Re: [boost] Compiler status for GCC 3.3

From: Beman Dawes (bdawes_at_[hidden])
Date: 2003-06-29 19:31:18

At 10:41 AM 6/29/2003, Reed Hedges wrote:
 >-----BEGIN PGP SIGNED MESSAGE-----
 >Hash: SHA1
 >
 >
 >Hello. I was building boost with GCC 3.3, so I thought I might as well
 >run the regression tests.  Here is the output:
 >
 >   http://interreality.org/~reed/boost-gcc-3.3
 >
 >I ran the tests on Linux (Debian 'unstable') and on Darwin (Mac OSX
 >10.2) with Boost CVS (just used the "tools/regression/run_tests.sh"
 >script).  It is possible that the Linux results are incomplete or
 >otherwise wrong since at one point I ran out of HDD space and just
 >re-ran the test after I had freed it up.
 >
 >There were some ICE's in Darwin GCC -- shall I submit bug reports to
 >Apple on that?  I think that the GCC 3.3 in Debian is a more recent
 >version, while the Apple GCC is a few bugfixes behind the version in
 >Debian, and those ICE's aren't happening on Debian.

Yes, please do submit bug reports to Apple if you think it is an Apple 
problem.

But some of the problems are clearly GCC bugs. For example, all the ublas 
tests are failing with this message:

...error: due to a defect in the G++ 3.2 ABI, G++ has assigned the same 
mangled name to two different types...

This is happening on Apple, Linux, and Windows, so that bug report should 
go to GCC.

GCC 3.3 (and 3.2.x, too, I think) also has some pretty bad behavior as far 
as memory usage goes. It uses 2.5 gigabytes on random_test (on both Win 2K 
and XP) before dying. It also outputs weird error messages about heap 
problems if you run bjam with the -j2 option (which tries to execute the 
build and test as multiple processes.) Too bad, because -j2 speeds bjam 
runs up at least 20% on my P4 with hyperthreading.

So by all means, do submit bug reports.

Thanks,

--Beman