MacPorts and clang

[Jack Howarth]

On Fri, Mar 04, 2011 at 05:48:56AM +1100, Joshua Root wrote:
> On 2011-3-4 02:40 , Jack Howarth wrote:
> >   Are there any plans to attempt to move MacPorts towards
> > building with clang once Xcode 4 is released? Considering that
> > we have so many packages supporting gcc4x variants to build with
> > FSF gcc, it seems strange not to do the same for clang.
> >              Jack
> > ps I assume that Xcode 4 won't be held up until Lion is released so
> > these clang variants could be first done on Snow Leopard. It is unclear
> > from http://developer.apple.com/technologies/tools/whats-new.html
> > if the clang c++ support in Xcode 4.0 will use the existing libstdc++
> > or the new libc++ instead. It would be nice if it were the latter.
> 
> We already have configure.compiler settings for llvm-gcc-4.2 and clang.
> Port maintainers can use these as desired. We default to whatever Apple
> calls /usr/bin/cc on each OS.
> 
> There's a default for darwin 11 in base already on spec (llvm-gcc-4.2).

 Wow. That would surprise me as I was under the impression that in llvm
development, llvm-gcc-4.2 was being given short shrift compared to clang.
Certainly my own runs of llvm-2.8 and llvm svn's gfortran with the
Polyhedron 2005 benchmarks has shown continued runtime performance regressions...

http://lists.cs.uiuc.edu/pipermail/llvmdev/2010-September/034780.html
http://lists.cs.uiuc.edu/pipermail/llvmdev/2011-February/038224.html

This compares to clang which, as measured by the himenoBMTxpa benchmark
anyway, appears to be improving towards FSF gcc's current performance.
            Jack

> We can change that if it turns out to be wrong when Lion ships.
> Otherwise this should all "just work".
> 
> - Josh