> Olliver Schinagl <oliver@schinagl.nl> writes:
> 
>> See, the hook is named 'post-checkout' and thus, it runs after
>> checkout has been performed. So we are now on the 'broken' commit we
>> do not want to test, git should have skipped this already, and not
>> checked it out.
> 
> You are not the only user of this feature (by the way, do not call
> this a "hook".  It should be per "git bisect" session)

Yep, it is stored as part of the bisect session when invoked via as a CLI.

> and others may need to actually inspect their working tree state before being
> able to say "nah, I do not want to test this version, please give me
> another one" by exiting with 125.

So your use of 'You' and 'others' is a bit confusing, if it's me, the person, and 'others' the script itself. But yes, I fully agree with what you are saying I think. Just 'please give me a nother one' could be done in the script itself, or via exit 125 of course (where I totally get that doing the 125 routine is much better).

Show 5 quoted lines

> That is why post-checkout is more
> useful in general.  Contrasted with that, a check that happens
> before the checkout is useful only in a much narrower "I can tell by
> looking only at the commit object name" use case, which I would not
> be interested in seeing.

So that leaves me in the same lack of understanding :p

If we run something post checkout, `bisect_checkout()` has run, and git bisect (the application) is (almost) done. The directory has been 'checked out' (as `bisect_checkout()` of course does `git checkout` (unless `no_checkout` is used) The tree is of course in a bisect session.

So if a script runs, and does the 'please give me a nother commit' thing, how would that work within `builtin/bisect.c` (or in `bisect.c`). This is the part that would puzzle me. The command returns with 125, then what? How can git 'go on and give you a nother one'. afaik the only return codes from `bisect_next()` etc are BISECT_OK and BISECT_FAIL?

Olliver

> 
> Thanks.
> 
>