[MMTK] Re: building MMTK on Mac OS X

imk@mrc-lmb.cam.ac.uk imk@mrc-lmb.cam.ac.uk
Wed, 30 May 2001 13:46:42 +0100


--On Tuesday, May 22, 2001 13:45 -0400 "Randy M. Wadkins" <rwadkin@jhmi.edu> 
wrote:

> Hi:
> I didn't get as far as you two did.  I get the following error
> message when trying to build MMTK.  Any thoughts how to fix this?
> Thanks,
> --Randy
>
> [...]
>In file included from Include/MMTK/forcefield.h:18,
>                   from Src/MMTK_deformation.c:8:
> /usr/include/pthread.h:241: parse error before `destructor'
> /usr/include/pthread.h:241: parse error before `)'

I finally had time to get back to this, and succeeded to compile. The error 
mentioned above does not have anything to do with MMTK, I remember having 
problems with this before, with other sofware using pthreads. Pthreads 
support is rather incomplete as yet in OS X. To get MMTK to compile, install 
Numeric 20.0, then replace arrayobject.h as Konrad recently described:

> - replace the file arrayobject.h from Numeric by the one available at
>   ftp://dirac.cnrs-orleans.fr/pub/MMTK/arrayobject.h

Then make a backup copy of /usr/include/pthreads.h and apply the following 
patch

--- /usr/include/pthread.h.orig Thu Apr 26 19:52:24 2001
+++ /usr/include/pthread.h      Tue May 29 20:09:43 2001
@@ -239,7 +239,7 @@
                                int policy,
                                 const struct sched_param *param));
 int       pthread_key_create __P((pthread_key_t *key,
-                            void (*destructor)(void *)));
+                            void (*)(void *)));
 int       pthread_key_delete __P((pthread_key_t key));
 int       pthread_setspecific __P((pthread_key_t key,
                              const void *value));

MMTK-2.2b6 seems to compile and work just fine after that, but I'm not 
completely sure of possible side-effects of the above modification, so if 
you have problems compiling other pthreads-using software, go back to the 
older version. I have ventured this after comparing with the IRIX header 
file.

Oh, and if you still get any trouble with "_PyArray_API not defined" you can 
use python with the -v option to find out the offending module. I had to 
recompile Scientific-2.2 after replacing arrayobject.h, but that may just be 
because I had been tinkering with that to get it to work already.

Amike salutas c^iujn,

  Ivo