future gcc releases on OS X -- they may be going away

[Brian Barnes]

Fellow OS X coders,

First, I apologize for this off-topic post, but I hope it is of  
interest to you all.  I use Aquamacs nearly every day to code in  
Fortran and Python (my Fortran codes are compiled using gcc/gfortran),  
and I just wanted to give you all a heads up to a recent development  
in gcc land.

The GNU steering committee has decided to remove OS X (intel) from  
their primary target list, and OS X (power pc) from their secondary  
target list.  This means that future releases of gcc may not work on  
OS X.  I presume many of you compile codes using gcc every day.  This  
would kill the availability of a free compiler for languages not  
supported by llvm/clang on OS X (which is a surprisingly lengthy list).

http://lists.macosforge.org/pipermail/macports-dev/2009-September/010209.html
http://gcc.gnu.org/ml/gcc-patches/2009-09/msg01523.html

I don't know what to do about this, and I certainly can't help  
maintain gcc.  But maybe someone reading this list will have an idea,  
or know someone who can help. Finally, Aquamacs is awesome -- thank  
you for this tool!

Sincerely,

Brian Barnes
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[davidswelt]

I don't think Aquamacs or Emacs compile with the original GCC 4.x, but  
only with Apple's GCC.

I also suspect that OS X users are the the largest or at least the  
second largest GCC user group (by platform).   OS X itself is compiled  
on gcc...
If Apple contributed more changes back to the gcc project, they  
wouldn't have to declare Apple's platform a "secondary" one.

--
http://aquamacs.org -- Aquamacs: Emacs on Mac OS X
http://aquamacs.org/donate -- Could we help you? Return the favor and  
support the Aquamacs Project!

On Sep 22, 2009, at 6:35 PM, Brian Barnes wrote:

> Fellow OS X coders,
>
> First, I apologize for this off-topic post, but I hope it is of  
> interest to you all.  I use Aquamacs nearly every day to code in  
> Fortran and Python (my Fortran codes are compiled using gcc/
> gfortran), and I just wanted to give you all a heads up to a recent  
> development in gcc land.
>
> The GNU steering committee has decided to remove OS X (intel) from  
> their primary target list, and OS X (power pc) from their secondary  
> target list.  This means that future releases of gcc may not work on  
> OS X.  I presume many of you compile codes using gcc every day.  
> This would kill the availability of a free compiler for languages  
> not supported by llvm/clang on OS X (which is a surprisingly lengthy  
> list).
>
> http://lists.macosforge.org/pipermail/macports-dev/2009-September/010209.html
> http://gcc.gnu.org/ml/gcc-patches/2009-09/msg01523.html
>
> I don't know what to do about this, and I certainly can't help  
> maintain gcc.  But maybe someone reading this list will have an  
> idea, or know someone who can help. Finally, Aquamacs is awesome --  
> thank you for this tool!
>
> Sincerely,
>
> Brian Barnes

_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

Brian Barnes wrote, On 23/9/09 10:35 AM:
> Fellow OS X coders,
>
> First, I apologize for this off-topic post, but I hope it is of  
> interest to you all. I use Aquamacs nearly every day to code in  
> Fortran and Python (my Fortran codes are compiled using gcc/gfortran),  

I guess it isn't off topic if it means that we soon will have to make
sure that emacs and Aquamancs can be built using llvm/clang

I do notice this ray of hope for Fortran on the Mac in the LLVM 2.5
release notes:

http://llvm.org/docs/ReleaseNotes.html

   LLVM 2.5 includes several major new capabilities:
    [ ... ]
    * llvm-gcc now generally supports the GFortran front-end, and the
precompiled release binaries now support Fortran, even on Mac OS/X.
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

Brian Barnes wrote, On 23/9/09 10:35 AM:
> Fellow OS X coders,
>
> First, I apologize for this off-topic post, but I hope it is of  
> interest to you all. I use Aquamacs nearly every day to code in  
> Fortran and Python (my Fortran codes are compiled using gcc/gfortran),  

