[Bug tree-optimization/78586] [7 Regression] Wrong code caused by printf-return-value

["jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78586

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Even for TYPE_PRECISION (dirtype) == TYPE_PRECISION (argtype), if one is signed
and the other is unsigned, are you sure that for VR_RANGE we do the right
thing?
Again, should be backed out by sufficient testsuite coverage.
E.g. consider:
__attribute__((noinline, noclone)) int
foo (unsigned int x)
{
  if (x < 64 || x > 2U * __INT_MAX__ - 10)
    return -1;
  char buf[4];
  return __builtin_sprintf (buf, "%d", x + 1);
}
Here we don't warn with -O2 -W -Wall, but if I change buf[4] to buf[1], I get:
warning: ‘%d’ directive writing between 2 and 3 bytes into a region of size 1
and
note: directive argument in the range [65u, 4294967285u]
But the directive is not writing in between 2 and 3 bytes, it is a number in
the range [65, INT_MAX] or [INT_MIN, -11], i.e. ~[-10, 64], thus it needs in
between 2 and 11 bytes.  For the warning that is a missed-warning case, but it
wouldn't surprise me if you couldn't construct a similar testcase perhaps with
sign -> unsigned or something similar (and octal or decimal, etc., many
options) where you'd be warning even on something that is safe.