Re: [Axiom-developer] Re: noweb / Lisp Skills

[Ralf Hemmecke]

On 05/05/2006 11:59 AM, Martin Rubey wrote:
> "Page, Bill" <address@hidden> writes:
>
> > On Thursday, May 04, 2006 10:47 AM Martin Rubey wrote:
> > > ... 
> > > Things to do on the interpreter side needed for the 
> > > axiom-combinat project:
> > >
> > > * make it understand Aldor:
> > >
> > >   * dependent types
> > >   * extend
> > >   * creating domains on the command line
> > Is what you are suggesting the same thing as making library
> > programming possible at the level of the interpreter? 
>
> Yes. In aldor -gloop this is possible.

> In fact, maybe it would be better to try to integrate the type guessing
> mechanism of Axiom into the Aldor interpreter. That is, throw away the axiom
> interpreter and replace it by aldor -gloop.

Hmm, I am not sure whether I would currently suggest "aldor -gloop" as a
replacement for the Axiom interpreter. The idea that the interpreter and
compiler languages are identical is attractive, but

1) "aldor -gloop" is slow and needs something like BNatural to make Aldor
simpler for endusers. But it might be a good start in the long run.

2) The aldor interpreter seems a bit buggy to me.

> > If so, they I supose that you realize that this is already an issue at the
> > level of SPAD and is not really an Aldor specific problem. It seems to me
> > that the differences between the interpreter and library compiler
> > environments were largely built-in as part of the original design of Axiom. I
> > think the designers decided that the interpreter *should* behave differently
> > than the compiler and that some of the things the compiler does should
> > specifically not be available to the interactive user.

> If this was intentional, it was a mistake.

I also think that. The interpreter should actually be a shell around the 
compiler and the libraries. One could even have several such shell 
similar to csh, bash, etc. (The question is whether different 
shells/interpreters would be desirable.) In particular, if an 
interpreter allows to read something like .input files (that are 
interpreted like the command line) than one cannot expect to get easy 
translation of .input to .as since the interpreter might have added 
automatic type coercion. The extreme would be that only pure Aldor (.as) 
files are allowed. But though it might be tempting to force people to 
think about the types they actually want, it would probably be much to 
hard for beginners. One has to get used to all this type stuff.

> > But the the 'extend' functionality is another issue because that whould
> > require providing the equivalent of the library compiler at the interpreter
> > level.

> Yes, this might have been wrong on my side. It seems that extend does not even
> work in aldor -gloop. However, I think it should.

Currently the Axiom interpreter can call the compiler and thus extend 
(well, not the Aldor "extend") the functionality of Axiom per session.

What would people think about a (library) function "compile" that takes 
a string as input (or better something of type "AldorCode" and considers 
 this to be aldor code that has to be compiled. The result would be the 
compiled function that then can be used as any other function. So one 
would say something like

f := compile(Integer -> Integer, "(i:Integer): Integer -> i+1")
n: Integer := 2;
f(n);

Of course I could write this without the quotes, but imagine that the 
string has been dynamically produced by a program.

Ralf