Re: [Outreachy][PATCH v5] Port helper/test-ctype.c to unit-tests/t-ctype.c

From: Josh Steadmon <steadmon@google.com>
Date: Jan 17, 2024, 05:37 UTC
Message-ID: <ZadnhRtYfrFeMCbX@google.com>
In-Reply-To: <41cf1944-2456-4115-a934-aff2306a26e5@web.de>

On 2024.01.16 20:27, René Scharfe wrote:

Show 14 quoted lines

> Am 16.01.24 um 16:38 schrieb Junio C Hamano:
> > Phillip Wood <phillip.wood123@gmail.com> writes:
> >
> >> Thanks for adding back the test for EOF, this version looks good to me.
> >
> > Thanks.  Let's merge it to 'next'.
> 
> OK.  I'm still interested in replies to my question in
> https://lore.kernel.org/git/a087f57c-ce72-45c7-8182-f38d0aca9030@web.de/,
> i.e. whether we should have one TEST per class or one per class and
> character -- or in a broader sense: What's the ideal scope of a TEST?
> But I can ask it again in the form of a follow-up patch.
> 
> René

I think that the scope of a TEST() should tend small: we want the minimal amount of setup required to test the invariants that we're interested in. For this particular unit test, since we're just testing simple predicates on static sets of characters, I would be OK seeing one TEST() per class/character. That would certainly make this unit test an outlier in the number of checks, but I'm less worried about that since this is testing system-provided functions that we don't expect to change regularly.

Additionally, the elimination of a level of macro indirection makes this more readable IMO.