Re: simultaneous open queries

Hi Ed,

On 03/18/2014 02:48 PM, Edd Barrett wrote:
> Hi,
>
> I am currently trying to get by head around the restrictions relating to
> simultaneous queries in a single threaded program that uses the SWI Prolog C
> API.
>
> We are told in the docs [0] that "a foreign context can have at most one
> active query" and that "we are "not allowed to open multiple queries and
> start generating solutions for each".
>
> So it seems that we are limited (at least without threading?) to one
> simultaneous query per calling context. Now, a calling context is an
> instance of module_t; the type used to refer to a Prolog module.

First of all, the module argument defines the module context of the
call. That is, the module used to resolve unqualified arguments to
meta-predicates. It refers to the predicate @/2. For example, the code
below will assert `t' in the module 'data':

   predicate_t pred = PL_predicate("assertz", 1, "system");
   term_t t = PL_new_term_refs(1);
   module_t datam = PL_new_module(PL_new_atom("data"));

   <create term in t>
   PL_call_predicate(datam, PL_Q_NORMAL, pred, t);

This module context has nothing to do with single or multiple queries
being active.

> I am having difficulty understanding what exactly this means in the
> sense of a calling context. Does it mean that in order for simultaneous
> queries to be open, the "callers" must be defined in different modules?
> I can't seem to find any documentation describing the role of a calling
> context (other than it is a module).

So, no.

One thread can not have multiple open queries, period ... Well, almost.
You *can* call Prolog, Prolog calls back C and then the C can open
another query. I.e., you can have C/Prolog mutually recursive functions.

You CAN NOT do

   q1 = PL_open_query(...);
   q2 = PL_open_query(...);

   for(;;)
     if ( PL_next_solution(q1) && PL_next_solution(q2) )
       ...

This would be coroutining, which is not provided by SWI-Prolog (dunno
whether there are systems that can do this). What you can do is create
an additional *engine* using PL_create_engine(). Now, you can switch
engines using PL_set_engine() and from a single C thread you can create
queries in both engines and ask both engines for solutions. I never
tried it ...

> Asked differently, is it possible to have simultaneous queries open in a
> single threaded C application at all?

So, yes, but engines do use the threading library. In fact, the code
running in the second engine claims to run in a thread. Anyway,
SWI-Prolog can not be built without threading support since quite a few
versions.

So, an engine is a set of Prolog stacks and a state that is able to
run Prolog queries.  An engine is normally connected to a unique
(native) thread, but it can be disconnected from the thread and
re-connected to another thread.

Hope this is clear ...

	Cheers --- Jan

> Perhaps we can use the outcomes of this discussion to improve the
> documentation.
>
> [0] http://www.swi-prolog.org/pldoc/man?section=foreign-create-query

Feel free to update.  Preferably as git patches by editing man/foreign.doc
(no, these are not Word files; Word didn't exist when I started the docs.
They are LaTeX that is pre-processed a little).

	Cheers --- Jan