[Bug tree-optimization/86572] unsafe strlen folding of const arguments with non-const offset

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86572

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
When we can detect undefined behavior we should try to behave as the user
intended.  For example for

  MEM[p_1 + 4, int] = 0;
  MEM[p_1, double *] = 1.;
  x_2 = MEM[p_1 + 4, int];

we see a must-alias and ignore that TBAA allows us to simplify x_2 to 0.

In this particular case folding happens with non-constant offset but
the outcome for large i could be considered a bigger security issue
than not folding (which likely would have crashed or resulted in a
smaller result).

MAX(0, 3 - (ssizetype)i)

might be indeed a way to limit the possible fallout.  Or even

len = 3 -(ssizetype)i;
len < 0 ? __builtin_trap () : len;