Compile #9 for the day
(If any of this doesn't make sense, please forgive the language skills
at this time of morning. I can try again later if you really need me
to)
I'm on glibc compile #9 trying to track down the i386-pc-linux-gnu
ldconfig miscompile.
I thought I had it. The only thing I can see different between my
build from the sources, and the package's is the configparam file.
I have so far also confirmed that:
If I run make by hand after having done ``debian/rules configure'',
the problem exists.
If I run make by hand after having done ``debian/rules setup'', the
problem does not exist.
I first took out the -fstrict-aliasing, thinking that it might be
causing a miscompile. That doesn't seem to be it.
Now I'm just trying to build having clobbered the configparam file by
hand. My default gcc is gcc-3.2 and it's a native build so everything
should be picked up right.
My gut guess is that we should leave the CFLAGS alone, and possibly
not tweak it in configparam at all. But all of my gut guesses so far
have been wrong...
I want to look at all of this stuff anyway once we get 2.3 deployed.
hurd-i386 cross-building got broken when we added the optimization
stuff in, and our link magic doesn't like the CVS directories - so
it's needs an overhaul to allow for these cases.
First let's get this one out the door. More info tommorow.
--
learning from failures is nice in theory...
but in practice, it sucks :)
- Wolfgang Jaehrling
Reply to: