Prolog integration and migration

This section provides suggestions for integrating and migrating plain Prolog code and Prolog module code to Logtalk. Detailed instructions are provided for encapsulating plain Prolog code in objects, converting Prolog modules into objects, and compiling and reusing Prolog modules as objects from inside Logtalk. An interesting application of the techniques described in this section is a solution for running a Prolog application that uses modules on a Prolog compiler with no module system. The wrapper tool can be used to help in migrating Prolog code.

Source files with both Prolog code and Logtalk code

Logtalk source files may contain plain Prolog code intermixed with Logtalk code. The Logtalk compiler simply copies the plain Prolog code as-is to the generated Prolog file. With Prolog modules, it is assumed that the module code starts with a module/1-2 directive and ends at the end of the file. There is no module ending directive that would allow us to define more than one module per file. In fact, most, if not all, Prolog module systems always define a single module per file. Some of them mandate that the module/1-2 directive be the first term in a source file. As such, when the Logtalk compiler finds a module/1-2 directive, it assumes that all code that follows until the end of the file belongs to the module.

Encapsulating plain Prolog code in objects

Most applications consist of several plain Prolog source files, each one defining a few top-level predicates and auxiliary predicates that are not meant to be directly called by the user. Encapsulating plain Prolog code in objects allows us to make clear the different roles of each predicate, to hide implementation details, to prevent auxiliary predicates from being called outside the object, and to take advantage of Logtalk advanced code encapsulating and reusing features. It also simplifies using its developer tools.

Encapsulating Prolog code using Logtalk objects is simple. First, for each source file, add an opening object directive, object/1-5, to the beginning of the file and an ending object directive, end_object/0, to the end of the file. Choose an object name that reflects the purpose of the source file code (this is a good opportunity for code refactoring if necessary). Second, add public/1 predicate directives for the top-level predicates that are used directly by the user or called from other source files. Third, we need to be able to call from inside an object predicates defined in other source files/objects. The easiest solution, which has the advantage of not requiring any changes to the predicate definitions, is to use the uses/2 directive. If your Prolog compiler supports cross-referencing tools, you may use them to help you make sure that all calls to predicates on other source files/objects are listed in the uses/2 directives. The Logtalk wrapper tool can also help in detecting cross-predicate calls. Compiling the resulting objects with the Logtalk unknown_predicates and portability flags set to warning will help you identify calls to predicates defined in other converted source files and possible portability issues.

Prolog multifile predicates

Prolog multifile predicates are used when clauses for the same predicate are spread among several source files. When encapsulating plain Prolog code that uses multifile predicates, it’s often the case that the clauses of the multifile predicates get spread between different objects and categories, but conversion is straight-forward. In the Logtalk object (or category) holding the multifile predicate primary declaration, add a predicate scope directive and a multifile/1 directive. In all other objects (or categories) defining clauses for the multifile predicate, add a multifile/1 directive and predicate clauses using the format:

:- multifile(Entity::Name/Arity).

Entity::Functor(...) :-
    ...

See the section on the multifile/1 predicate directive for more information. An alternative solution is to simply keep the clauses for the multifile predicates as plain Prolog code and define, if necessary, a parametric object to encapsulate all predicates working with the multifile predicate clauses. For example, assume the following multifile/1 directive:

% city(Name, District, Population, Neighbors)
:- multifile(city/4).

We can define a parametric object with city/4 as its identifier:

:- object(city(_Name, _District, _Population, _Neighbors)).

    % predicates for working with city/4 clauses

:- end_object.

This solution is preferred when the multifile predicates are used to represent large tables of data. See the section on parametric objects for more details.

Converting Prolog modules into objects

Converting Prolog modules into objects may allow an application to run on a wider range of Prolog compilers, overcoming portability problems. Some Prolog compilers don’t support a module system. Among those Prolog compilers that support a module system, the lack of standardization leads to several issues, notably with semantics, operators, and meta-predicates. In addition, the conversion allows you to take advantage of Logtalk more powerful abstraction and reuse mechanisms, such as separation between interface and implementation, inheritance, parametric objects, and categories. It also allows you to take full advantage of Logtalk developer tools for improved productivity.

Converting a Prolog module into an object is simplified when the directives used in the module are supported by Logtalk (see the listing in the next section). Assuming that this is the case, apply the following steps:

1. Convert the module module/1 directive into an object/1 opening object directive, using the module name as the object name. For module/2 directives apply the same conversion and convert the list of exported predicates into public/1 predicate directives. Add a closing object directive, end_object/0, at the end of the source code.

