Re: Claude AI code audit of GNUstep core stack — 150 fixes, 12 perf optimizations, all available for upstream

On 13/04/2026 04:35, Todd White wrote:
> Hi GNUstep Team,
>
> As an exercise to test out the latest Claude AI capabilities, we=20
> recently completed a comprehensive, bottom-up code audit of the=20
> GNUstep core stack =E2=80=94 all seven repositories =E2=80=94 covering l=
ibobjc2,=20
> libs-base, libs-corebase, libs-opal, libs-quartzcore, libs-gui, and=20
> libs-back.

Thanks for that, it looks like very worthwhile work.=C2=A0 I can't really=
=20
comment on other packages, but I'd certainly like to incorporate a lot=20
of that into GNUstep-base.=C2=A0 In the areas of the base library I can se=
e=20
quite a mixture of stuff, some of which is already known, other parts=20
are quite new.=C2=A0 For instance, at one extreme having server verificati=
on=20
off by default was a decision from many years back, but one where a=20
change was long overdue, while at the other extreme the check for=20
cross-thread autorelease pool draining deals with such a perverse=20
situation (I find it hard to concieve of any way to accidentally do=20
that) that it's quite delightful to have a check/fix for it (though=20
classifying it as a critical issue seems silly).

In an ideal world, I'd want to see an individual patch for each issue.=C2=
=A0=20
The patch would contain the source code change, the corresponding=20
testcases, and a ChangeLog entry, but the cloned repos contain only the=20
first of those.

Can the LLM take the testcases it has generated and rewrite them as=20
portable (excluding objc2 and/or platform specific features or=20
controlling them with preprocessor macros) code as used by the gnustep=20
regression testing framework, and add them to the regression tests in=20
each package?

Thanks