[incremental] Patch: FYI: more late gimplification clean up

[Diego Novillo]

Tom Tromey wrote:

> The first is diagnose_omp_structured_block_errors.  I just turned this
> into an ordinary pass, which is run near the start of
> all_lowering_passes.  I don't really understand why this couldn't be
> done this way on the trunk.  But, since a lot of the test suite is not
> working on my branch, perhaps the answer is waiting to be found.

IIRC, it may have been a remnant of when we would expand OMP constructs 
directly during gimplification, but I may be misremembering.  As long as 
we do this before final OMP expansion, this can be turned into a pass.

> The second is c_gimple_diagnostics_recursively.  Here I took a
> potentially controversial route: I added a new langhook for a
> post-gimplification pass.

Oh noes! ;)

Seriously, it does not matter.  OpenMP validation and initial processing 
is highly front end specific, so none of these activities will ever be 
done from language-independent layers.  A langhook here is just fine.


Diego.