more discussion about your combine issue.

[Kenneth Zadeck]

I was thinking a little more about the combine problem you were having
and i was wondering why it did not show up on other platforms.

While doing that i was replying to Richard Sandiford's problems with my
last patch
(see http://gcc.gnu.org/ml/gcc-patches/2007-09/msg01360.html).

Note the part of the mail where I comment about richard's need to
bracket a set of split insns with DF_NO_INSN_RESCAN.  I am wondering if
the reason that this comes up on the alpha is not that something is
being split but that something is being split that creates a REG_EQUAL
note. 

If it is, the the proper fix is the one that i suggested for richard:
change df_note_rescan so that it just returns if the insn does not have
an associated basic block (i.e. the insn was created but not inserted
into the stream).

This is the only plausible explanation that I have as to why this bug
has not surfaced before.

If this fixes the problem, the patch is preapproved.

Kenny