[svn:parrot] r39873 trunk/include/parrot

[François Perrad]

2009/7/4 MoC <moc at coders-haven.net>:
>> Index: /trunk/include/parrot/thr_windows.h
>> ===================================================================
>> --- /trunk/include/parrot/thr_windows.h (revision 37201)
>> +++ /trunk/include/parrot/thr_windows.h (revision 39873)
>> @@ -124,11 +124,11 @@
>>  typedef void (*Cleanup_Handler)(void *);
>>  -#ifndef _STRUCT_TIMESPEC
>> -#  define _STRUCT_TIMESPEC
>> +#ifndef HAVE_STRUCT_TIMESPEC
>> +#  define HAVE_STRUCT_TIMESPEC
>>  struct timespec {
>>     time_t tv_sec;
>>     long tv_nsec;
>>  };
>> -#endif /* _STRUCT_TIMESPEC */
>> +#endif /* HAVE_STRUCT_TIMESPEC */
>>   #  undef CONST
>
> I'd just like to add that I don't really think that Parrot's to blame here,
> but rather pthreads-win32. I first got in touch with the "struct timespec"
> issue when trying to cross-compile ruby-svn, which uses pthreads, with MinGW
> GCC 3.4.5. (The pthreads headers weren't bundled yet, though the library
> files were already included since they seem to have built MinGW itself with
> pthreads support.) Back then I came to a similiar conclusion though wanted
> to check on who's to blame. So I searched my VirtualBox Ubuntu's include
> directory (since I'm not that familiar with the structure of *nix systems)
> and found out that:
> 1) _STRUCT_TIMESPEC appears to be used on *nix OS (I have to admit that only
> checking Ubuntu might not allow me to say that, but since Ruby also used
> this variant I think I should be right.)
> 2) _STRUCT_TIMESPEC is actually defined somewhere within sys/time.h (or one
> of the included headers) and is only defined in pthreads-win32's pthread.h
> because by default it's not defined within MinGW.
> So I think that it's rather pthreads-win32 that's responsible for this issue
> than Parrot and I suggest we should contact pthreads-win32 and/or MinGW
> about it.
> Any other opinions?
>

This issue comes with the newest MinGW which supplies a include pthread.h,
and the segregation between pthread & Windows thread is not perfect.

We have 2 platform dependent includes thr_windows.h & thr_pthread.h
- thr_windows.h includes windows.h (Parrot doesn't use pthread-win32,
because we target all compiler available on Windows, not only Mingw)
- thr_pthread.h includes pthread.h.
Both are included by the generated include platform.h.
parrot.h includes platform.h & pthread.h if exists.

So, I suggest to remove the conditional inclusion of pthread.h in parrot.h
and the definition of struct timespec in thr_windows.h.

See, the attached patch.
This patch needs validation on many platform before commit (not like r39873).

François.

Note:
- the guard _STRUCT_TIMESPEC is used in thread.h, which is a common include.
- the guard _STRUCT_TIMESPEC is not used by MinGW


> --MoC
>
> _______________________________________________
> http://lists.parrot.org/mailman/listinfo/parrot-dev
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pthread-mingw.patch
Type: application/octet-stream
Size: 1514 bytes
Desc: not available
URL: <http://lists.parrot.org/pipermail/parrot-dev/attachments/20090704/aa1e3a79/attachment-0001.obj>