Re: GCC 3.0 Release Criteria

["Charles M. Hannum" <root at ihack dot net>]

I've just gotten back from yet another round of travel, so I have not
been reading this discussion as it occured.

That said, I must say that I am *deeply* disturbed.


It has already been the case for some time that Linux has largely
driven egcs releases.  This has, from my POV, been to the exclusion of
almost everything else.  I've found it difficult at times even to get
OS-independent bug fixes (e.g. to the ARM port) in because (AFAICT)
people were too busy arguing about yet another Linux/i386 kernel
problem.

The difficulty of getting bug fixes in, combined with the fact that
the releases have often specifically been timed to benefit the Linux
community, has *already* caused miniature forks.  E.g., the NetBSD
project has been forced to maintain our own GCC branch, because no
release of the compiler has ever worked well enough on all the
platforms we support.  This is certainly not due to any lack of effort
on our part; I've personally spent at least 2 weeks last year *just*
tracking down random compiler bugs.  And that doesn't include all the
other work we've collectively done, none of which anyone involved in
NetBSD got paid for.

I am also disturbed that, despite contributing considerable work
(including substantial ARM-related fixes, several fairly important
frontend fixes, all of the VAX and NS32k updates for egcs, etc.), that
no NetBSD platforms are considered `interesting' enough to even rate
on the `second tier' list.  This suggests to me that no matter how
much we contribute in good faith, that the GCC steering committee is
interested only in where it perceives the money to be.  I suppose from
Cygnus/RedHat's perspective this makes sense, but I do not think it's
the way that the steering committee should be headed.


To be quite honest, if this path toward even stronger connection with
Linux continues, I envision that we will eventually end up with a
completely separate compiler -- not because we want to, but because we
will have no *choice*.

I implore the egcs/GCC steering committee to consider what we can do
to avoid this happening.  (I have several suggestions, but I will
discuss them offline for the moment.)