I've now set up t2.math so it is able to build Sage again. The software I've
added includes
* gcc 4.5.1
* emacs 23.2
* vim 7.3 (vi clone, but handles big files better than vi)
* screen 4.00.03
Please add something like:
if [ `uname -n` = t2 ] ; then
. /usr/local/bin/t2-setup
fi
to your .profile and removal a line which you may have which points to an
earlier version of gcc.
Note I have *not* installed the GNU coreutils, which means it's not possible to
build a binary of Sage using 'sage -bdist' as that makes use of the unportable
'-a' option to 'cp'. However, that bit of bdist can be modified to use 'tar'.
In fact, by using tar in a more efficient way, I think 'sage -bdist' can be
speeded up for everyone. That's because part of bdist creates a tar file, then
extracts that file. It would be sensible to pipe the output of one tar command
to the input of another, which should save a lot of I/O, as much of this will be
done in memory rather than on disk.
Note the following variable is set:
SAGE_FORTRAN_LIB=/usr/local/gcc-4.5.1/lib/libgfortran.so
If you wish to build a 64-bit version of Sage on Solaris 10 (which does build,
but is pretty unstable), then you will need to set that to
SAGE_FORTRAN_LIB=/usr/local/gcc-4.5.1/lib/sparcv9/libgfortran.so
and export SAGE64=yes.
Please look at /usr/local/bin/t2-setup to see exactly what variables are set.
You might choose to unset some of them, or set them to something else. However,
those settings work for building Sage.
Dave
There must be another file that is trying to load that. I think .bahrc
will take preference over .profile. There must be something in your
setup, whch is trying to load that filel
> My .profile is formed by the lines
> if [ `uname -n` = t2 ] ; then
> . /usr/local/bin/t2-setup
> fi
Try adding that to .bashrc
> I suppose the t2-setup will be responsible for choosing the sun linker
> as above. I guess the missing linker would also be the reason why the
> upgrade of my Sage-copy on t2 failed.
The reason the upgrade failed is almost certainly you dont have the
right gcc, or some other option which was specified is messing things
up.
> Moreover, the welcome message tells:
> There is a pre-built 32-bit copy of Sage at:
> /usr/local/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS.tar.gz
/usr/local got removed. I need to remove that message
Simon,
I'm not at home, and are using someone elses computer without an ssh
client, so I can't log into t2.
There is not a binary on the system created by me. The buildbot might
have crated one, but IIRC the directories created are not publically
viewable, so you might have to either ask someone who has access to
the buildbot acount, or build it from source yourself. If you do the
latter, set SAGE_ATLAS_LIB to be /usr/local/ATLAS32 and it will stop
building atlas, but use a pre-built version. Building packages in
parallel, without ATLAS, the build is not as painfully slow as it used
to be.
dave
This should now be fixed for newer builds -- I recently set 'umask =
022' in the buildslaves' configurations.
> the buildbot acount, or build it from source yourself. If you do the
> latter, set SAGE_ATLAS_LIB to be /usr/local/ATLAS32 and it will stop
> building atlas, but use a pre-built version. Building packages in
> parallel, without ATLAS, the build is not as painfully slow as it used
> to be.
There's a 4.6.1.alpha1 t2 binary in
http://sage.math.washington.edu/home/buildbot/binaries/sage/4.6.1.alpha1