Re: [Axiom-developer] Failed to build latest axiom tree

[root]

> > I should be done later today and then I'll announce that the changes
> > are available.
> 
> Oops sorry. Later time I'll wait for your announcement.
> 
> Best regards,
> d.

No problem. I'm going to need someone to test this version anyway. 
The fact that *I* can build it doesn't mean anyone else can. I'm
testing the build on a clean machine just to make sure I don't
announce something totally broken (which, clearly, the current upload is).

While I compile it here I'll take the time to tell you what has changed.
The bulk of the work is in the src/algebra directory. There are about
350 files (.spad.pamphlet) that contain approx 1100 domains, packages,
and categories. The task has been to get the algebra to compile from
scratch (previously you required a compiled version of the algebra).
In order to do that I need to determine which domains are primitive
(depend on nothing) and which depend on each other. This lattice currently
contains about 900 of the files. You can see the lattice by looking at
src/algebra/Makefile.pamphlet.

On my 1Ghz machine it takes about 90 minutes to compile. Great effort
is being made in the Makefiles to ensure that once you've compiled
something no redundant work occurs next time you type "make". All
useful intermediate work is cached (splitting out pamphlets for source,
compiling source to lisp, etc). It might not seem worth it now as it
greatly complicates the Makefiles but I can assure you from experience
that the Makefile complexity pays for itself immediately.

Further progress is blocked by a bug. The bug causes an infinite
recursion during "has" category lookup. It is not clear where the bug
occurs but I suspect it was an optimization for CCL. I've been
ignoring it as long as possible so I can make progress on the rest of
the lattice but the time has come to hunt it down.

Nevertheless, the system is now capable of doing limited algebra.
For example, 1+1 works. That exercises a fair piece of the machinery.
The failing algebra is, unfortunately, scattered around the lattice
so it keeps the system from doing marvelous things. The symptom is a
message about being unable to load "FOO.o" which means that "FOO.o" is
one of the algebra files that won't compile yet.

My next task is to find and fix the infinite recursion bug and finish
building the algebra lattice and the associated databases. At that
point we will have a working algebra system. Assuming the bug gets
fixed tonight (ha) the rest of the lattice build should go smoothly
(ha^2). I'll do another upload at that point. Once other people can
get the full algebra code running it will be time to switch to using
the real website rather than tenkan and switch to really using CVS
rather than my current "bulk" method.

This build does not have any "user" packaging yet so the "axiom" command
does not yet really exist. If the build works you should be able to run
(yourpath)/obj/linux/bin/interpsys

I should caution you that, before you build, you need to change 3 things:

1) modify the (yourpath)/Makefile and change the SPD=(yourpath)
2) modify the (yourpaht)/Makefile.pamphlet and change the SPD=(yourpath)
3) export AXIOM=(yourpath)/mnt/linux

missing any of these three things will cause obscure bugs. I'll include
a note to that effect in the top level directory but I doubt it will get
read until after the fact :-)

I want to drive the first two changes off the AXIOM shell variable but
"make" seems to have changed since last I used it and my old trick fails.

Once this gets "announced" please let me know if you can compile it.

Tim