Re: Backport Policy Question

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

François-Xavier and Jerry,

> But the second point means that we need to decide now what features we
> want to integrate into 4.3, and start working on those rather quickly,
> because stages 1 & 2 don't last forever.
I propose that we hold a #gcc meeting to "decide" these matters.  It 
would be very helpful to have all the developers present and a menu of 
possible new features to discuss.

Aside from what appears on the Wiki, I seem to recall some weeks or 
months back that there was somebody who was going to implement C 
indexing for gfortran.  If that is an active project, we want to see it 
asap.  Also, we should consider sorting out the stride of pointers, so 
that we can pointer assign or pass to procedures components of derived 
type arrays.  I am painfully aware of the kludge that I introduced into 
trans-expr.c for the latter and that there is a PR out because it 
doesn't behave entirely as it should.

Speaking entirely for myself, I would rather like to see 4.3 as a 
consolidation phase, where we all but eliminate rejects-valid, 
wrong-code and ICE-on_valid's, where we address at least some the 
diagnostic failures in the Polyhedron diagnostic test (FX has already 
been busy there) and that we have another performance bee (I know that 
there is backend work going on, which is relevant here but I 
suspect/know that we have some unresolved business too - MOD for 
example).  On the latter, it would already be a big step forward to 
identify where the bottlenecks are in the five poor performers in the 
Polyhedron suite, when run on 64 bit machines.

'nuff said!

Paul