Re: Patch: malloc function attribute

[Jeffrey A Law <law at cygnus dot com>]

  In message <199911290635.BAA01815@caip.rutgers.edu>you write:
  >  > From: Anthony Green <green@cygnus.com>
  >  > 
  >  > Jeff wrote:
  >  > > Otherwise it looks great.  Please install it.
  >  > 
  >  > Done (with your suggestions).  Thanks,
  >  > AG
  > 
  > Hey great!  I suggested this in September hoping someone would
  > consider it to be worthwhile, but the discussion trailed off when no
  > one could decide if __restrict__ would already satisfy the
  > functionality.  (Any idea about that?)
I don't remember getting a conclusive answer on that issue.  Also note
that attributes are probably more likely to work in a language like Java
than __restrict.  ie, __attribute__ is *supposed* to be a fairly language
independent way to tag items.  __restrict on the other hand is fairly
C/C++ centric.

  > QUESTION: should realloc and similar functions which could possibly
  > return the same pointer more than once have this attribute?  I don't
  > think so, so I didn't mark xrealloc or spaces.
I believe xrealloc is safe.  Consider the conditions under which is
can return the same pointer, I believe all are safe.
 
  > list.  (BTW, what the heck is spaces good for anyway?)
No idea.  Who added it?

  > gcc:
  > 	* calls.c (special_function_p): Add `strdup' to the list of
  > 	functions which have attribute malloc by default.
  > 
  > include:
  > 	* ansidecl.h (ATTRIBUTE_MALLOC): New macro.
  > 
  > 	* libiberty.h (buildargv, dupargv, concat, choose_temp_base,
  > 	make_temp_file, xmalloc, xcalloc, xstrdup, xmemdup): Add
  > 	ATTRIBUTE_MALLOC.
  > 	(xatexit): Remove __GNUC__ check, add ATTRIBUTE_NORETURN.
The only concern I have is people bootstrapping with older snapshots may
get errors because there was no malloc attribute supported in older snapshots.

How do you propose to handle that?

jeff