Journeyman configure/build options for gcc & glibc?

I thought I had the whole soft-float thing on 8xx worked out long ago, until I
upgraded our root filesystem to the one from MontaVista's Journeyman kit, and
found that my glibc-2.1.3.so (which my binaries work when linked against)
doesn't work in place of the Journeyman glibc-2.1.2.so.  It's the usual
symptom: A simple floating-point printf gives the wrong answer.

So, my question: what configure options were used to build the gcc-2.95.2 &
glibc-2.1.2 in Journeyman?  In particular:

- Did the gcc configure include "-with-cpu=860" (or one of the other cpus's
  to make -msoft-float the default).

- Has the gcc specs file been hacked to pass -D_SOFT_FLOAT in the absence of
  -mhard-float (note that -mcpu=860 won't do this by default in gcc-2.95.2)

- Was CFLAGS set to include -msoft-float before building glibc?

- Any other suggestions :-(

Thanks!
Graham
--
Graham Stoney
Principal Hardware/Software Engineer
Canon Information Systems Research Australia
Ph: +61 2 9805 2909  Fax: +61 2 9805 2929

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Graham Stoney wrote:

I thought I had the whole soft-float thing on 8xx worked out long ago, until I
upgraded our root filesystem to the one from MontaVista's Journeyman kit,

The lastest CDK 1.2 has been released, so you may want to use that.

The MontaVista CDK is created to work as a complete kit.  I don't know
all of the details, but the binutils, compilers, run-time start up,
libraries, header files, and applications are all built to work together.
There is an 8xx set of everything that "just works" together.  Mixing
among various tools/libraries has never worked at any time in history and
that continues to be the case.


	-- Dan

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

On Tue, 15 Aug 2000, Graham Stoney wrote:

I thought I had the whole soft-float thing on 8xx worked out long ago, until I
upgraded our root filesystem to the one from MontaVista's Journeyman kit, and
found that my glibc-2.1.3.so (which my binaries work when linked against)
doesn't work in place of the Journeyman glibc-2.1.2.so.  It's the usual
symptom: A simple floating-point printf gives the wrong answer.

So, my question: what configure options were used to build the gcc-2.95.2 &
glibc-2.1.2 in Journeyman?  In particular:

- Did the gcc configure include "-with-cpu=860" (or one of the other cpus's
  to make -msoft-float the default).

- Has the gcc specs file been hacked to pass -D_SOFT_FLOAT in the absence of
  -mhard-float (note that -mcpu=860 won't do this by default in gcc-2.95.2)

- Was CFLAGS set to include -msoft-float before building glibc?

I believe the compiler has been modified to default to both
-msoft-float and -D_SOFT_FLOAT for ppc_8xx.

- Any other suggestions :-(

I recall the printf problem as something we specifically fixed.  I
asked the engineer who worked on it, and he said it was a varargs
problem, so the fix is in the the gcc package.

--
Joe Green [off-list ref]
MontaVista Software, Inc.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Hi Joe,

On Tue, 15 Aug 2000, Graham Stoney wrote:

I thought I had the whole soft-float thing on 8xx worked out long ago, until I
upgraded our root filesystem to the one from MontaVista's Journeyman kit, and
found that my glibc-2.1.3.so (which my binaries work when linked against)
doesn't work in place of the Journeyman glibc-2.1.2.so.  It's the usual
symptom: A simple floating-point printf gives the wrong answer.

Joe Green writes:

I recall the printf problem as something we specifically fixed.  I
asked the engineer who worked on it, and he said it was a varargs
problem, so the fix is in the the gcc package.

Thanks for the response.  I tried upgrading my root filesystem from Journeyman
to 1.2, and now binaries built with my compiler work correctly with the HHL1.2
libraries.  I never saw any varargs problems between my gcc-2.95.2 and
glibc-2.1.3 once the other 8xx fixes were applied (the same ones that are
now in the 1.2 SRPMs), so perhaps whatever got "fixed" in Journeyman was a bad
mod that got backed out again for 1.2.

Anyway, I'm happy.  My only gripe about the otherwise excellent Hard Hat kit
is the ugly symlink from /opt/hardhat/devkit/ppc/8xx/powerpc-linux -> /
which is necessary on the _target_.  If you added the following after
-dynamic-linker in the gcc specs file, this wouldn't be necessary:
    --rpath /lib --rpath-link /opt/hardhat/devkit/ppc/8xx/powerpc-linux/lib

Then the cross-compiled binaries will look for their dynamic libraries in
/lib on the target, rather than /opt/hardhat/devkit/ppc/8xx/powerpc-linux/lib.

(Well, actually I'm not thrilled with the gcc/cccp.c relocation hack either;
 this sounds like a job for configure --prefix to me, but aside from that I'm
 happy :-)

Keep up the good work!

Thanks,
Graham
--
Graham Stoney
Principal Hardware/Software Engineer
Canon Information Systems Research Australia
Ph: +61 2 9805 2909  Fax: +61 2 9805 2929

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

On Wed, 16 Aug 2000, Graham Stoney wrote:

I tried upgrading my root filesystem from Journeyman
to 1.2, and now binaries built with my compiler work correctly with the HHL1.2
libraries.  I never saw any varargs problems between my gcc-2.95.2 and
glibc-2.1.3 once the other 8xx fixes were applied (the same ones that are
now in the 1.2 SRPMs), so perhaps whatever got "fixed" in Journeyman was a bad
mod that got backed out again for 1.2.

Anyway, I'm happy.

OK, I'm glad to hear that.

 My only gripe about the otherwise excellent Hard Hat kit
is the ugly symlink from /opt/hardhat/devkit/ppc/8xx/powerpc-linux -> /
which is necessary on the _target_.

Yes, we agree.  We actually have already fixed this, but it missed the
cutoff for the CDs; it was easy to fix, but was unfortunately overlooked
until too late.  It will be one of the first updates we post.

(Well, actually I'm not thrilled with the gcc/cccp.c relocation hack either;
 this sounds like a job for configure --prefix to me, but aside from that I'm
 happy :-)

We have customers who want the installation location configurable at
install time.  gcc seems to be the only package that has a problem with
this.  I think we're working on a better solution for the next release.

Keep up the good work!

Thanks
--
Joe Green [off-list ref]
MontaVista Software, Inc.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/