Re: [Acl-devel] [PATCH attr 2/8] build: Do not assume we have a config.h header around

[Guillem Jover]

On Sat, 2022-11-12 at 21:30:27 +0700, Mike Frysinger wrote:
> On 20 Jul 2019 04:14, Guillem Jover wrote:
> > We are not installing it anyway, and to be able to use it we would need
> > a config.h generated during a configure step, which we do not provide.
> > Just remove the conditionals and assume everything required is present.
> 
> i think there's value in leveraging the system settings found in there,
> like the various system extensions.  the normal way to make this logic
> conditional is to check for HAVE_CONFIG_H defined.

I've never been a fan of the HAVE_CONFIG_H pattern, because that means
if the build system is misconfigured (bad -I or similar), you might
end up with files that do not include the config.h that apparently is
being included, which can be rather confusing to debug.

> we could also integrate building of the tool so you could run `make
> examples/copyattr` ...

Sure, I guess the main question is what's the purpose of this code,
and what expectations it needs to abide by.

If this is intended to be a compilable example and not leave the source
tree, then integrating more tightly with the build system would make
more sense yes. Part of the patch already matches acl/examples/copyperm.c,
but I'd indeed revert the removal of <config.h>, and hook it into the
build system to build it, and remove the Makefile.

If this is intended to be a simple «project» example alongside the
documentation that would ideally be installed as part of the development
files, then instead of this patch the build system would install it and
its Makefile, but in my mind not providing a simple configure.ac would be
a bit confusing, but then that also might seem too much scaffolding?

I'm fine providing patches in either direction TBH if desired.

Thanks,
Guillem