On Wed, Jul 20, 2011 at 09:26:46AM -0700, Jeremy Huddleston wrote:
>
> 4) compiler codegen issues
>
> As discussed on this list over the past month or so, we've run into some
> codegen issues when using llvm-gcc-4.2. This will appear as a segfault,
> timeout, bad visual artifacts, or just plain weird behavior. If something
> new like this coprs up, try rebuilding the relevant ports with a different
> compiler (preferably gcc-4.2 since that was what created the "known good"
> version). If you find that works, bisect the relevant ports to find which
> port was built wrong. Once you know the actual port that has codegen issues,
> try using clang instead of llvm-gcc-4.2. If that doesn't work, please file a
> bug with Apple and llvm (and reference the bug number in the Portfile), then
> fall back on gcc-4.2 until it is addressed.
>
Jeremy,
We avoided those issues on fink by defaulting 10.7 to use a
path-prefix-clang which maps cc/gcc to clang
and cxx/g++ to clang++. So far we have seen very few instances of bad codegen
(and we are adding InfoTest
whenever possible to validate the clang builds of packages against their
testsuites). The only challenge has
been to correct existing code for the stricter language compliance of
clang/clang++. Also, the fact that
clang/clang++ doesn't handle the *PATH environmentals has caused us to rework
some builds that utilized that
feature. Overall the switch to the clang compilers is considered a net win
(especially with the faster
compile times and lower memory footprint).
Jack
_______________________________________________
macports-dev mailing list
macports-dev@lists.macosforge.org
http://lists.macosforge.org/mailman/listinfo.cgi/macports-dev
