On Sun, Nov 09, 2003 at 05:56:42PM +0100, Santiago Vila wrote: > > > Still, there are 19 tests which fail even when threads are disabled. > > > These are the build log and the patch I used to disable threads:
> > The only ones I'd take any time exploring are the ndbm tests. I would > > be worried about it produce bad databases or corrupting them. (There > > isn't enough here to see what the failures are). It might also be nice > > to look at the math errors. The other things (socket errors, pipe > > errors, and such) are all ignorable. > Interesting. Do these ignorable bugs mean bugs in the Hurd or something? Yup! Our socket implementation is broken in some odd ways. > > I'm hoping that glibc 2.3.2.ds1-10 will be suitable for hurd-i386 again, > > which includes a CVS update. Mostly it's now waiting on me getting a > > working cross compiler. > What do you mean, exactly? That you do not consider native compiling > reliable? That you do not consider native compiling reliable for glibc > specifically? That glibc does not allow native compiling now? That > glibc needs gcc 3.4? :-) Oh, it ought to support native compiling fine - But that's assuming your Hurd box lives that long. Mine often doesn't, so I tend to cross-build the glibc and gcc packages. You can also stop it and ``debuild -B -nc'' it occasionally. I first got into cross-compiling the Debian glibc packages when the filesystem used to randomly eat blocks. I think we've cured it of that habit though. I mention gcc-3.4 specifically because they've introduced a lovely new feature "--with-sysroot" that allows the compiler to look at a particular directory for the libraries, include files, etc. Building a cross compiler before was an annoying but of making sure the linker scripts were tweaked, and that new libraries went in the right directories and such. Gcc-3.4 solves this in a sane way. The patch has been proposed for the next gcc-3.3 release, too (due in January or so, but I'm watching for it to land on the branch) Tks, Jeff Bailey
