Re: libffi4, ffi_prep_cif on mips and mipsel
[CCing debian-mips and debian-gcc]
The gcc-4.2 source uses the libffi backport from the trunk. To
investigate further, please look at libffi's config.log to see why the
test for ffi_prep_cif fails.
Matthias
Neil Williams writes:
> For once, this isn't cross-build related - not even slightly. :-)
>
> I'm wondering if you can help me with a package problem that shows up
> only on mips and mipsel and is related to the ffi_prep_cif function in
> libffi4, built from gcc-4.2.
>
> g-wrap looks for this function during the build and the latest upload
> (which only changed the dependency information regarding libglib1.2 vs
> libglib2.0) breaks gnucash on mips and mipsel because g-wrap is unable
> to find ffi_prep_cif in libffi4 for mips and mipsel. I've checked
> previous logs and this has not been a problem with previous builds of
> g-wrap.
>
> http://buildd.debian.org/fetch.cgi?pkg=g-wrap;ver=1.9.6-3.1%2Bb1;arch=mips;stamp=1183106948
> checking ffi.h usability... yes
> checking ffi.h presence... yes
> checking for ffi.h... yes
> checking for ffi_prep_cif in -lffi... yes
>
> http://buildd.debian.org/fetch.cgi?pkg=g-wrap;ver=1.9.6-3.2;arch=mips;stamp=1184676067
> checking ffi.h usability... yes
> checking ffi.h presence... yes
> checking for ffi.h... yes
> checking for ffi_prep_cif in -lffi... no
>
> This causes g-wrap to try to use an internal libffi that appears to
> work but causes a FTBFS within gnucash (which is about the only
> application now using g-wrap in Debian). Disabling the internal libffi
> support in g-wrap could cause a FTBFS in g-wrap on mips and mipsel.
>
> This *only* happens on mips and mipsel. All other architectures
> correctly locate ffi_prep_cif within the libffi library. If previous
> builds of g-wrap on mips/mipsel had used the internal ffi support, I
> would work on fixing that but this does appear to be a change in how
> libffi4 has been built on mips and mipsel.
>
> The 1.9.6-3.2 builds were made using gcc-4.2 4.2-20070712-1.
> (Tue Jul 17 12:41:07 2007)
>
> The 1.9.6-3.1 builds were made using gcc-4.2 4.2-20070627-1
> (Fri Jun 29 08:49:08 2007)
>
> I'm a relatively new DD and I'm not sure how to proceed on this bug -
> it was my NMU that exposed it but I can't see how my changes would have
> caused it.
>
> Can you help me? Maybe just a few pointers at where the real bug lies?
> Is it a bug in libffi4 on mips and mipsel from the gcc-4.2 source? I
> don't mind reporting it as such but I would like to check that it
> reasonable to assign this to the libffi build from gcc-4.2 as it would
> be RC due to the FTBFS in gnucash.
Reply to: