Re: Errors on trunk since a while

[Richard Guenther <rguenther at suse dot de>]

On Tue, 29 Aug 2006, Paolo Carlini wrote:

> Richard Guenther wrote:
> 
> >I get since a few days
> >  
> >
> What do you mean by "I get"? When? Where? Apparently, *nobody* else is seeing
> that error on testresults during normal make, make check...

Sorry for being imprecise.  This is from the nightly builds of trunk used
for the c++bench tester, a single file testcase is tramp3d, but failures
include boost wave and DLV - so I suspected it must be a common problem.
(Try http://www.suse.de/~rguenther/tramp3d/tramp3d-v4.cpp.gz)

> >any ideas?
> >  
> >
> I don't know, recently __convert_from_v has been changed in mainline to a
> vararg function, maybe you could start from that... But the entire issue is
> deeply mysterious: __convert_from_v is used eminently by num_put of floats (vs
> money_put) and you should also see it there... A specific feature of money_put
> is that we are passing a constant string literal, "%.*Lf", maybe also
> something to consider.
> 
> Are you sure the tool chain is solid, no experimental linkers on that machine,
> things like that?!?

Hm, the trunk gcc is not bootstrapped but built by gcc 4.1.x (from suse 
10.1), maybe there is a problem.  Also this machine is somewhat flaky, 
but svn status or svn diff don't show differences (would they?).

I'll do a fresh checkout of the tree to make sure it is not contaminated.

Thanks,
Richard.

--
Richard Guenther <rguenther@suse.de>
Novell / SUSE Labs