[Axiom-developer] Re: Axiom and Sage

[William Stein]

On Dec 9, 2007 1:00 PM, root <address@hidden> wrote:
> >The main plus of this packaging for sage is that it builds from
> >source quickly (in a few minutes) using precompiled clisp files.
>
> Well, on my 2Ghz machine with 2 Gig of memory running VMWare and
> using the sage vmware image (but upgrading the VM to have 1G memory)
> I started the package-install at 3:30am this morning. It is now
> 2:10pm and the build is still "in-process". They are heavy things,
> your minutes :-)

On a 1.8Ghz opteron (sagemath.org) it takes 18 minutes (I just tested
the install).   So it taking longer is a bug.  Thanks for reporting
the problem.
This is now
   http://trac.sagemath.org/sage_trac/ticket/1439

> Likely a portion of the problem is due to starting the package-install
> from the notebook. I'm running native windows and sage in the VM and
> connecting thru the browser.

That is very likely the problem.  The output of installing packages through
the notebook is way too verbose, so this is in fact a likely source of
the problem (which could be remedied).   Better would be to login as "manage",
do "sudo su" (password sage), then type "sage -i fricas-0.3.1".

> I suspect a lot of CPU is going into running jsMath to redraw the
> output page. The Fricas build has a lot of output (which could be
> suppressed during package-install) and jsMath is not fast. Axiom

jsmath isn't likely not relevant -- the output of the fricas build has
nothing to do with jsmath...

> has a NOISE variable in the Makefiles to suppress most output.

Excellent.

> You might consider a note suggesting that installs be done from
> inside the virtual machine rather than thru the notebook interface.

Even better would be to fix things so they work through the notebook well.

    http://trac.sagemath.org/sage_trac/ticket/1439

> I didn't realize that package-install was going to try a dynamic
> fetch and build. Shipping an Axiom source tarball would be faster
> and cheaper.

<joke> Would you give me a special software-developer's discount :-) </joke>

> Three other semi-related comments...
>
> Re: Lisp builds: Axiom is designed to be constructed directly "from
> the lisp" by shipping the pre-built "int" directory which caches all
> possible machine-generated, system-independent work during builds.
> This was used to minimize build times on 6 megahertz machines.
> Fricas uses a different build mechanism so I'm not sure what
> Waldek does.  I'll check to see if Axiom's mechanism has any bugs
> since I don't use it regularly despite the design point.

Let me know.

> Re: jsMath: I started using jsMath but Arthur Ralfs modified Axiom to
> output mathML directly (using )set out mathml). All Axiom needs is a
> mathml enabled browser with the correct fonts. I found that this is
> very much faster and lighter-weight than jsMath. The browser has
> machine code to handle mathml but jsMath is interpreted. You might
> want to reconsider this design decision because it will impact the
> long term speed of your notebooks. Arthur (copied on this email)
> might be able to give some advice.

Two points:
  (1) "All Axiom needs is a mathml enabled browser with the correct fonts."
   This is a nontrivial assumption to make. E.g., I have this on none of
my web browsers and mathml doesn't ship with browsers yet.
Of course, I can't see any reason for somebody to not develop
this for Sage as an option (that the user could configure in their sage notebook
preferences panel -- which doesn't exist yet).  One would use
some sort of latex --> mathml converter probably.
  (2) jsmath looks identical to tex, since it is 100% implementation of the
tex layout engine.  I don't think the same is true of mathml plugins,
i.e., mathml
doesn't look as good to me.

> Re: Source Code Control: On a more religious note, it appears the
> package was downloaded from a mercurial repository (asssuming I
> understood correctly).
> After trying many of the repo software packages
> (cvs, svn, darcs, arch, sourcesafe, but not mercurial) I find that git
> is blindingly faster than anything else. Git is so fast that when I
> first started using it I used to refetch the sources to make sure it
> actually worked.  Axiom uses Arch, CVS, SVN, and git but I shadow the
> other systems from the git master. Git has changed the way I work which
> is not something that happens often, given that I'm an old curmudgeon.

Mercurial is in some cases faster than GIT and other cases slower.
See this page on the GIT website (hence possibly skewed in their favor):

   http://git.or.cz/gitwiki/GitBenchmarks

In any case, Mercurial is plenty fast for the needs of Sage, and the code
of mercurial is easy to read and understand if need be.

 -- William