homepage

On Solaris, alloca() is a #define which is inside <alloca.h>. 
Ctypes fails to compile because the #define is missing. Please fix by
adding the following at the front of these 2 files:
#if defined (__SVR4) && defined (__sun)
# include <alloca.h>
#endif

forgot to mention the files:
Modules/_ctypes/callproc.c
Modules/_ctypes/libffi/src/sparc/ffi.c 

A better solution would be to use the HAVE_ALLOCA and HAVE_ALLOCA_H
defines that fficonfig.h provides to decide whether or not to include
alloca.h. And in callproc.c whether or not to provide a workaround
using malloc (I'm assuming non-gcc sparc compilers also support alloca
for sparc/ffi.c, but I don't know for sure).

Turns out callproc.c forgot to include <ffi_common.h> after <ffi.h>
which conditionally includes alloca.h. So it's a one-line fix.

> Greg Couch added the comment:
> 
> Turns out callproc.c forgot to include <ffi_common.h> after <ffi.h>
> which conditionally includes alloca.h. So it's a one-line fix.
> 
This would not work. <ffi_common.h> is a file private to libffi;
when Python is configured to use the system ffi-library (--with-systemffi)
this file is not available.
I would tend to replace the three alloca() calls in the function
_CallProc() in callbacks.c with calls to malloc(). All the other
occurrences of alloca() are only compiled on windows.

That's a disappointment. <ffi_common.h> has the right logic in it
already. Perhaps it should be copied to callproc.c. I'm less concerned
about alloca not being there at all because it seems to be a pervasive
extension in non-gcc compilers, but using malloc is fine too. Please go
ahead and fix this as you see fit. I've started issue 1516 about using
non-gcc compilers to compile _ctypes/libffi.

Fix is inline.

I applied the patch to SVN trunk as rev 61520. It would probably be
better to have a configure test for alloca.h.