[tree-ssa]: GIMPLE lowering array references too early?

[law@redhat.com]

In message <Pine.LNX.4.44.0307291246230.3531-100000@nondot.org>, Chris Lattner 
writes:
 >
 >> On Tue, 29 Jul 2003, Daniel Berlin wrote:
 >>> We used to keep array references around (instead of transforming into
 >>> address arithmetic), the thinking being that loop opts want them.
 >
 >> This seems to be the fault of the c front end, not the gimplifier.
 >> It only builds array_refs in certain cases.
 >
 >I have also observed this as well, and would like this to change.  Does
 >anyone know what the repercussions of changing the CFE would be?  In
 >particular, it now only generates an ARRAY_REF if the first operand is
 >actually an array, and now it could generate it if the first operand was a
 >pointer as well.
 >
 >Would this break everything?
It's a source of annoyance.  I had started cobbling together some code to
do this, but things kept breaking downstream.  Probably due to my lack of
knowledge about reform the ARRAY_REFs/ARRAY_TYPEs.

I've seen real codes which suffer because of the inconsistencies in how
we generate trees for array references depending on whether or not the
underlying type is a pointer or an array.

jeff