[Bug sanitizer/62307] -fsanitize=undefined doesn't pay attention to __attribute__((returns_nonnull))

[zackw at panix dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62307

--- Comment #2 from Zack Weinberg <zackw at panix dot com> ---
(In reply to Marc Glisse from comment #1)
> -fsanitize=null seems to imply -fno-delete-null-pointer-checks, so I assume
> this is on purpose. It would actually be quite natural for the sanitizer to
> insert an extra check after every call to a returns_nonnull function,
> checking that the result is indeed !=0. Otherwise yes, sanopt is way too
> late for any other optimization to take place.

I had a different mental model of how ubsan was supposed to function -- I was
expecting it to insert explicit checks before every construct with potentially
runtime-undefined behavior (not due to array bounds or pointer lifetimes), but
then for the compiler to try as hard as possible to *eliminate* the checks
where provably unnecessary.  I was hoping, in fact, to be able to use
-fsanitize=undefined -fdump-tree-optimized as a poor man's "tell me all the
places where the program can't be proven *not* to have runtime-undefined
behavior."