PR fortran/18899: ubound() failing for assumed-shape arrays

[Richard Sandiford]

Steven Bosscher <stevenb@suse.de> writes:
> This entire test case optimizes away to nothing for me with Richard
> Guenther's array aliasing patch.  You should probably add an explicit
> interface for test and not make test CONTAINED.

Is that a bad thing?  We're only trying to test what the frontend
produces, and if the frontend produces something that the optimisers
realise is a no-op, that's fine.

If any of the ubound()s didn't evaluate to the expected values,
we'd have a call to abort(), and no optimiser has the right to
get rid of that ;)

Richard