[m-users.] Performance of solutions module

[Zoltan Somogyi]

On Fri, 2 Nov 2018 15:04:15 -0400, Charlie McGee <c4cypher at gmail.com> wrote:

> >If you can construct a list of solutions directly, that will usually be
> faster
> >than using solutions/2, since the latter has nontrivial overhead. Whether
> >it will be *measurably* faster depends on what fraction of the program's
> >overall runtime this task takes, and on whether your direct method
> >for construction of solutions requires a post-pass to sort the solutions,
> >or to eliminate duplicates (two tasks that solutions/2 does for you,
> >and which form part of its overheads).
> 
> Building on Zoltan Somogyi's suggestion to construct your own list instead
> of relying on solutions/2, if you want to get an idea of how to do that,
> you can look on how the solutions library implements its calls by looking
> through solutions.m
> <https://github.com/Mercury-Language/mercury/blob/master/library/solutions.m>
> in the library folder of the Mercury source code (The Mercury Lang github
> repository is very useful for this).

Actually, I would advise against doing this. The code of solutions/2 has to do
something that cannot be done in pure Mercury: communicating information
between different branches of a disjunction. It therefore has to use impure code,
even though the interface it is implementing has a pure semantics. This works,
but I wouldn't recommend the code to nonexperts.

TL/DR: if you have no simple way to construct your list directly, then use
the Mercury library's solutions/2, or one of the other predicates in the solutions
module; do NOT try to roll your own.

Zoltan.