Re: Patch to detect invalid and missing ATTRIBUTE const/pure [take 2]

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu> 
 > 
 > Next task will be to start adding ATTRIBUTES where suggested by gcc
 > and test test test.

Hmm, so of course its not that simple.  Things appear to be strange
under the hood.  Consider my last patch:
http://gcc.gnu.org/ml/gcc-patches/2002-03/msg01477.html
and the following testcase:

When compiling with -Wmissing-pure -Wmissing-const, I'm getting this
one warning:

 > foo.c:14: warning: function `p2' may be candidate for attribute `pure'

Here's what I consider wrong:

1.  Function p2 *is* declare "pure" by the prototype, but
    mark_constant_function doesn't realize that when parson p2's
    definition.  For some reason, "const" is carried through
    prototypes to the function definition, but "pure" isn't.  If I
    mark p2's function definition itself pure, then it appears to
    work, but that seems like a bug in attribute merging.

2.  I would have thought that foo_p1/foo_c1/foo_p2/foo_c2 would all be
    suggested candidates for either pure or const.  None of them are.
    Is it not the case that a "pure" or "const" function may call
    another respectively "pure" or "const" function?

		--Kaveh

----------------------------
#define APURE __attribute__ ((__pure__))
#define ACONST __attribute__ ((__const__))

extern int i;
extern int p1 (int) APURE;
extern int c1 (int) ACONST;

extern int p2 (int) APURE;
extern int c2 (int) ACONST;

int p2 (int a)
{
  return i + a;
}

int c2 (int a)
{
  return a * 3;
}

int foo_p1 (int a, int b)
{
  return p1(a) + b;
}

int foo_c1 (int a, int b)
{
  return a + c1(b);
}

int foo_p2 (int a, int b)
{
  return p2(a) + b;
}

int foo_c2 (int a, int b)
{
  return a + c2(b);
}