[Axiom-developer] Re: libaxiom.a

[Ralf Hemmecke]

Hi Bill,

let me CC this to axiom-developer since there might be some more people 
who know how Axiom works internally.

A sort summary. Last night I tried to compile libaxiom.al -> libaxiom.a. 
That did not work because the Aldor compiler told me.

[...]
ar x libaxiom.al basecliq.ao;
/home/hemmecke/software/Aldor/aldor/linux/latest/bin/aldor -Y . -fo 
-csmax=0 -laxiom *.ao;
Compiler bug...Bug: Bad case 3 (line 4588 in file ../src/genc.c).
make: *** [libaxiom.a(basecliq.o)] Error 1

Why would someone want libaxiom.a in the first place?

Well, that is needed to produce stand alone programs. And I was and I am 
still not sure how library code is executed internally inside Axiom. Is 
it always interpreted LISP or does AXIOMsys contain object code from the 
 ALGEBRA libraries? (More below.)

...

> You mean stand alone Aldor programs that use Axiom domains?

Right.

> This is possible now in Aldor isn't it?  I haven't tried to
> do this, but there is a small section in the Aldor users guide
> about it.

Oh, which one? I've just searched through the .pdf and there does not 
appear the string

  "axiom"

(including quotes) and found nothing. :-(

But, hey, what is this?
http://www.aldor.org/docs/HTML/chap18.html#12

That simply means that the online version and the up-to-date 
cvs-repository-version of the userguide at aldor.org are not identical. 
The online version is older.

Under 18.13 it says...
  At present, code from the AXIOM library will only run inside an
  AXIOM  workspace. Consequentely Aldor programs extending the
  AXIOM library will only run inside AXIOM. For a later version,
  it is planned that the entire AXIOM library will be translated
  into Aldor, at which point it should be possible to call it
  from C, Fortran, or any other environment in which Aldor operates.

The reason is that currently there is no libaxiom.a. And I have no idea 
whether the .o files under mnt/linux/algebra/ can simply be packaged to 
libaxiom.a. I rather believe that they cannot, since the names differ 
from the names in libaxiom.al.

It means currently, there is no way to get a standalone program that 
uses libaxiom. So there is also no way to call Axiom routines from a C 
program. (That is possible if you just take libaldor and libalgebra.)

> > and I also guess, that they are a bit faster, since all this
> > is binary format and not interpreted by the LISP machine.
>
> I had assumed that when compiled in stand alone mode, the
> necessary Axom domains would be compiled to binary if
> specified by the compiler options.
>
> > But maybe, you tell me that inside Axiom the domains living
> > in libaxiom are executed in a binary fashion anyway and only
> > the domains etc. that are entered into the interpreter are
> > interpreted.
>
> Even these are compiled and run as machine code if you use
>
> )set functions compile on
>
> however these are not saved permanently.

I have to trust you that this would be machine code and not some 
intermediate code which is approximately at the level of Java byte code.

> Axiom is quite different from Maple because it is based on
> Lisp. Whether the Lisp code is interpreted or compiled is
> (mostly) transparent to Axiom. In Axiom, both SPAD and Aldor
> generate Lisp code that is then (usually) compiled to machine
> object code and effectively become a dynamic extension of Axiom
> itself. There is no "Axiom kernel" as such except of course
> there is a lot of special purpose code that implements the
> Axiom interpreter and the SPAD compiler all of which is written
> in Lisp (plus Boot, which is a kind of high-level Lisp).
> Very few mathematical primitives are "built-in" at the Lisp/Boot
> level - almost all of the mathematics is compiled from SPAD
> code.

OK, I should have been more precise. By "Axiom kernel", of course, I 
meant the current Algebra stuff (basically everything that goes into 
libaxiom.al). I know that Axiom actually has NO kernel in the sense of 
Maple (except a very tiny little bit of built-in knowledge). But what 
you are saying sounds like: If I call a function from a domain that 
currently exists in Axiom (say, UnivariatePolynomial Integer), then that 
code is executed in object form (.o). So the .o files of all the Axiom 
domains should already be compiled into AXIOMsys, right?

Now if I compile a new function via

)set functions compile on

next time I use that function, the spad compiler and gcl should be 
invoked to get a (temporary) .o file. And more, that file should be 
loaded dynamically. Is that really the case?

Ralf