Created attachment 442028 [details, diff] tree-vect-data-refs correctness patch for gcc Attempting to build media-libs/libraw on the musl libc tickles a bug in vect_analyze_data_ref_accesses on GCC 4.9 and 5.4 (only ones I tested, apparently 5.3 is also affected). The attached patch, from upstream, resolves this issue. As upstream seems to be unwilling to push out a new GCC 4.9 version, and 5.4 did not include this particular fix, it may be beneficial to carry this patch in the Gentoo patch set.
(In reply to A. Wilcox (awilfox) from comment #0) > Created attachment 442028 [details, diff] [details, diff] > tree-vect-data-refs correctness patch for gcc > > Attempting to build media-libs/libraw on the musl libc tickles a bug in > vect_analyze_data_ref_accesses on GCC 4.9 and 5.4 (only ones I tested, > apparently 5.3 is also affected). > > The attached patch, from upstream, resolves this issue. As upstream seems > to be unwilling to push out a new GCC 4.9 version, and 5.4 did not include > this particular fix, it may be beneficial to carry this patch in the Gentoo > patch set. Looks good to me. @vapier, any reason not to back port this to 4.9 and 5.x?
(In reply to Anthony Basile from comment #1) > > Looks good to me. > > @vapier, any reason not to back port this to 4.9 and 5.x? Actually I should say I'll add this to the gcc ebuild on the musl overlay for now.
(In reply to A. Wilcox (awilfox) from comment #0) > Created attachment 442028 [details, diff] [details, diff] > tree-vect-data-refs correctness patch for gcc > > Attempting to build media-libs/libraw on the musl libc tickles a bug in > vect_analyze_data_ref_accesses on GCC 4.9 and 5.4 (only ones I tested, > apparently 5.3 is also affected). > > The attached patch, from upstream, resolves this issue. As upstream seems > to be unwilling to push out a new GCC 4.9 version, and 5.4 did not include > this particular fix, it may be beneficial to carry this patch in the Gentoo > patch set. Can you get me a reference to the upstream commit for this?
(In reply to Anthony Basile from comment #3) > > Can you get me a reference to the upstream commit for this? Never mind, we found it. I think we're going to go directly from gcc-4 to gcc-6 where this patch is not needed. So I think pushing this onto the musl overlay is sufficient. sys-devel/gcc::musl is going away when we stabilize gcc-6.
(In reply to Anthony Basile from comment #4) > (In reply to Anthony Basile from comment #3) > > > > Can you get me a reference to the upstream commit for this? > > Never mind, we found it. I think we're going to go directly from gcc-4 to > gcc-6 where this patch is not needed. So I think pushing this onto the musl > overlay is sufficient. sys-devel/gcc::musl is going away when we stabilize > gcc-6. With gcc-6 on the way to being stabilized, this is no longer needed.