Re: [Yap-users] Communicating multiple-precision integers to foreign code and back
Vitor Santos Costa wrote:
I suppose such an interface is really in order. I haven't done much because (1) no one asked before, and (2) it will make the interface depend on GMP, which is not always available. But (1) is broken now, and (2) should not be strong enough.
I think that the interface should not depend on GMP: the ability to communicate Prolog integers back and forth should be present anyway and not depend on the size of those Prolog integers. In other words, I believe the presence of GMP should only influence the availability of unbounded integers, not the way integers of any size are communicated (on YAP systems compiled without the GMP, trying to convert an externally-provided integer that is too large will simply cause the C function implementing the conversion to return an error code).
Which brings me to another question: what is the best way for foreign code to know whether it is interfaced to a YAP system with bounded or unbounded integers? And to obtain the minimum and maximum integers? Is there a sensible alternative to calling current_prolog_flag/2? For instance, for SWI-Prolog you can write the following C code:
Prolog_min_integer = PL_query(PL_QUERY_MIN_INTEGER); Prolog_max_integer = PL_query(PL_QUERY_MAX_INTEGER);
whereas this is valid C for interfacing with GNU Prolog
Prolog_min_integer = INT_LOWEST_VALUE; Prolog_max_integer = INT_GREATEST_VALUE;
That's a good point, actually. Do you know how other systems cope with that?
SICStus Prolog offers SP_put_integer_bytes() and SP_get_integer_bytes(), SP_put_number_chars() and SP_get_number_chars(). I suppose the former are more efficient than the latter.
Ciao Prolog has ciao_put_number_chars() and ciao_get_number_chars(). Cheers,
Roberto
Sorry for the previous message, I actually grepped for a message from you and Christian I did not answered at the time and found the other message, which I suppose I never did reply :-(.
I think that the interface should not depend on GMP: the ability to communicate Prolog integers back and forth should be present anyway and not depend on the size of those Prolog integers. In other words, I believe the presence of GMP should only influence the availability of unbounded integers, not the way integers of any size are communicated (on YAP systems compiled without the GMP, trying to convert an externally-provided integer that is too large will simply cause the C function implementing the conversion to return an error code).
Yes, YAP_WriteBuffer and YAP_ReadBuffer should do what you want. They are a bit heavier than what one would like, but I suspect the actual overhead of generating a text representation for a bignum is more than the overhead of parsing.
Which brings me to another question: what is the best way for foreign code to know whether it is interfaced to a YAP system with bounded or unbounded integers? And to obtain the minimum and maximum integers? Is there a sensible alternative to calling current_prolog_flag/2? For instance, for SWI-Prolog you can write the following C code:
Prolog_min_integer = PL_query(PL_QUERY_MIN_INTEGER); Prolog_max_integer = PL_query(PL_QUERY_MAX_INTEGER);
whereas this is valid C for interfacing with GNU Prolog
Prolog_min_integer = INT_LOWEST_VALUE; Prolog_max_integer = INT_GREATEST_VALUE;
No, Yap should have something like PL_query, that's quite easy to do and other people have asked for it, I'll see if I can have today (you can always call Prolog to call call current_prolog_flag, but it is kind of overkill).
SICStus Prolog offers SP_put_integer_bytes() and SP_get_integer_bytes(), SP_put_number_chars() and SP_get_number_chars(). I suppose the former are more efficient than the latter.
Ciao Prolog has ciao_put_number_chars() and ciao_get_number_chars(). Cheers,
I suppose I could have a specialised routine eventually...
Cheers,
Vitor
participants (2)
-
Roberto Bagnara -
Vitor Santos Costa