[Gllug] updating red hat

[Nix]

On Mon, 29 Oct 2001, Ms. Lene Jensen said:
> Red Hat never shipped a broken gcc,

I dispute that. The GCC which was shipped with RH7 (as opposed to the
copy that eventually emerged after all the errata) was definitely *not*
release-quality (and so saith the Steering Committee). It generated
noticeably worse code than either GCC-2.95.2 or the eventual GCC-3.0,
was rather slower than 2.95.x, and frequently ICEd. The fixes folded
into that branch made it *much* more usable as time went by, and the
torture-testing which it triggered and enabled was the direct cause of
many GCC bugfixes, but it was not what I'd call `working'. Certainly it
was not working *well*.

And it broke C++ binary compatibility *again* when the GCC team wanted
to inflict at most one more such breakage on the long-suffering world
(GCC 3.0's). (Thank God that's over --- nearly. libstdc++ is still not
frozen, but the ABI itself is).

>                                     what we did, for RH7 only, was to ship
> a gcc which was newer than the kernel packages needed.

Newer than *any* existing software needed (well, except for evil
template-expression monsters like Blitz++, which RH's GCC behaved worse
with than did GCC-2.95.x... but Blitz++ is so pathologically wierd that
no blame attaches anywhere for that :) )

I'm still not sure why RH did this; I heard that it was because of IA64
support, but IA64 support is hardly functional even in the RH 2.96
branch, and getting it functional would be a huge job. So that can't be
it.


But I've ranted about this before and I shouldn't be doing it now --- my
excuse is that I'm so tired I can barely think, and it's making me run
off at the keyboard ;)

> kernel, you needed an old version, which we called kgcc.  For RH7.1 and
> RH7.2 we are using the same version as anyone else.  The gcc was never
> broken, it was just too new for the kernel.  Our gcc was version 2.96, the

There is no such version. 2.96 was an internal GCC development snapshot
number, and corresponded to no released version. RH's GCC was a branch
from a weekly snapshot mixed with bits of GNUPro, heroically brushed up
to something like functioning by (IIRC) Jeff Law, rth, and a few others.
Personally, I call it 2.96-rh; certainly it was not 2.96 (which
numbering implies an official release, which it was not).

> kgcc was 2.95, which all the other distros used at the time.

IIRC, at that point, 2.95 was not kernel-certified either (although in
practice it worked OK).

Oh, and you shipped with a beta glibc too (2.1.92, IIRC).


Definitely not production-ready. A bleeding-edge release, 7.0 was. I'm
amazed it worked, let alone as well as it did. That it worked despite
the compiler and libc versions is a credit to you all :)

-- 
`You're the only person I know who can't tell the difference
 between a pair of trousers and a desk.' --- Kieran, to me

-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug