Re: C++ compiler assertion errors

[Jim Wilson <wilson at cygnus dot com>]

In article <CMM.0.92.0.988497273.beebe@suncore.math.utah.edu> you write:
>During this work, I've encountered fatal compiler assertion failures
>on IA-64 in both g++ (version 2.9-ia64-000216)

This compiler is obsolete.  The IA-64 Linux project (aka Trillian) stopped
using it about 6 months ago.  This compiler is based on a Jan 2000 snapshot,
and is known to have serious bugs in the C++ front end, as the C++ front end
was undergoing a major rewrite at the time.  It likely also has a lot of
IA-64 specific bugs, as the IA-64 backend was still pretty immature at that
time.  The vast majority of these bugs were fixed long ago in the FSF gcc
sources, and will not be fixed in this obsolete release.

The compiler that the Trillian group currently uses is based on a June 2000
snapshot, and was last updated in November.  All current GNU/Linux releases
use this compiler.   This compiler should work much better.  I know that
HP was interested in releasing a new version of NUE using this compiler, but
I don't know when or if that will happen.  This compiler is not ABI compatible
with the older one, which means your NUE code won't run on any current
GNU/Linux distributions, and this compiler won't work with your NUE.  Also,
this compiler is in the process of being obsoleted, since I no longer maintain
it.  At this point, I'm not very interested in C++ problems reported against
this release.

The preferred compiler at this time is gcc-3 (pre-release).  However, it
hasn't been finished yet.  There are still some instabilities due to
development work, and there will be some more ABI changes that haven't been
implemented yet.  These ABI changes primarily affect C++ programs, but they
may also affect shared libraries in inconvenient ways.  So there is a good
chance that there will again be zero compatibillity with the previous
compiler.  It will probably be 6 months or so before there are GNU/Linux
releases that are based on it.  I expect that all GNU/Linux distributions will
move to this compiler when practical.  If there are C++ problems in this
compiler, then we definitely care.

Once we have gcc-3, there should not be any more major ABI changes.  Anything
before gcc-3 should be considered experimental.  It is possible that HP
decided to wait until gcc-3 was stable before updating NUE.  It is also
possible that HP decided that NUE was no longer worth the trouble of
maintaining given that IA-64 hardware will soon be publicly available.
You'd have to get an answer to that question from HP.

Jim