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

[phillip.wood123@gmail.com]

On 06/04/2024 20:17, Olliver Schinagl wrote:
> Hey Phillip,
> 
> On 06-04-2024 15:50, Phillip Wood wrote:
>> Hi Olliver
>>
>> On 06/04/2024 11:06, Olliver Schinagl wrote:
>>> On 06-04-2024 03:08, Junio C Hamano wrote:
>>>> Olliver Schinagl <oliver@schinagl.nl> writes:
>> If you search builtin/bisect.c you'll see some existing callers of 
>> strbuf_read_file() that read other files like BISECT_START. Those 
>> callers should give you an idea of how to use it.
> 
> Yeah, I found after Junio's hint :) What threw me off, as I wrote 
> earlier, get_terms(). I wonder now, why is get_terms() implemented as it 
> is, and should it not use the same functions? Or is it because terms is 
> a multi-line file, whereas the others are all single line (I didn't 
> look, though I see addline functions for the strbuf functions. Should 
> this be refactored?

get_terms() wants to read the first line into `term_bad` and the second 
line into `term_good` so it makes sense that it uses two calls to 
`strbuf_getline()` to do that. It does not want to read the whole file 
into a single buffer as we do here.

> So with the name, I started to think some more about it, and after 
> playing with some names, I settled on 'bisect-post-checkout'. Things 
> then sort of fell more into place. It is still a hook/commandline 
> option, but it's a much smaller change (since we don't have any special 
> code to check the exit code anymore) as we can (obviously) run `git 
> bisect skip` instead of `exit 125` as well of course.

Does that mean you will be starting "git bisect skip" from the script 
run by the current "git bisect" process. I don't think calling git 
recursively like that is a good idea as you'll potentially end up with a 
bunch of "git bisect" processes all waiting for their post checkout 
script to finish running.

Best Wishes

Phillip