Did you know ... Search Documentation:
Deal with modules

Modules provide encapsulation, which is particularly useful for Prolog because the language often needs helper predicates. This implies that many predicates are there just to support the implementation of some more generally useful predicate.

In most modern languages, importing a module adds the name of the module to the current name space. Subsequently we access symbols in the module using <module><sep><symbol>, where <sep> is e.g., `. or ::`, etc. To avoid having to type the long module name over and over it is often possible to rename the module during the import, so we get e.g. Python

import long_module as m
m.my_function()

The SWI-Prolog module system works differently. This applies to all Prolog systems who's module system is based on Quintus Prolog's module system designed in the late 80s, the majority of today's popular Prolog systems. In these Prolog systems, importing a module adds the predicates of the module to the current name space. So, after importing the long_module, you can access its my_predicate/1 as

:- use_module(long_module).
t :- my_predicate(X).

The disadvantage of this is that it can lead to name conflicts if two modules export the same predicate and it is not immediately clear from which module a predicate is imported. To deal with this, exported predicates should have a fairly long descriptive name, often with a prefix that is an abbreviation of the module name, e.g., rb_insert/4 from library(rbtrees). As the name space for modules is flat, the name of a module as it appears in the :- module(Name,Exports) directive should also be descriptive. For example, avoid util as it will stop your code loading together with another package that provides a util module.

Importing from (library) modules

SWI-Prolog defines 5 ways to access predicates from a (library) module

  1. auto loading. This is defined for most of the standard library and allows calling predicates from the library without any declaration. You may define a predicate with the same name, which overrules the library predicate.
  2. :- autoload(File). is a more controlled version of general autoloading. It does not load File, but registers all exported predicates as first candidate for auto loading.
  3. :- autoload(File, Imports). As autoload/1, the file is not immediately loaded. It is used to lazily bind all predicates that appear in Imports. It is not allowed to define predicates from Imports locally.
  4. :- use_module(File). Loads File immediately, making all predicates available to be lazily activated. I.e., it is possible to define an exported predicate locally.
  5. :- use_module(File, Imports). Load Fileimmediately and imports the predicates in Imports. Any other attempt to define these predicates leads to an error.

Typically, (1) is practical for interactive use and prototyping. (2) and (4) are fairly controlled and good during development. (3) and (5) are good for the maintenance phase of a project where changes are relatively infrequent. (2) and (3) are SWI-Prolog specific and may be used to reduce the startup time of large applications that are executed from source.

Analysis of dependencies is provided by gxref/0, the PceEmacs command Control-C Control-D (Dependencies) or the GNU Emacs Sweep mode command Control-C Control-U (Undefined). The latter two should be synchronized.

Dealing with predicate name conflicts

Should you still have two modules exporting the same predicate that you wish to use in a single module, you have several options. The classical way is to use use_module/2, not importing the conflicting predicate. So, if m1 and m2 export p/1, use

:- use_module(m1, [q/1, r/1]). % all you need __except__ p/1
:- use_module(m2, [x/2]).

t :- m1:p(X), m2:p(X).

The alternative is to import with a different name, e.g.

:- use_module(m1, [p/1 as m1_p]).
:- use_module(m2, [p/1 as m2_p]).

t :- m1_p(X), m2_p(X).