Re: guile and slib, old guile versions, missing warnings with compilation

On Mon, 17 Aug 2026, Greg Troxel <[email protected]> wrote:
[...]
>
> The warnings (with guile 3.0.11):
>
[...]
>   ;;; /usr/pkg/share/slib/require.scm:180:37: warning: possibly unbound variable `syncase:load'
>   ;;; /usr/pkg/share/slib/require.scm:181:42: warning: possibly unbound variable `macwork:load'
>   ;;; /usr/pkg/share/slib/require.scm:188:53: warning: possibly unbound variable `load-ciexyz'
>   ;;; /usr/pkg/share/slib/require.scm:191:26: warning: possibly unbound variable `load-color-dictionary'
>   ;;; /usr/pkg/share/slib/require.scm:261:10: warning: possibly unbound variable `transcript-on'
>   ;;; /usr/pkg/share/slib/require.scm:263:10: warning: possibly unbound variable `transcript-off'
>   ;;; compiled /home/gdt/.cache/guile/ccache/3.0-LE-8-4.7/usr/pkg/share/slib/require.scm.go
>   ;;; compiling /usr/pkg/share/slib/mklibcat.scm
>   ;;; compiled
>   /home/gdt/.cache/guile/ccache/3.0-LE-8-4.7/usr/pkg/share/slib/mklibcat.scm.go

I had the same warning as you.  I think they are not harmful, but maybe
SLIB need to re-write portions to make the compiler happier.

[...]

> The larger issue is that slib has guile.scm and guile-2.scm, with
> guile.scm switching on "guile-2" to use guile-2.scm.    It's unclear to
> me if "guile-2" vs "not guile-2" is about the <=1.8 and >=2.0 sets, or
> if there really should be a load file for every major version.
>
> That makes me wonder how far back people think we should be going.  In
> pkgsrc, we have packages for 18, 20, 22, 30.  Things that need 18 at
> this point I see as woefully unmaintained, and in pkgsrc at least
> lilypond is still using 20.  I would therefore guess that for slib and
> things like that, dropping support for <= 1.8 is fine, and it should
> continue to support >= 2.0 at least as long as that isn't painful.

Guile 1 is long gone.  At least, I won't invest my time in fixing thing
for it.  Guile 2 is stable and I don't think release have been made for
it for year.  Though, it is still possible to backport patches to it, if
needed.

> A further related issue: when building the package again, the warnings
> are omitted, I think because the compiled version of the file is used.
> This feels like a bug; ccache when translating .c to .o not only
> produces the output but prints out the stored warnings.  I think guile's
> compiler should do that too.

I understand the possible confusion here.  But the warning is emitted by
the compiler.  We don't keep the warning in the final ELF image.  Do you
get GCC warnings when executing your final a.out ?  That is the same
thing here.

Thanks,
Olivier
-- 
Olivier Dion