2. Convert any export/1 directives into public/1 predicate directives.

3. Convert any use_module/1 directives for modules that will not be converted to objects into use_module/2 directives (see next section), replacing the file spec in the first argument with the module name.

4. Convert any use_module/1-2 directives referencing other modules also being converted to objects into Logtalk uses/2 directives.

5. Convert each reexport/1 directive into a uses/2 directive and public/1 predicate directives (see next section).

6. Convert any meta_predicate/1 directives into Logtalk meta_predicate/1 directives by replacing the module meta-argument indicator, :, with the Logtalk meta-argument indicator 0 for goal meta-arguments. For closure meta-arguments, use an integer denoting the number of additional arguments that will be appended to construct a goal. Arguments that are not meta-arguments are represented by the * character. Do not use argument mode indicators such as ?, or +, or - as Logtalk supports mode directives.

7. Convert any explicit qualified calls to module predicates to messages by replacing the (:)/2 operator with the (::)/2 message-sending operator when the referenced modules are also being converted into objects. Calls in the pseudo-module user can be encapsulated using the {}/1 Logtalk external call control construct. You can also use instead a uses/2 directive where the first argument would be the atom user and the second argument a list of all external predicates. This alternative has the advantages of not requiring changes to the code making the predicate calls and of better visibility for the documenting and diagramming tools.

8. If your module uses the database built-in predicates to implement module-local mutable state using dynamic predicates, add both private/1 and dynamic/1 directives for each dynamic predicate.

9. If your module declares or defines clauses for multifile module predicates, replace the (:)/2 functor by (::)/2 in the multifile/1 directives and in the clause heads for all modules defining the multifile predicates that are also being converted into objects; if that is not the case, just keep the multifile/1 directives and the clause heads as-is.

10. Compile the resulting objects with the Logtalk unknown_predicates, and portability flags set to warning to help you locate possible issues and calls to proprietary Prolog built-in predicates and to predicates defined on other converted modules. In order to improve code portability, check the Logtalk library for possible alternatives to the use of proprietary Prolog built-in predicates.

Before converting your modules to objects, you may try to compile them first as objects (using the logtalk_compile/1 Logtalk built-in predicates) to help identify any issues that must be dealt with when doing the conversion to objects. Note that Logtalk supports compiling Prolog files as Logtalk source code without requiring changes to the file name extensions.

Compiling Prolog modules as objects

A possible alternative to porting Prolog code to Logtalk is to compile the Prolog source files using the logtalk_load/1-2 and logtalk_compile/1-2 predicates. The Logtalk compiler provides partial support for compiling Prolog modules as Logtalk objects. This support may allow using modules from a backend Prolog system in a different backend Prolog system, although its main purpose is to help in porting existing Prolog code to Logtalk in order to benefit from its extended language features and its developer tools. Why partial support? Although there is an ISO Prolog standard for modules, it is (rightfully) ignored by most implementers and vendors (due to its flaws and deviation from common practice). In addition, there is no de facto standard for module systems, despite otherwise frequent misleading claims. Key system differences include the set of implemented module directives, the directive semantics, the handling of operators, the locality of flags, and the integration of term-expansion mechanisms (when provided). Another potential issue is that, when compiling modules as objects, Logtalk assumes that any referenced module (e.g., using use_module/1-2 directives) is also being compiled as an object. If that’s not the case, the compiled module calls being compiled as message-sending goals will still work for normal predicates but will not work for meta-predicates called using implicit module qualification. The reason is that, unlike in Logtalk, calls to implicitly and explicitly qualified module meta-predicates have different semantics. Follows a discussion of other limitations of this approach that you should be aware of.

Supported module directives

Currently, Logtalk supports the following module directives:

module/1

The module name becomes the object name.

module/2

The module name becomes the object name. The exported predicates become public object predicates. The exported grammar rule non-terminals become public grammar rule non-terminals. The exported operators become public object operators but are not active elsewhere when loading the code.

use_module/2

This directive is compiled as a Logtalk uses/2 directive in order to ensure correct compilation of the module predicate clauses. The first argument of this directive must be the module name (an atom), not a module file specification (the adapter files attempt to use the Prolog dialect level term-expansion mechanism to find the module name from the module file specification). Note that the module is not automatically loaded by Logtalk (as it would be when compiling the directive using Prolog instead of Logtalk; the programmer may also want the specified module to be compiled as an object). The second argument must be a predicate indicator (Name/Arity), a grammar rule non-terminal indicator (Name//Arity), a operator declaration, or a list of predicate indicators, grammar rule non-terminal indicators, and operator declarations. Predicate aliases can be declared using the notation Name/Arity as Alias/Arity or, in alternative, the notation Name/Arity:Alias/Arity. Similar for non-terminal aliases.

