Re: PR4114, 4113, 4082, 4078, (part of) 4096 plus other reports in gcc-bugs

[Benjamin Kosnik <bkoz at redhat dot com>]

> Yes.  Based on all the input collected over the weekend, I will
> attempt to produce a patch for this issue tonight...

Cool. It looks like Alexandre has a good way to fix it. 

> Yes, since ref4 and ref5 usually produce identical results, I have
> switched ref5 to test the gcc mainline instead of 3.0.X.

Much appreciated, thanks.

> Beast, the alpha machine, is now taking over 24 hours to bootstrap and
> check on some days (due to other load on the machine) thus I have
> already scaled it back to 3 times a week.  If it would help, I could
> have it run mainline twice a week and the 3.0.X branch once a week.

Even once a week is fine. Since this is a donated resource, I don't want 
to step on any toes. Now that the libstdc++ testsuite has some kind of 
sane memory protection, hopefully people won't be scared to run 'make check'

:)

I check gcc-testresults on a daily basis, for every posted platform. I've 
found it's a pretty good way to avoid stupid mistakes. 

-benjamin