Re: Preserving query data in foreign interface

2013/8/11 Abramo Bagnara <[email protected]>

> Il 11/08/2013 13:41, Paul Singleton ha scritto:
> > On 11/08/2013 11:39, Abramo Bagnara wrote:
> >
> >> I'm calling from Java a Prolog query and I need to avoid that returned
> >> data is lost.
> >
> > The "high level" interface does that for you
> >
> >> I'm using low-level interface to avoid term_t <-> Term copying and the
> >> very deep stack recursion that this implies when handling long lists.
>

I don't know JPL... but the C++ interface has PlTail, a wrapper around the
non recursive list accessor implemented by PL_get_list, PL_get_nil.
Using it, copying a list is easy. For instance

        PlTerm Completions, Delete, word;
        if (PlCall("prolog", "complete_input", PlTermv(Before, After,
Delete, Completions)))
            for (PlTail l(Completions); l.next(word); )
                strings.append(t2w(word));

Maybe PlTail could be implemented in JPL ?


> >
> > Then you'll have to implement your own non-recursive term_t <-> Term
> > copying - please donate this to JPL :-)
>
> We have decided (successfully) to use only low level interface so to
> avoid this problema and to avoid creation of java objects (we produce
> very big prolog compound in java and we elaborate them in Prolog).
>
> >> If I use Prolog.close_query the binding is lost and we have no
> >> Prolog.cut_query available.
> >
> > Hmm, if your hypothetical Prolog.cut_query retains the binding, then we
> > can't reuse the engine until you explicitly close it; I don't think this
> > is the way to go...
>
> It is not hypothetical: C interface has it.
>
> >> That apart, I also have noted that in both java interface and C
> >> interface it is missing an helper to do deep copy of answers so to
> >> preserve multiple results of nondet queries.
> >
> > JPL has this; Query#allSolutions returns an array of maps from variable
> > names to copied bindings
> >
> >> I've missing something? There are other ways to do what above?
> >
> > Until term_t <-> Term copying is reimplemented non-recursively (or
> > otherwise ingeniously), the best workaround for copying long lists is to
> > use a Prolog-side wrapper for long-list-returning predicates which
> > converts them to trees (or atoms?), then convert them again Java-side.
>
> Rather ugly, isn't it?
>
> Currently, as fortunately the data obtained from the Prolog query is
> rather small, I've written a workaround that extract the data from
> results when query is still open in Java objects, then close the query
> and rebuild the Prolog result from Java objects.
>
> It works, but nevertheless it seems to me rather uncomfortable. I wonder
> if SWI Prolog C interface has the same troubles when trying to save
> returned data from queries that should be closed.
>
> As a stupid example (and forgetting for a moment JPL high level
> interface that gives support for that copying every Prolog term in java
> objects), suppose we want to build from C (or Java) a list of Prolog
> terms. One of them should be A as obtained from Prolog query `A is 1
> rdiv 2 * 3 rdiv 4'.
>
> Are there simpler (and cheaper when returned data is *big*) way than the
> one described above?
>
> --
> Abramo Bagnara
>
> BUGSENG srl - http://bugseng.com
> mailto:[email protected]
> _______________________________________________
> SWI-Prolog mailing list
> [email protected]
> https://lists.iai.uni-bonn.de/mailman/listinfo.cgi/swi-prolog
>
-------------- next part --------------
HTML attachment scrubbed and removed