Re: mingw, ptw32, and struct timespec

[Eric Blake]

On 06/02/2010 10:50 AM, Eric Blake wrote:
> I'm getting a compilation failure when trying to cross-compile from
> Linux to mingw, because I have the mingw pthreads-win32 (ptw32) library
> installed (Fedora 13 system, with mingw32-pthreads and mingw32-gcc
> packages).  The ptw32 library installs a pthread.h that blindly defines
> struct timespec if HAVE_STRUCT_TIMESPEC is not defined:
> 
> #ifndef HAVE_STRUCT_TIMESPEC
> #define HAVE_STRUCT_TIMESPEC 1
> struct timespec {
>         long tv_sec;
>         long tv_nsec;
> };
> #endif /* HAVE_STRUCT_TIMESPEC */
> 
> That library also has the nasty behavior of installing a config.h file,
> which it then includes as part of pthread.h (recommended practice is
> that config.h must NEVER be installed, and that if you are installing a
> public header that depends on configuration options, you should install
> a header by a different name that only your package will refer to, and
> that you should make an extreme effort to make the file be independent
> of compiler options that might differ between when your library was
> configured vs. when a client of your library is configured).

It gets worse.  The ptw32 library pthread.h is broken, with unsafe
constructs like:

#define localtime_r( _clock, _result ) \
        ( *(_result) = *localtime( (_clock) ), \
          (_result) )

that evaluate _result multiple times within the function call.  The
complete list of broken *_r functions incorrectly defined in <pthread.h>:

strtok_r
asctime_r
ctime_r
gmtime_r
localtime_r
rand_r

Maybe the better thing for gnulib to do would be to flat-out reject the
ptw32 pthread.h as useless, until pthreads-win32 cleans up the mess?

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature