[Boost-users] [atomic count] boost on apple gcc 4.2

From: David Abrahams (dave_at_[hidden])
Date: 2008-08-05 23:26:45

Well, first, I think you prolly ought to be posting to the developers'
list, and second I think you need a tag in the subject like "atomic
count" so that the right people will look at it :-)

on Thu Jul 31 2008, "Mat Marcus" <mat-lists-AT-emarcus.org> wrote:

> On Mon, Apr 28, 2008 at 3:22 PM, Peter Dimov <pdimov_at_[hidden]> wrote:
>> Daniel Oberhoff:
>>> Hi,
>>>
>>> Has anyone successfully used boost with the "new" apple gcc 4.2? I
>>> would like to since gcc42 has openmp, but I cant seem to be able to use
>>> boost. I compiled and installed boost with apple gcc 4.2 by directly
>>> using bjam with --toolset=darwin-4.2. When I try to use it I get an
>>> error (detailed output of gcc below) which hints towards problems with
>>> inline assembly. Actually, to get it compiled at all I had to set
>>> -fasm-blocks. Am I missing a flag of sorts? Or is it "just" another woe
>>> with apple's modified assembler?
>>
>> Is the new Apple GCC documented somewhere? I wasn't able to find anything
>> related to a new assembler syntax or -fasm-blocks. Can you ask Apple to
>> explain the issue? Stock g++ 4.2.x compiles the code, so the problem must be
>> Apple-specific.
>
> Ok, I've got a little bit more information on the Apple gcc 4.2
> situation.  Here is a snippet from an email dialog between a user and
> an apple compiler engineer:
>
> user appears below with with prefix ">>"
> apple compiler engineer appears below with with prefix ">"
>
>>> Thank you very much for the detailed reply.
>>> Based on your information I deduce:
>>>
>>> 1) Using gcc 4.2 with 4.0 (c++) libraries is a supported
>>> configuration by Apple
>>
>> Right.
>>
>>> I then conclude that the C++ ABI is fully compatible between 4.0 and
>>> 4.2
>>
>> Right.
>
> [snip, then user hypothesizes as to why there are problems with boost
> under gcc 4.2/libstdc++ 4.0 on current Apple systems]
>
>>> This is the case because boost assumes that a 4.2 compiler implies
>>> 4.2 libraries in flies such as boost/detail/atomic_count_gcc.hpp:
>>>
>>> #if (__GNUC__ > 4) || ((__GNUC__ == 4) && (__GNUC_MINOR__ >= 2))
>>> # include <ext/atomicity.h>
>>> #else
>>> # include <bits/atomicity.h>
>>> #endif
>>
>> Yes, this is a bug or faulty assumption in boost.
>
> [snip]
>
>> The 4.2 library issue is a bug in Boost, not a problem with
>> our compiler.  I'd suggest raising the issue on one of the boost
>> mailing lists and see what they say.
>
> Comments?
> Mat
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com