I guess it isn't off topic if it means that we soon will have to make
sure that emacs and Aquamacs can be built using llvm/clang

I do notice this ray of hope for Fortran on the Mac in the LLVM 2.5
release notes:

http://llvm.org/docs/ReleaseNotes.html

   LLVM 2.5 includes several major new capabilities:
    [ ... ]
    * llvm-gcc now generally supports the GFortran front-end, and the
precompiled release binaries now support Fortran, even on Mac OS/X.
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[davidswelt]

On Sep 25, 2009, at 7:26 PM, Sidney Markowitz wrote:
> I guess it isn't off topic if it means that we soon will have to  
> make sure that emacs and Aquamancs can be built using llvm/clang

Yes, it's be nice to see it compiled with clang.  It might even be a  
bit faster.  Give it a shot if you have time.

Emacs 22 and also the 23/Appkit port compile with clang thanks to a  
few small changes already... See Yamamoto Mitsuharu's e-mail on emacs-
devel (Sep 8 2009).

That said... some gcc people saying that Mac/Intel64 is becoming a  
secondary platform in terms of their support doesn't mean that gcc 4.2  
as deployed with every DevTools install world-wide is magically going  
to stop working...  Apple has been maintaining their fork of gcc  
(which is reasonably far away from the mainstream, actually) and  
they're going to keep using it, or, as you suggest, switch to clang.  
iPhone apps are already compiled with clang.

What was and is causing big pain has been the switch to 10.6, though.  
Different story.

- D


--
http://aquamacs.org -- Aquamacs: Emacs on Mac OS X
http://aquamacs.org/donate -- Could we help you? Return the favor and  
support the Aquamacs Project!

_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

David Reitter wrote, On 26/9/09 11:37 AM:
> That said... some gcc people saying that Mac/Intel64 is becoming a  
> secondary platform in terms of their support doesn't mean that gcc 4.2  
> as deployed with every DevTools install world-wide is magically going  
> to stop working...

That's true ... We have a working gcc on our machines, so it is not a
big deal if a new version of gcc that we never get doesn't support those
machines. It is more of a concern if 10.7 arrives without a gcc that
supports it and there are programs that cannot be built using whatever
compiler does come with it. And that's more of a concern to people who
use Fortran. I'm sure that 10.7 will not arrive without any support for C++.

I haven't been following the issues about LLVM and clang, or even yet
understand the difference. This post from an Apple dev seems to indicate
well what is going on with it:

  http://lists.apple.com/archives/xcode-users/2009/Sep/msg00272.html

It looks like the right thing to do is confirm that Aquamacs builds with
llvm-clang and keep checking that it does so we are ready if gcc for
Darwin does go away some day.

_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

David Reitter wrote, On 26/9/09 11:37 AM:
> Yes, it's be nice to see it compiled with clang.  It might even be a  
> bit faster.  Give it a shot if you have time.

A quick test did not work right out of the box. I downloaded the latest
released clang-llvm, 0.221, put a symbolic link to the bin/clang that
came with that in /usr/local/bin/clang, verified that I could compile
and run a hello world app with

  /usr/local/bin/clang -o hello hello.c

the edited build23.sh to insert CC=/usr/local/bin/clang in the configure
line.

Running the build.sh script got through compilation with nothing worse
than more warnings than I would like, then broke at

`/bin/pwd`/temacs --batch --load loadup bootstrap
make[1]: *** [bootstrap-emacs] Segmentation fault


I probably won't have time to really look at it to figure out what isn't
working.

  -- sidney


_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

Sidney Markowitz wrote, On 26/9/09 1:25 PM:
> `/bin/pwd`/temacs --batch --load loadup bootstrap
> make[1]: *** [bootstrap-emacs] Segmentation fault

Using the llvm-gcc compiler from the LLVM 2.5 download ended with a
segmentation fault in the same place in the build. That doesn't mean
that it is the same error, only that the temacs executable dies with a
fault the first time in the build that it is run.

