Re: [Axiom-developer] Unions in Spad

[Stephen Wilson]

"Bill Page" <address@hidden> writes:

> On 08 Jul 2007 15:21:52 -0400, Stephen Wilson wrote:
> >
> > Spad, internally, has two notions of Unions.  The first variety might
> > be termed `regular', the other is, not surprisingly, termed `new'.
> >
> > Regular Unions do not have tags associated with branches. For example:
> >
> >       Union(Integer, String)
> >
> > New Unions have tags, as in
> >
> >       Union(int: Integer, str: String)
> >
> > The shortcoming is how these two flavors interact.  Currently, Unions
> > are of one kind or the other.  But this fact is nowhere asserted, and
> > blending the two types leads to obscure errors.  For example:
> >
> >      Union(int: Integer, String)
> >
> > I would appreciate any input if the above `hybrid' Union is useful or
> > desirable.
> >
> 
> I think it is undesirable.

OK.  I personally agree.

> > If so, I have a strategy to make such Unions usable (but it may be a
> > while before I find the time to implement it). If not, then there is
> > an easy fix to assert the uniformity of Union branches.
> >
> 
> I think mixed union constructions should be consider syntax errors.

I personally agree as well.  Unfortunately, the current Spad parser
is, well, difficult to change.  I can check the uniformity of Union
branches at a post-parsing stage to at least get the assertion into
the system.  Perhaps in time my new parser could be used as a full
replacement for the current one.  

> Also it seems to me that the untagged union construction should be
> flagged by the SPAD compiler as deprecated so that the library code
> can eventually be updated to use the new tagged form that is
> compatible with Aldor.

There is a lot of code which uses the untagged union, so the
conversion would be a fair bit of work.  I could experiment with such
a change in the Axisp repo and solicit everyones help in changing the
Algebra.  In time, we might be able to get a changeset together which
could be applied to Silver.  Any thoughts on this?

> One might be tempted to argue that the untagged form is more readable
> but I think the complications associated with the semantics of things
> like:
> 
>    Union(Integer,Integer)
> 
> out weigh any linguistic advantage.

Absolutely, but this too could be flagged as an error by the compiler.

> 
> Regards,
> Bill Page.


Take care,
Steve