Re: PATCH to remove unnecessary NOTE_INSN_DELETED

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Bernd" == Bernd Schmidt <bernds@cygnus.co.uk> writes:

    Bernd> I have to agree here.  This is just asking for problems
    Bernd> when debugging the compiler.  Why don't we just add another
    Bernd> uid field to every insn?  We'd have one that stays constant

I disagree.  I expect the number of bugs that occur only when
instructions are renumbered, and not when they are not, to be very
small.  (Say, smaller than the number of bugs that occur when -O2 -g
is used but not when just -O2 is used.)  So, most of the time, you can
just turn off flag_renumber_insns.

In the bad case, you can use the tables to work out what's going on.
Yes, this will be a pain, but it's not that bad.  I've certainly
tracked worse processes, like keeping track of all the entries in the
CSE hash-table.

One of the wins of this pass is that we can scrap the LUID tables in
most cases; just run renumber_insns before the patch, and set max_luid
equal to luid.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com