[tree-ssa] Simple out of ssa diagnostics

[Jan Hubicka]

> On Thu, 2003-06-12 at 17:13, Andrew MacLeod wrote:
> 
> > Oh, and I have no idea if I did the diagnostic bit right, but it works
> > for me on c , which is the only place I had a testcase producing the LOE
> > problem. :-)
> > 
> I know we talked about this earlier, but I've changed my mind :)  I
> don't mind the -debug bit, too much, but then I question what is the
> difference between -details and -debug?  Both are used when debugging
> mis-optimizations.  So, I wouldn't mind having it in -details.  In fact,
> we may want to have more than one level of details.  I'm not quite sure.
> 
> What I have changed my mind on is the warnings().  I think warnings are
> reserved for things that are related to the user's program directly. 
> So, emitting a warning on stderr for something that the optimizers
> messed up doesn't look right to me.  I would vote to emit these warnings
> in the dump files directly.  They're only interesting to us, anyway.

I don't have very good experience with this practice.  For profiling
code I do use same way (dumping profile corruption into dump files) but
apparently no one worry about that and the profile updating bugs tends
to stay unnoticed.

If at all possible, I would recommend to do ICE and require optimizer to
cleanup properly here.  (or are there cases where optimizer can't do so?
It looks trivial if I understand it right)

Honza

> 
> 
> Diego.