[Bug tree-optimization/110603] [14 Regression] GCC, ICE: internal compiler error: in verify_range, at value-range.cc:1104 since r14-255

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110603

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu.org

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Aldy Hernandez from comment #4)
> Now the reason we're passing swapped endpoints seems to originate in
> get_range_strlen_dynamic().  It is setting a min of 2, courtesy of the
> nonzero characters in the memcpy:
> 
> memcpy(a, "12", sizeof("12") - 1);

Guess in a program without UB both the bounds are valid, a zero terminated
string in
char[2] array can't have strlen longer than 1 and when '1' and '2' characters
are memcpyed at the start of some buffer then the string length will be at
least 2.
But the program would invoke UB if this code is reached, so the question is how
to resolve it.
The old behavior of VRP/ranger with swapping the boundaries avoided the ICE but
wasn't
right, this case isn't that the string length will be in [1, 2] range, but that
the argument will never be a valid zero terminated string.
So, guess either we shouldn't set minlen or maxlen (whatever is found second)
if it violates the other bound, or check it after the fact and pick just one of
them or set the other to one of them.