Re: Compiler for Red Hat Linux 8

[Sergey Ostrovsky <sostrovs@kns.com>]

"H . J . Lu" wrote:

> I think that is the problem. When was the last time any Linux
> distributions used an unmodified *released* GCC? Probably never. None
> of the *released* GCCs was good enough to build a whole Linux
> distribution.

Quite a few people have built their own systems from sources using only
released
source tarballs of everything ( gcc including ). Check out
linuxfromscratch.org.
I've built 4 such systems, each has of approximately 90 packages installed
( yes, and X,
and teTeX, Qt, KDE, all the bells and whistles ), each
one built from sources. Gcc compilers used are 2.95, 2.95.2, 2.95.3.
Glibc versions used are 2.1.3, 2.2, 2.2.1, 2.2.2. Kernels that glibc was
built against are 2.2.15 - 2.4.4.
Binutils range from 2.9.x to 2.11.x.
Few packages had glibc-2.1 - 2.2 compatibility or pre-ANSI C problems, so
my modest knowledge of C was just sufficient to fix.
And I have yet to see my first segfault, oops, or whatever.

After glibc successfully built.

Because if the build of glibc breaks, there is not much sense in trying
anything else.
If you've built glibc, than the toolchain is good for everything, including
the kernel.
I found the gcc-binutils coupling pretty weak.
Almost every time when new glibc is out, it takes several attempts to chose
one of recent binutils
( H.J. Lu makes bugfix releases quite often, which is appreciated very
much) which will be OK to
1. build binutils
2. build gcc using (1) binutils
3. build glibc using (1) (2) guys.
In these attempt everything seems to matter - with or without -march=, -O#
switches
in CFLAGS, etc.
So I take is as kind of black magic.
Hope this "pure user" point of view helpful.

Regards,

Sergey Ostrovsky.