>
> Not the compiler is broken but the source code! See
> http://www.bero.org/gcc296.html for more information. After a close look
> of some of my own source and contemplating about Beros comments I
> finally got to the conclusion Red Hat did a Good Thing (TM) deciding to
> use gcc 2.96-RH. But decide yourself. It may take *some* fixing in the
> XFS code but after that you're standards compliant, so your code should
> work with any compiler (that is, if that compilers is ANSI C compliant!
> Not all are...)
>
> FWIW I asked on the Red Hat mailing list about XFS and RHL, but no
> employee of Red Hat shed any light to the issue...
>
> Best regards,
> Martin Stricker
Actually, the examples given on this page do not apply to xfs, what does
appear to be the case with xfs is that it exercises the 64 bit handling
of the compiler more than most things do. 90% of the code base is also
the Irix code which works just fine with the SGI compilers on a 64 bit
architecture.
The challenging problem is going from a kernel that crashes due to code
not being compiled as expected, to finding the line of code in the 122
thousand lines which make up XFS and it support code (comments not stripped
in that number).
Oh and add to that the fact that some build types seem to run fine and
some do not, that suggests bad code generation to me.
Unfortunately, getting to the bottom of this is a big task, and there are
compilers which are:
a) still the recommended kernel compiler
b) do not exhibit problems with xfs.
Steve
|