Re: [PATCH 0/3] revision: refactor ref_excludes to ref_visibility

From: John Cai <johncai86@gmail.com>
Date: Jun 23, 2023, 19:16 UTC
Message-ID: <941CCF5B-1FE6-46BE-9ED7-77C11E943E2E@gmail.com>
In-Reply-To: <ZJRDZ7NhyNpTV8jD@nand.local>

Hi Taylor,

On 22 Jun 2023, at 8:49, Taylor Blau wrote:

Show 26 quoted lines

> On Thu, Jun 22, 2023 at 08:42:24AM -0400, Taylor Blau wrote:
>> On Wed, Jun 21, 2023 at 07:35:09PM +0000, John Cai via GitGitGadget wrote:
>>> The ref_excludes API is used to tell which refs should be excluded. However,
>>> there are times when we would want to add refs to explicitly include as
>>> well. 4fe42f326e (pack-refs: teach pack-refs --include option, 2023-05-12)
>>> taught pack-refs how to include certain refs, but did it in a more manual
>>> way by keeping the ref patterns in a separate string list. Instead, we can
>>> easily extend the ref_excludes API to include refs as well, since this use
>>> case fits into the API nicely.
>>
>> After reading this description, I am not sure why you can't "include" a
>> reference that would otherwise be excluded by passing the rules:
>>
>>   - refs/heads/exclude/*
>>   - !refs/heads/exclude/but/include/me
>>
>> (where the '!' prefix in the last rule is what brings back the included
>> reference).
>>
>> But let's read on and see if there is something that I'm missing.
>
> Having read this series in detail, I am puzzled. I don't think that
> there is any limitation of the existing reference hiding rules that
> wouldn't permit what you're trying to do by adding the list of
> references you want to include at the end of the exclude list, so long
> as they are each prefixed with the magic "!" sentinel.

To be honest, I had no idea "!" would have this effect--so thanks for bringing it to my attention.

Show 21 quoted lines

>
> I think splitting the list of excluded references into individual
> excluded and non-excluded references creates some awkwardness. For one:
> excluded references already can cause us to include a reference, so
> splitting that behavior across two lists seems difficult to reason
> about.
>
> For example, if your excluded list contains:
>
>   - refs/heads/foo
>   - refs/heads/bar
>   - !refs/heads/foo/baz
>
> and your included lists contains:
>
>   - refs/heads/bar/baz/quux
>
> I am left wondering: why doesn't the rule pertaining to
> refs/heads/foo/baz show up in the included list? Likewise, what happens
> with refs/heads/bar/baz/quux? It is a child of an excluded rule, so the
> question is which list takes priority.

Now knowing the effect of "!", I understand the concerns about having a separate list for included references. However, I do think it would be odd if there is a ref_included() function to also use ref_excluded() to include a ref.

Also, even with the current API, I think it can be confusing to reason about what takes precedence if there is a mix of inclusions and exclusions. For example, if the excluded list contains:

- refs/heads/foo/baz
- !refs/heads/foo

would the inclusion take precedence, or the exclusion?

Show 5 quoted lines

>
> Mostly, I am wondering if I am missing something that would explain why
> you couldn't modify the above example's excluded list to contain
> something like "!refs/heads/bar/baz/quux", eliminating the need for the
> include list entirely.

I think my one reservation however, is with usability of the current API. It's not very intuitive to include references by adding them with ref_excluded(&exclusions, "!ref/to/be/included").

I do think it would be easier to reason about if we kept two separate lists, one for inclusion and one for exclusion. The existence of the "!" magic sentinel does make things much more confusing however. I almost want to remove support for the magic sentinel in favor of keeping two distinct lists that cannot be mixed. Wondering your thoughts on that approach?

>
> Thanks,
> Taylor

thanks! JOhn