export/1

Exported predicates are compiled as public object predicates. The argument must be a predicate indicator (Name/Arity), a grammar rule non-terminal indicator (Name//Arity), an operator declaration, or a list of predicate indicators, grammar rule non-terminal indicators, and operator declarations.

reexport/2

Reexported predicates are compiled as public object predicates. The first argument is the module name. The second argument must be a predicate indicator (Name/Arity), a grammar rule non-terminal indicator (Name//Arity), an operator declaration, or a list of predicate indicators, grammar rule non-terminal indicators, and operator declarations. Predicate aliases can be declared using the notation Name/Arity as Alias/Arity or, in alternative, the notation Name/Arity:Alias/Arity. Similar for non-terminal aliases.

meta_predicate/1

Module meta-predicates become object meta-predicates. All meta-predicates must be declared using the meta_predicate/1 directive using Logtalk syntax for normal arguments and meta-arguments. Note that Prolog module meta-predicates and Logtalk meta-predicates don’t share the same explicit-qualification calling semantics: in Logtalk, meta-arguments are always called in the context of the sender. Moreover, Logtalk is not based on the predicate-prefixing mechanism common to most Prolog module systems.

A common issue when compiling modules as objects is the use of the atoms dynamic, discontiguous, and multifile as operators in directives. For better portability, avoid this usage. For example, write:

:- dynamic([foo/1, bar/2]).

instead of:

:- dynamic foo/1, bar/2.

Another common issue is missing meta_predicate/1, dynamic/1, discontiguous/1, and multifile/1 predicate directives. The Logtalk compiler supports detection of missing directives (by setting its missing_directives flag to warning).

When compiling modules as objects, you probably don’t need event support turned on. You may use the events compiler flag to deny in the Logtalk compiling and loading built-in methods for a small performance gain for the compiled code.

Unsupported module directives

The reexport/1 and use_module/1 directives are not directly supported by the Logtalk compiler. But most Prolog adapter files provide support for compiling these directives using Logtalk’s first stage of its term-expansion mechanism. Nevertheless, these directives can be converted, respectively, into a sequence of :- use_module/2 and export/1 directives and use_module/2 directives by finding which predicates exported by the specified modules are reexported or imported into the module containing the directive. For use_module/1 directives, finding the names of the imported predicates that are actually used is easy. First, comment out the directive and compile the file (making sure that the unknown_predicates compiler flag is set to warning). Logtalk will print a warning with a list of predicates that are called but never defined. Second, use this list to replace the use_module/1 directives by use_module/2 directives. You should then be able to compile the modified Prolog module as an object.

Modules using a term-expansion mechanism

Although Logtalk supports term and goal expansion mechanisms, the usage semantics are different from similar mechanisms found in some Prolog compilers. In particular, Logtalk does not support defining term and goal expansions clauses in a source file for expanding the source file itself. Logtalk forces a clean separation between expansion clauses and the source files that will be subject to source-to-source expansions by using hook objects. But hook objects also provide a working solution here when the expansion code is separated from the code to be expanded. Logtalk supports using a module as a hook object as long as its name doesn’t coincide with the name of an object and that the module uses term_expansion/2 and goal_expansion/2 predicates. Assuming that’s the case, before attempting to compile the modules as objects, set the default hook object to the module containing the expansion code. For example, if the expansions are stored in a system module:

| ?- set_logtalk_flag(hook, system).
...

This, however, may not be enough, as expansions may be stored in multiple modules. A common example is to use a module named prolog for system expansions and to store the user-defined expansions in user. The Logtalk library provides a solution for these scenarios. Using the hook_flows library we can select multiple hook objects or hook modules. For example, assuming expansions stored on both user and system modules:

| ?- logtalk_load(hook_flows(loader)).
...

| ?- set_logtalk_flag(hook, hook_set([user, system])).
...

After these queries, we can try to compile the modules and look for other porting or portability issues. A well-know issue is Prolog module term-expansions calling predicates such as prolog_load_context/2, which will always fail when it’s the Logtalk compiler instead of the Prolog compiler loading a source file. In some of these cases, it may be possible to rewrite the expansion rules to use the logtalk_load_context/2 predicate instead.

File search paths

Some Prolog systems provide a mechanism for defining file search paths (this mechanism works differently from Logtalk own support for defining library path aliases). When porting Prolog code that defines file search paths, e.g. for finding module libraries, it often helps to load the pristine Prolog application before attempting to compile its source files as Logtalk source files. Depending on the Prolog backend, this may allow the file search paths to be used when compiling modules as objects that use file directives such as use_module/2.