  -- sidney
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[davidswelt]

On Sep 25, 2009, at 10:03 PM, Sidney Markowitz wrote:

> Sidney Markowitz wrote, On 26/9/09 1:25 PM:
>> `/bin/pwd`/temacs --batch --load loadup bootstrap
>> make[1]: *** [bootstrap-emacs] Segmentation fault
>
> Using the llvm-gcc compiler from the LLVM 2.5 download ended with a  
> segmentation fault in the same place in the build. That doesn't mean  
> that it is the same error, only that the temacs executable dies with  
> a fault the first time in the build that it is run.

try the patch below.
- D



Begin forwarded message:

> From: YAMAMOTO Mitsuharu <mituharu@...>
> Date: September 8, 2009 9:04:13 PM EDT
> To: CHENG Gao <chenggao@...>
> Cc: emacs-devel@...
> Subject: Re: Emacs 22 Carbon+AppKit port and Emacs 23 Mac port
>
>>>>>> On Tue, 08 Sep 2009 18:38:03 +0800, CHENG Gao  
>>>>>> <chenggao@...> said:
>
>> Out of curiosity, I tried to build Mac port (1.92) with clang, and
>> found it works well, while building 32bit ns port fails.
>
> The Mac port and the Carbon+AppKit port have contained a small kludge
> for LLVM-based compilers (llvm-gcc for 10.5, llvm-gcc and clang for
> 10.6) bundled with Xcode since their first versions.
>
> RCS file: /cvsroot/emacs/emacs/src/m/intel386.h,v
> retrieving revision 1.42.2.4
> diff -c -p -r1.42.2.4 intel386.h
> *** src/m/intel386.h 27 Feb 2008 23:32:04 -0000 1.42.2.4
> --- src/m/intel386.h 6 Sep 2008 04:51:04 -0000
> *************** NOTE-END */
> *** 215,221 ****
>  #endif
>
>  #if defined (MAC_OSX) || defined (DARWIN)
> ! #ifdef _LP64
>  /* For Intel Mac, with CC='gcc -arch x86_64'.  */
>  #define NO_ARG_ARRAY
>  #endif
> --- 215,221 ----
>  #endif
>
>  #if defined (MAC_OSX) || defined (DARWIN)
> ! #if defined (_LP64) || defined (__llvm__)
>  /* For Intel Mac, with CC='gcc -arch x86_64'.  */
>  #define NO_ARG_ARRAY
>  #endif
>
> I'm not sure if it is applicable to all versions of LLVM-based
> compilers or all target architectures.  So I just put it as a "kludge"
> for unofficial ports.
>
> You can also turn on link-time optimization option (-O4) on intel
> processors though it takes much more time to build.  IIRC, it didn't
> work well for ppc64.
>
>     YAMAMOTO Mitsuharu
> mituharu@...
>
>

_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

David Reitter wrote, On 26/9/09 5:29 PM:
> try the patch below.

That did it! I only tried it with clang, not llvm-gcc, but that's the
more extreme case anyway.

  -- sidney
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel

[Sidney Markowitz]

David Reitter wrote, On 27/9/09 3:56 AM:

> On Sep 26, 2009, at 7:36 AM, Sidney Markowitz wrote:
>
>> David Reitter wrote, On 26/9/09 5:29 PM:
>>> try the patch below.
>> That worked with clang (didn't try llvm-gcc, but presumably that  
>> would work too), but my email to aquamacs-devel saying so has not  
>> shown up after about four hours so far.
>
> That's really great news (that it compiles, that is).
> Is it noticeable faster?  Startup?

All I tried to compare speed was to start up Aquamacs several times
looking at the clock ticking off seconds in the menu bar, and starting
up Slime. To within that precision for those two actions, they ran at
the same speed.

I'm Cc'ing this to the dev list because as far as I can tell my first
reply saying that clang worked with that patch never arrived on the list.

  -- sidney
_______________________________________________
Aquamacs-devel mailing list
Aquamacs-devel@...
http://lists.aquamacs.org/mailman/listinfo.cgi/aquamacs-devel