axiom-developer
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Renaming at inheritance (was: Re: [Axiom-developer] Curiosities with


From: Ralf Hemmecke
Subject: Re: Renaming at inheritance (was: Re: [Axiom-developer] Curiosities with Axiom mathematical structures)
Date: Thu, 02 Mar 2006 13:38:01 +0100
User-agent: Thunderbird 1.5 (X11/20051201)

Maybe, I am a bit behind ...

On 03/01/2006 05:26 AM, Bill Page wrote:
On February 27, 2006 11:08 AM Gabriel Dos Reis wrote:

...
Ralf Hemmecke <address@hidden> writes: ... | One reason for having Monoid and AbelianMoniod (one with "*" and
| the other with "+") is that neither SPAD nor Aldor allows to
| rename functions during inheritance.

yes, that is true for many (all?) object-oriented languages.  I can
understand the language limitation, but that is hardly an excuse for
chosing in advance that "+" is for Abelian monoid and "*" for
non-Abelian ones.


I don't see why renaming is necessary.
| Note that one would have to say something like
| | define Monoid(
|    *: (%, %) -> %;
|    1: %
| ): Category == ...
| | define Group(
|    *: (%, %) -> %,
|    inv: % -> %,
|    1: %
| ): Category == ...
| | define Ring(.....): Category == with {
|    Monoid(*, 1);
|    Group(+, -, 0);
| }

Ring should be both Monoid and Group, whether they shoud be categories, I don't really know.

Well, think of Axiom's Category as just a list of function (and constant) signatures.

Well, as far as I know, something that is of type Category in Aldor/Axiom should be near to a mathematical category. (But I am not the original designer and unfortunately there is no page on the AxiomWiki that explains it clearly.

Anyway, what I suggested above is perhaps not so close to mathematics as I and probably other mathematicians would like it to be.

The point is that one speaks of a "category of rings, groups, monoids, etc.), but not of a "category of rings(*, 1, +, -, 0)". So defining those structures in Axiom without arguments would be highly preferable.

However, there is no way in Axiom to express that a Ring is a Monoid (SemiGroup if you like) "with respect to multiplication" and a Group (which is also a Monoid) "with respect to addition".

So the compromise in Axiom currently is to have Monoid and AbelianMonoid where AbelianMonoid does not inherit from Monoid and one is with respect to "*" and the other with respect to "+". However, being a mathematician, it is not so nice that just because of language limitations I have to construct another algebra hierarchy that doesn't match the mathematical one.

About two years ago I have spoken to Stephen Watt about a possible way to rename functions, but he said that would be a major change to the compiler and very hard to implement. :-(

In Axiom (SPAD) we can write:

Monoid(m:Symbol,u:Symbol): Category == with
       m: (%,%) -> %       ++ returns the product of x and y
       u: () -> %          ++ unit
       associative(m)      ++ m(a,m(b,c)) = m(m(a,b),c)
       identity(u)         ++ m(a,u) = m(u,a) = a

As you see below, that will not work in Aldor.

--begin mon.as
#include "aldor"
define Monoid(m:Symbol,u:Symbol): Category == with {
        m: (%,%) -> %;
        u: () -> %;
        associative(m);
        identity(u);
}
--end mon.as

aldor mon.as
"mon.as", line 5:         associative(m);
                  ........^
[L5 C9] #1 (Error) There are no suitable meanings for the operator `associative'.

"mon.as", line 6:         identity(u);
                  ........^
[L6 C9] #2 (Error) There are no suitable meanings for the operator `identity'.

It simply does not match the Aldor language specification.
"associative(m)" is interpreted as call to a function "associative" which simply does not exist.

And now, something interesting. The following file compiles.
--begin mon1.as
#include "aldor"
define Monoid(m:Symbol,u:Symbol): Category == with {
        m: (%,%) -> %;
        u: () -> %;
}
--end mon1.as

My first guess was that the compiler would complain, because m is a parameter of type Symbol and now we turn it into a function. But I was wrong. There are simply TWO m there. One of type Symbol and the other of type "(%, %) -> %" and the have NOTHING to do with each other. They just happen to have the same name.

That seems to be a difference between SPAD and Aldor.

Then we can write:

  )sh Ring(+,-,"0"::Symbol,*,"1"::Symbol)
  Ring(+,-,"0"::Symbol,*,"1"::Symbol) has commutative(+)

See the example at http://wiki.axiom-developer.org/SandBoxMonoid

That looks interesting, but as I said above, I would not like Ring to carry parameters. With parameters, it looks a bit more like a universal algebra.

Anyway, I think in the above example Axiom's "Symbol" domain (written *in* the SPAD language) is treated as part of the SPAD language. I cannot say that I like that. I prefer Aldor's approach even if it (still) forbids me to do "renaming during inheritance". But I can understand the code from the language definition.

See also my additions on
  http://wiki.axiom-developer.org/SandBoxMonoid

| If such renaming could be implemented into the Aldor language
| that would make it even better suited for mathematics. However, | then there is need that someone implements these ideas into the
| compiler. Who?

Could you explain what you mean by "renaming" and how it might
be used to express mathematics?

Well, clearly a mathematian would always say that an abelian monoid is a monoid. These properties are properties of the structure and not of the names of the functions that form that structure. We cannot express this in Aldor in general, and it is also not done in Axiom.

Anyway, it is a pity that "Axiom" cannot express axioms in its language.
By the category "AbelianMonoid" one has just a tag that domains that implement this category have to implement "+: (%,%)->%" in a commutative fashion. The only way to state commutativity is in the documentation, but not formally in a language construct.

There are some experiments, however...
http://citeseer.ist.psu.edu/poll98adding.html
http://www.cs.kent.ac.uk/people/staff/sjt/Atypical/AldorExs/index.html

> Not all, Eiffel can rename methods when inheriting:
>
> http://www.maths.tcd.ie/~odunlain/eiffel/intro.html
> """
>    class C inherit
>
>         A rename x as x1, y as y1 end;
>
>         B rename x as x2, y as y2 end
>
>   feature...
>
> Here, if both A and B have features named x and y, class C would be
> invalid without the renaming.

I don't know Eiffel, but I guess that feature is necessary for multiple inheritance.

In Aldor multiple inheritance is already possible for categories, but forbidden for domains.

But I would like to say something like

define SemiGroup: Category == with {
  *: (%, %) -> % -- or any other name.
}
Ring: Category == with
  SemiGroup;
  SemiGroup where {* == +};
}

Oh that will be getting complicated if I inherit from something like

define Structure: Category == with {
  foo: % -> %;
  foo: (%, %) -> %;
}

Where are the language designers?

Ralf




reply via email to

[Prev in Thread] Current Thread [Next in Thread]