Bootstrapping and the Kernel

[Greg Schafer]

Hi Guys

I've wanted to raise this issue for quite a while. Stumbling across this
post reminded me of it: (Hmm, poster's name looks familiar! :-)

  http://sources.redhat.com/ml/glibc-bugs/2006-03/msg00029.html

I believe Jon is spot on. We do not place enough emphasis on the role of the
running kernel. It's really quite ridiculous that we measure the
effectiveness of a Glibc build by running its testsuite against a kernel we
didn't even compile ourselves. I mean honestly, Glibc is so tightly bound to
the running kernel it ain't funny.

To fully achieve a proper bootstrap in the true sense of the word, we need
to run the Glibc testsuite (and most everything else for that matter) under
a running kernel we built ourselves as part of the bootstrapping process.
Not sure this is achievable considering it's pretty much impossible to
script. But I think the issue needs to be highlighted so I'll try and work
something into the docs. Theoretically, GCC-Pass1 (even tho' it's a "mini"
GCC) should well be capable of building a kernel.

Incidentally, this is why I believe so called "live" CD's are fatally flawed
when used for the purpose of bootstrapping a new system. AFAICT, it's
impossible to boot your own self-compiled kernel in a Live CD scenario (tho'
tools like kexec might be able to help here. dunno.)

Regards
Greg