Fix PR45586 (type confusion ICEs), take 4

[Mikael Morin]

On Friday 18 February 2011 20:11:07 Michael Matz wrote:
> Hi,
> 
> On Thu, 17 Feb 2011, Mikael Morin wrote:
> > > Regstrapped on x86_64-linux, no regressions.  I'm assuming that the
> > > testcase for pr47455 (typebound_proc_20.f90) goes into the tree when
> > > that bug is fixed for good, so I'm not adding it for this one.
> > 
> > Given how much time some of the bugs need to get fixed, I'm a bit
> > inclined to commit the testcase as compile only test now. What do others
> > think ? (It's not much of a problem as long as we have Dominique around
> > ;) )
> 
> Next iteration.  Still fixes the bug, but doesn't regress either test
> from pr47455 (thanks Dominique!).  I'm now deferring to build the variant
> until the input type really is final (has been layed out), otherwise some
> fields might still be missing (the one case of pr47455 ==
> typebound_proc_20.f90), or at the very least the fields don't have their
> place (offset/size) yet (other case of pr47455 == typebound_proc_21.f90).
> 
> Now I also don't need to call mirror_fields from outside trans-types.c,
> hence made it static again, and not renamed to gfc_mirror_fields.  It's
> called only once, but I thought it be a good abstraction hence didn't fold
> it back into gfc_nonrestricted_type.
> 
> I've added both testcases that Dominique pointed out, as compile only.
> 
> Regstrapped on x86_64-linux, no regressions.  Let's see if Dominique finds
> another problem ;)  Otherwise okay for trunk?
> 
Yes, OK
Dominique, we're holding our breath...

Mikael