[Bug middle-end/93582] [10 Regression] -Warray-bounds gives error: array subscript 0 is outside array bounds of struct E[1]

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93582

--- Comment #17 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #16)
> I'll play with it next week.

What can be handled specially is certainly size2i == maxsizei for the
known_subrange_p case.  Then we can pun the RHS directly without
going through native encode/interpret.  But I always thought we have
too much special-casing there anyways - more generalized bit-precision
and offset aware native encode/interpret handling would be appreciated.

As for the more complex case of handling non-constants mixed with constants
I do have some patches in the queue for stage1 for the partial overlap case
but only for byte offsets/sizes again.

Note there's always the question as to what replacements we want VN to make
since an extra load can be quite cheap sometimes compared to having a live
register that needs to be combined.  VN doesn't know whether the memory
can be elided later to offset such cost :/

For bitfields there's also the ever present bitfield-lowering idea...

And the fact that optimize_bit_field_compare is done way too early.