plans and wishes

Ryan.Oliver at Ryan.Oliver at
Thu Jan 15 18:12:55 PST 2004

Alex Groenwood wrote:
> Ryan Oliver wrote:
> > I'm wary of altering ch5
> I'm not.  But then I've got great confidence in the bootstrap of GCC.

And you think I don't ?

I have great confidence in the toolchain build (hell, I wouldn't
have given it to you guys if I didn't :-) ). It's the rest I'm wary of
at times (sometimes unfoundedly).

Initial build was published the way it was because
a) everything was tested with the testsuites
b) everything PASSED said testsuites.
c) still held true from many varied starting hosts,
   even from thoroughly broken hosts (you should see some of my
   ancient systems, you'd cry).
d) also worked on sparc32 (my testing), ppc, alpha (later testing)

Nothing was left to chance. This will be the yardstick.

BTW big thankyou should go out to Toby Thain ( Telegraphics ) and Finn
Thain for the loaner of a DEC Alphastation 400 for the testing effort.

> Yet we use the make and sed and gawk and grep and cmp and install,
> and what not, from the host to compile and install Binutils, GCC,
> Glibc and Coreutils, the most basic building blocks of the temporary
> system.  If those programs are good enough for the basic toolchain,
> then they are good enough for the rest of the tools too, I should
> think.

Try using an old sed with our current instructions all the way through.

Binutils GCC and Glibc when built in the correct order don't give a damn
about the host.

Binutils and gcc have been architected to be able to be built on
near any OS and hardware for near any OS and hardware with gnu and
non-gnu tools.

(binutils builds with egcs, gcc looks after itself with its bootstrap
 regardless of starting compiler, glibc is built with our new tools
 and it's only concern is make version).

Just cos the toolchain builds no matter the state of host tools doesn't
mean the rest will (trust me, I lost a fair few months of my life to
this fighting with older hosts).

90% of the time it may work, that still leaves 10% clogging support
(not counting PEBKAC errors).

> Including a list of minimum required versions wouldn't be a bad
> idea.
> I've been able to build an LFS4.8 system (a half-updated 4.1) using
> gcc 2.95.2, binutils 2.9.5, glibc 2.1.3, make 3.79.1 (Suse 7.0).

No dice from that setup nowadays without first building a new make.

Well, looks like us test guys have a bit of work on our hands :-)

Anyone want to join in I have a few little requirements...

1) build the test tools first (TCL, Expect, Dejagnu) if your hosts
   are not current.
2) run testsuites for everything
3) keep full configure build test and install logs (preferably
3) prepare to lose a month or two out of your lives.
4) NO CHEATING, strict book instructions (unless you're not on x86)

hopefully it will all just work...

Ideally you'll find the oldest Distro you can get your hands on,
(not too old, circa RH6. You'll probably need to update hosts make)

Time to pull the old sparcstation out of mothballs again...


More information about the lfs-dev mailing list