Dealing with proprietary Prolog directives and predicates

Most Prolog compilers define proprietary, non-standard directives and predicates that may be used in both plain code and module code. Non-standard Prolog built-in predicates are usually not problematic, as Logtalk is usually able to identify and compile them correctly (but see the notes on built-in meta-predicates for possible caveats). However, Logtalk will generate compilation errors on source files containing proprietary directives unless you first specify how the directives should be handled. Several actions are possible on a per-directive basis: ignoring the directive (i.e., do not copy the directive, although a goal can be proved as a consequence), rewriting and copying the directive to the generated Prolog files, or rewriting and recompiling the resulting directive. To specify these actions, the adapter files contain clauses for the internal '$lgt_prolog_term_expansion'/2 predicate. For example, assume that a given Prolog compiler defines a comment/2 directive for predicates using the format:

:- comment(foo/2, "Brief description of the predicate").

We can rewrite this predicate into a Logtalk info/2 directive by defining a suitable clause for the '$lgt_prolog_term_expansion'/2 predicate:

'$lgt_prolog_term_expansion'(
        (:- comment(F/A, String)),
        (:- info(F/A, [comment is Atom]))
) :-
    atom_codes(Atom, String).

This Logtalk feature can be used to allow compilation of legacy Prolog code without the need of changing the sources. When used, it is advisable to set the portability compiler flag to warning in order to more easily identify source files that are likely non-portable across Prolog compilers.

A second example, where a proprietary Prolog directive is discarded after triggering a side effect:

'$lgt_prolog_term_expansion'(
        (:- load_foreign_files(Files,Libs,InitRoutine)),
        []
) :-
    load_foreign_files(Files,Libs,InitRoutine).

In this case, although the directive is not copied to the generated Prolog file, the foreign library files are loaded as a side effect of the Logtalk compiler calling the '$lgt_prolog_term_expansion'/2 hook predicate.

Calling Prolog module predicates

Prolog module predicates can be called from within objects or categories by simply using explicit module qualification, i.e. by writing Module:Goal or Goal@Module (depending on the module system). Logtalk also supports the use of use_module/2 directives in objects and categories (with the restriction that the first argument of the directive must be the actual module name and not the module file name or the module file path). In this case, these directives are parsed in a similar way to Logtalk uses/2 directives, with calls to the specified module predicates being automatically translated to Module:Goal calls.

As a general rule, the Prolog modules should be loaded (e.g., in the auxiliary Logtalk loader files) before compiling objects that make use of module predicates. Moreover, the Logtalk compiler does not generate code for the automatic loading of modules referenced in use_module/1-2 directives. This is a consequence of the lack of standardization of these directives, whose first argument can be a module name, a straight file name, or a file name using some kind of library notation, depending on the backend Prolog compiler. Worse, modules are sometimes defined in files with names different from the module names, requiring finding, opening, and reading the file in order to find the actual module name.

Logtalk allows you to send a message to a module in order to call one of its predicates. This is usually not advised as it implies a performance penalty when compared to just using the Module:Call notation. Moreover, this works only if there is no object with the same name as the module you are targeting. This feature is necessary, however, in order to properly support the compilation of modules containing use_module/2 directives as objects. If the modules specified in the use_module/2 directives are not compiled as objects but are instead loaded as-is by Prolog, the exported predicates would need to be called using the Module:Call notation but the converted module will be calling them through message-sending. Thus, this feature ensures that, on a module compiled as an object, any predicate calling other module predicates will work as expected, either these other modules are loaded as-is or also compiled as objects.

For more details, see the Calling Prolog predicates section.

Loading converted Prolog applications

Logtalk strongly favors and advises users to provide a main loader file for applications that explicitly load any required libraries and the application source files. In contrast, Prolog applications often either scatter loading of source files from multiple files or use implicit loading of source files via use_module/1-2 directives. Due to this frequent ad-hoc approach, it’s common to find Prolog applications with duplicated loading directives, and where loading order ignores the dependencies between source files. These issues are easily exposed by the Logtalk linter when compiling Prolog files as Logtalk files. Also common are Prolog files with multiple circular dependencies. While this should not affect the semantics of the ported code, it may cause some performance penalties as it prevents the Logtalk compiler from optimizing the message sending goals using static-binding. It also makes the application architecture more difficult to understand. The definition of explicit loader files provides a good opportunity for sorting out loading order and circular dependencies, with the linter warnings providing hints for possible code refactoring to eliminate these issues. The diagrams tool supports directory and file loading and dependency diagrams that are also useful in understanding applications architecture.