On Wed, 28 Oct 2026 10:53:26 +0100,
Stuart Henderson <[email protected]> wrote:
> 
> On 2026/10/28 02:41, Kirill A. Korinsky wrote:
> > I'd like to drop ports-gcc from build dependency of libxml.
> 
> > Why? To use it as dependency of gcc which is required for cobol.
> > 
> > To do it, I need to switch libxml to autotools instead of meson.
> 
> Something is strange here because libxml was still using autoconf when
> it was switched to ports-gcc to fix build. (Perhaps libxml required a
> newer C standard at one point and then switched back, this does seem
> unlikely though. Also I would think that if libxml does currently build
> with base-gcc, we can't really expect it to keep doing so, and we don't
> have bulk build machines using base-gcc any more to alert us to this
> happening).
>

May something had happened with gcc-4 in base? or bfd? I don't know, and
without the error it is hard to say.

Anyway, their README states:

  All code must conform to C89 and pass the GitLab CI tests. Add regression
  tests if possible.

https://github.com/GNOME/libxml2/blob/master/README.md

what reads good enough to be on safe side.

> > Here a diff which was tested on my hppa VM.
> > 
> > It builds, and it doesn't need gcc.
> > 
> > Not sure about lib bump, I increased it to be on safe side.
> > 
> > Ok?
> 
> I'd really like to avoid adding more dependencies to the build of gcc
> as used by the majority of the ports tree (i.e. for those ports using it
> for C, C++, Fortran).
> 
> If supporting other languages in GCC adds new dependencies can those
> langs be knocked out via a bootstrap flavour for the usual ports builds?
> (this also helps avoid bottlenecks if they add much additional build time).
> 
> 

So far cobol is the first language (and seems the only one) which brings new
dependency.

I can limit it only on non GCC in base platforms, but it makes a bit
different build for that beast that I don't like neigher.

Adding here one more copy with minimal patchset from gcc makes support it a
bit more tricky.

-- 
wbr, Kirill

Reply via email to