Re: term_expansion/2 with DCG clauses

On 08/13/2013 05:19 PM, Paulo Moura wrote:
> Hi Jan,
>
> Sorry to somehow hijack the thread but...
>
> On 13/08/2013, at 15:12, Jan Wielemaker<[email protected]>  wrote:
>
>>> It is still not very clear what you want.  If you want to produce
>>> DCG's from term-expansion, this is just fine.  DCG expansion is
>>> done automatically if term-expansion produces (Head --> Body)
>>> terms.
> ... the above behavior come as a surprise to me. A simple example:
>
> $ swipl ... ?- assertz(term_expansion(a(X), (X --> []))). true.
>
> ?- expand_term(a(z), T). T = (z(_G2260, _G2260):-true).
>
> I would expect T = (z-->[]). GNU Prolog, XSB, YAP, Logtalk all
> produce T = (z-->[]). The Quintus Prolog 3.5 manual states (pp 1315)
> that:
>
> "expand_term/2 calls term_expansion/2 first; if it succeeds, the
> standard grammar rule expansion is not tried."
>
> Is SWI-Prolog the only one deviating from the original Quintus
> definition?

Looks like it. I've forgotten the history. SWI-Prolog uses a layered
translation model, first working in the current module, then feeding the
result to the rules in user, then to the rules in system and finally to
the DCG translator. Each step may produce a term or lists of terms. If
it produces a list, the terms are handed one-by-one to the next layer.

This seems a logical way to access DCG expansion. The only price is that
you cannot created rules for -->/2 (nor for []/0 and ./2). Do these
systems have an agreed way to creates DCGs from expansion rules? I guess
Richard will know the rationale for not doing DCG after user expansion.
I can't think of one right now.

	Cheers --- Jan

P.s.	Note that having the DCG translation at the end doesn't stop
	writing your own.