potential for using Sage build farm for Maxima ports

Robert Dodier wrote:
> Michael,

Hi Robert,

> If I am not mistaken, the Sage project has a collection of different
> systems which are used for testing Sage ports.

Yep.

> What is the potential for using the Sage build farm for testing
> Maxima ports? 

I see no problem doing so since we are helping ourselves here, too :)

> Specifically I am interested in testing Maxima + ECL
> and Maxima + GCL.

Mhhh, we are finally switching this week to ECL after talking about it 
for more than six months. I have gotten Maxima 5.16.3 to pass its test 
suite by building clisp 2.47 on Solaris 10/Sparc with a gcc 3.2.3. If 
anybody wants that as a binary (since the blastwave site has an ancient 
2.28 clisp only) I am more than happy to provide it.

We have been setting up all the needed bits (i.e. buildbot and many 
[D]RCS clients on those boxen) and are in the process of adding various 
projects to the list we test. Note that the truly interesting boxen are 
behind firewalls and have severely limited connectivity, i.e. only one 
box can even get to the outside on every port and the other ones are 
usually unable to resolve external IPs, i.e. no email to the outside. 
But given that I need to check on the status of my builds I am there 
nearly daily anyway and can forward things to the list/someone like you 
who will deal with the results then.

The regular stuff (about 20 VMWare images running 24/7 on a nice new 24 
Core/128 GB box with 12 Linux distros, FreeBSD, various Solarises, 
Cygwin and more in the future) are easily accessible and I can set up a 
buildbot that can also test on OSX 10.4, 10.5 and so on. In the near 
future you should also be able to log into those VMware images with you 
sage.math account directly. If you need an account just (recently all 
account holders who had not logged in for six months had their access 
disabled) ping me off list.

Re GCL: I have build 2.6.7 as well as 2.6.8CVS on all the "interesting" 
machines and at least the naive "configure && make && make install" 
fails on all boxen, but one Itanium. When I complain about gcl's build 
system it is exactly from that experience since in most cases it was 
instantly clear from the error message that libbfd.a was missing from 
the system or some equally trivial configure problem was to blame. I 
guess when trying again I ought to use the internal libbfd and see how 
far I get. I have some sbcl, cmucl on various Solaris boxen and can 
likely build on more of them as needed.

There is also someone now working on packaging ecl in 32 and 64 bit 
Windows and building Maxima on top of that ought to be interesting. We 
plan to not use the in tree gmp and boehm_gc, but the latest upstream 
there and also build ecl with the good options, i.e. --small-scons and a 
couple other flags I do not recall at the moment to get the best 
performance out of it and also configure boehm_gc to use the large mem 
setting since IIRC the default is still a max 1GB heap. Since my 64 bit 
Windows box has 16GB Ram that seems like a waste and given the fact that 
RAM is dirty cheap I see little benefit from limiting Maxima in Sage's 
Windows port to 1GB. Packing Maxima + ECL to an MSI installer should 
also be something we can provide in the not too distant future.

> For the record, aside from ensuring that run_testsuite succeeds,
> I am hoping to find out whether the stuff around the edges
> (plotting, on-line documentation, anything that causes file i/o,
> console interaction, maybe other stuff) works as expected.

Hmm, what requirements (X, gnuplot, ....) are needed here? I do have 
shell access only, but can tunnel into the boxen to test things via X, 
but if it isn't automated I will likely only be able to do this rarely.

> Thanks for your help,

I won't be getting to this until at least next Tuesday since I am at 
Sage Days 12 and have been sleeping about 5 hours a night for the last 
three nights and way too many tickets to fix until I leave on Monday. 
You have been to SD 8.5, so you should get an idea what it is like :)

Either way, please ping me toward the middle of next week if you don't 
see any sign of life from me end since I tend to always have to put out 
fires and things like the above then gets preempted. It goes like the 
old saying that the squeaky wheel gets greased, so no need to be subtle 
when things don't happen.

Depending on the signal/noise ratio for the rest of the list we can take 
this private if you so desire.

> Robert Dodier

Cheers,

Michael

> _______________________________________________
> Maxima mailing list
> Maxima at math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
>