Re: Question about libffi closure

[Andrew Haley <aph at redhat dot com>]

David Edelsohn writes:
 > >>>>> Andreas Jaeger writes:
 > 
 > Andreas> Should we do instead of:
 > 
 > Andreas> static ffi_closure cl;
 > Andreas> ffi_closure *pcl =3D &cl;
 > 
 > Andreas> use:
 > Andreas> ffi_closure *pcl;
 > 
 > Andreas> pcl = mmap (NULL, sizeof(ffi_closure), PROT_READ|PROT_EXEC,
 > Andreas> MAP_PRIVATE|MAP_ANONYMOUS, fd, 0);
 > 
 > Andreas> Ok, I'll try to change the testsuite...
 > 
 > 	And the above is more portable because...?  libffi is used by more
 > than just GNU/Linux.

We need to map executable memory somehow.  We are, as ever, open to
all suggestions of ways to do this portably.

 > 	Also, this area has caused problems on systems when GCC tried to
 > optimize through the pointer, so you need to test on a large variety of
 > systems before this can be changed.

We're only changing the testsuite.   We can ifdef where necessary.
  
Andrew.