Re: gcc and C++0x on Mac OS X

[John Swensen]

On Oct 26, 2010, at 10:47 AM, John Swensen wrote:

> 
> On Oct 26, 2010, at 10:38 AM, Jordi Gutiérrez Hermoso wrote:
>> 
>> I'm a little worried about this. Apple is quite clearly trying to move
>> away from gcc. They stopped using gcc when it became GPLv3 and are
>> pouring all of their efforts in clang instead. I would like to be able
>> to start using C++0x constructs in Octave, but this is going to be
>> impossible with gcc on Mac OS X if you can't update it.
>> 
>> Is this an actual concern, or am I worried over nothing? How backwards
>> compatible should Octave's code remain? C++0x will mostly likely be
>> finalised by 2012, by the estimates I've seen, and gcc already
>> implements much of it. By contrast, clang is way behind. Many C++0x
>> features could clean up Octave code. Will we be able to use them?
>> 
>> - Jordi G. H.
> 
> My recent compilations have all been using the GCC 4.4.5 as provide by 
> MacPorts.  The only dependency I have to compile on my own is FLTK, as the 
> NO_X11 variant for MacPorts doesn't provide a OpenGL capable compilation of 
> FLTK.  MacPorts also provides GCC 4.5.1 and GCC 4.6 (I'm not sure when all 
> the C++0x feature became available), but I haven't compiled Octave with 
> either one of them.
> 
> On the other hand, getting Octave to compile with clang and llvm *may have* 
> some advantages.  I am not a compiler expert by any stretch of anyone's 
> imagination, but earlier discussion on the mailing list indicated that LLVM 
> may provide the shortest path to JIT compilation via a transformation of the 
> Octave Parser tree into an LLVM compatible format.  Has anyone successfully 
> compiled even the most recent sources with clang/clang++/llvm?  Does anyone 
> know how the GCC frontend for LLVM works?  Would it still provide the JIT 
> possibilities without using clang/clang++?
> 
> John Swensen
> 

Also, to amend my previous email, as of GCC 4.5 there is a plugin system and 
what used to be called the llvm-gcc is now a plugin called DragonEgg.  So I 
guess this kindof also answered my earlier question about whether some of the 
niceties of an LLVM backend would be available for JIT with GCC as a frontend.

John