As reported in PR 13839 I am (since quite some time now) unable to build a
working libc.so.

What I tried in between is all I could thing of:

 - used the same source and header files, booted 1.5.2 and rebuild
   libc -> got a working libc.so (that is: compiled with the old toolchain
   everything works)

 - downloaded mrg's last snapshot and replaced my local libraries & binaries
   with that, then rebuild includes (from scratch), build tools and then
   libc -> fails.

 - Matt offered me his tools directory, I used that and rebuild libc -> fails.

 - as a strawman, thinking maybe ntpd got in the way (we've seen that once)
   I rebuild everything running in single user -> fails (but with a bit 
   different, less often alignement errors)

My machine is working absolutely reliable. Running a current kernel and 1.5.2
userland works without any problem (and can rebuild a working libc.so from
current sources). Running with old -current kernels and -current userland
gives the same failure (so I suppose it is not a kernel issue). Using
"foreign" binaries and rebuilding on my machine gives the same failure -
which seems to imply it's not my (broken) binaries but something specific to
my machine or setup that makes it fail. I've once see a kernel compiled with
my home made tools break (kernel trap very early) - a kernel from the same
sources compiled with Matts tools worked.

I'm running out of ideas for (a) things to try, (b) possible explanations
and (c) methods to track this down.

Any hints?

Martin