[PATCH] PR optimization/12324

[Jan Hubicka]

> On Fri, Oct 17, 2003 at 08:59:37PM +0200, Jan Hubicka wrote:
> > Actually you should have a followup containing patch that disables the
> > function deferring.  It looks like we fixed all the problems I had when
> > I tried that month or two ago.
> 
> Does that mean that the other couple of places we fiddle 
> output defering can be removed?

What places do you have in mind?  We still must deffer forward
declaration as they may re-appear (I will probably add interface for
that in varpool in future).  We also must maintain non-unit-at-a-time
mode that needs to keep the old scheme around unfrotunately.
But with the exception of forward declarations,
we can ignore the deferred flags in unit-at-a-time mode now.
> 
> > Concerning the C++ function deferral,  did you made some progress on
> > your plan you mentioned last month?
> 
> No, I never got back to it.
Real shame...  I still have some code moving the function deferring from
non-unit-at-a-time C++ frontend into cgraphcode and letting it to
deferred unreferenced functions as currently done via DECL_NEEDED_P hack
(so we also build tree representation of all used functions even when we
don't use them).  How much this is in conflict with your plans?

Honza
> 
> 
> r~