Re: GUB error with unpure-pure-container.cc

[David Kastrup]

Graham Percival <address@hidden> writes:

> On Sun, Mar 10, 2013 at 06:22:57PM -0000, Phil Holmes wrote:
>> /home/gub/gub/target/darwin-ppc/src/lilypond-git.sv.gnu.org--lilypond.git-release-unstable/lily/unpure-pure-container.cc:140:
>> error:   initializing argument 2 of 'void
>> scm_set_smob_apply(scm_t_bits, scm_unused_struct* (*)(...), unsigned
>> int, unsigned int, unsigned int)'
>> 
>> and GUB fails.  I don't understand why it should fail on GUB and not
>> on a normal make, but there you go.
>
> GUB uses gcc 4.2.1 or something like that.  The whole development
> toolchain is a few years behind current linux distributions.  So
> if new code relies on current compiler behaviour (whether that's
> using C++11 or just plain bugfixes from previous versions), it can
> easily work with "make" but not in GUB.  Also, the version of
> guile in GUB might not be the latest -- again, if GUB has guile
> 1.8.1 (I'm making this up) but that new code relies on a bugfix in
> guile 1.8.2, then it would also fail.  That said, guile hasn't
> been updated for a while, so I doubt it's a problem with libguile
> specifically.
>
> I have no clue what's happening with scm_set_smob_apply(), so I
> can't help more than that.

scm_set_smob_apply gets passed a function taking a variable number of
arguments, so it needs to be cast appropriately from the actual function
(which takes a fixed number of arguments).  It would appear that the
cast I used was not acceptable to older gcc versions.  I'll try digging
through the definitions and hopefully coming up with something working
on more versions.  Unfortunately we can't just revert the respective
patch since others depend on it already.

-- 
David Kastrup