Predicatesï
Predicate directives and clauses can be encapsulated inside objects and categories. Protocols can only contain predicate directives. From the point-of-view of a traditional imperative object-oriented language, predicates allows both object state and object behavior to be represented. Mutable object state can be represented using dynamic object predicates but should only be used when strictly necessary as it breaks declarative semantics.
Reserved predicate namesï
For practical and performance reasons, some predicate names have a fixed interpretation. These predicates are declared in the built-protocols. They are: goal_expansion/2 and term_expansion/2, declared in the expanding protocol; before/3 and after/3, declared in the monitoring protocol; and forward/1, declared in the forwarding protocol. By default, the compiler prints a warning when a definition for one of these predicates is found but the reference to the corresponding built-in protocol is missing.
Declaring predicatesï
Logtalk provides a clear distinction between declaring a predicate and defining a predicate and thus clear closed-world assumption semantics. Messages or calls for declared but undefined predicates fail. Messages or calls for unknown (i.e. non declared) predicates throw an error. Note that this is a fundamental requirement for supporting protocols: we must be able to declare a predicate without necessarily defining it.
All object (or category) predicates that we want to access from other objects (or categories) must be explicitly declared. A predicate declaration must contain, at least, a scope directive. Other directives may be used to document the predicate or to ensure proper compilation of the predicate clauses.
Scope directivesï
A predicate scope directive specifies from where the predicate can be
called, i.e. its visibility. Predicates can be public, protected,
private, or local. Public predicates can be called from any object.
Protected predicates can only be called from the container object or
from a container descendant. Private predicates can only be called from
the container object. Predicates are local when they are not declared in
a scope directive. Local predicates, like private predicates, can only be
called from the container object (or category) but they are invisible
to the reflection built-in methods (current_predicate/1
and predicate_property/2) and to the message error handling
mechanisms (i.e. sending a message corresponding to a local predicate
results in a predicate_declaration
existence error instead of a scope
error).
The scope declarations are made using the directives public/1, protected/1, and private/1. For example:
:- public(init/1).
:- protected(valid_init_option/1).
:- private(process_init_options/1).
If a predicate does not have a (local or inherited) scope declaration, it is assumed that the predicate is local. Note that we do not need to write scope declarations for all defined predicates. One exception is local dynamic predicates: declaring them as private predicates may allow the Logtalk compiler to generate optimized code for asserting and retracting clauses.
Note that a predicate scope directive doesnât specify where a predicate is, or can be, defined. For example, a private predicate can only be called from an object holding its scope directive. But it can be defined in descendant objects. A typical example is an object playing the role of a class defining a private (possibly dynamic) predicate for its descendant instances. Only the class can call (and possibly assert/retract clauses for) the predicate but its clauses can be found/defined in the instances themselves.
Scope directives may also be used to declare grammar rule non-terminals and operators. For example:
:- public(url//1).
:- public(op(800, fx, tag)).
Note that, in the case of operators, the operator definitions donât become
global when the entity containing the directives is compiled and loaded.
This prevents an application breaking when e.g. an updated third-party
library adds new operators. It also allows loading entities that provide
conflicting operator definitions. Here the usual programming idiom is to
copy the operator definitions to a uses/2
directive. For example, the
lgtunit tool makes available a (=~=)/2
predicate
(for approximate float equality) that is intended to be used as an infix
operator:
:- uses(lgtunit, [
op(700, xfx, =~=), (=~=)/2
]).
Thus, in practice, the solution to use library entity operators in client entities is the same for using library entity predicates with implicit message sending.
Mode directiveï
Often predicates can only be called using specific argument patterns. The valid arguments and instantiation modes of those arguments can be documented using the mode/2 directive. For example:
:- mode(member(?term, ?list), zero_or_more).
The first directive argument describes a valid calling mode. The minimum information will be the instantiation mode of each argument. The first four possible values are described e.g. in the ISO Prolog Core standard [ISO95]). The remaining two can also be found in use in some Prolog systems.
+
Argument must be instantiated (but not necessarily ground).
-
Argument should be a free (non-instantiated) variable. When bound, the call will unify the computed term with the given argument.
?
Argument can either be instantiated or free.
@
Argument will not be further instantiated (modified).
++
Argument must be ground.
--
Argument must be unbound. Used mainly when returning an opaque term (e.g. a stream handle).
Note that the +
and @
instantiation modes have the same meaning
for atomic arguments. E.g. you can write either +atom
or @atom
but the first alternative is preferred.
These six mode atoms are also declared as prefix operators by the
Logtalk compiler. This makes it possible to include type information
for each argument as in the example above. Some possible type
values are: event
, object
, category
, protocol
,
callable
, term
, nonvar
, var
, atomic
, atom
,
number
, integer
, float
, compound
, and list
. The
first four are Logtalk specific. The remaining are common Prolog types.
We can also use our own types that can be either atoms or ground
compound terms. See the types library documentation
for an extensive list of pre-defined types that cover most common use
cases.
The second directive argument documents the number of proofs, but not
necessarily distinct solutions, for the specified mode. As an example,
the member(X, [1,1,1,1])
goal have only one distinct solution but four
proofs for that solution. Note that different modes for the same predicate
often have different determinism. The possible values are:
zero
Predicate always fails (e.g. the
false/0
standard predicate).one
Predicate always succeeds once (e.g. the
flush_output/0
standard predicate).zero_or_one
Predicate either fails or succeeds (e.g. the
atom/1
standard predicate).zero_or_more
Predicate has zero or more proofs (e.g. the
current_predicate/1
standard predicate).one_or_more
Predicate has one or more proofs (e.g. the
repeat/0
standard predicate).zero_or_error
Predicate either fails or throws an error.
one_or_error
Predicate either succeeds once or throws an error (e.g. the
open/3
standard predicate).zero_or_one_or_error
Predicate succeeds once or fails or throws an error (e.g. the
get_char/1
standard predicate).zero_or_more_or_error
Predicate may fail or succeed multiple times or throw an error (e.g. the
retract/1
standard predicate).one_or_more_or_error
Predicate may succeed one or more times or throw an error.
error
Predicate will throw an error (e.g. the
type_error/2
built-in method).
The last six values support documenting that some call modes may throw an
error or will throw an error despite the call arguments complying with the
expected types and instantiation modes. As an example, consider the open/3
standard predicate:
:- mode(open(@source_sink, @io_mode, --stream), one_or_error).
In this case, the mode directive tells the user that a valid call can still throw an error (there may be e.g. a permission error opening the specified source or sink).
Notice that using the zero
, one
, zero_or_one
, zero_or_more
, or
one_or_more
modes is not only for predicates that never throw an exception;
they can also be used for any predicate that doesnât throw an exception when
the arguments are valid. For example, the current_predicate/1
standard
predicate throws an exception if the argument is neither a variable nor a
predicate indicator; but it succeeds zero or more times when its argument is
valid:
:- mode(current_predicate(?predicate_indicator), zero_or_more).
Some predicates have more than one valid mode, thus implying several mode
directives. For example, to document the possible use modes of the standard
atom_concat/3
predicate we would write:
:- mode(atom_concat(?atom, ?atom, +atom), one_or_more).
:- mode(atom_concat(+atom, +atom, -atom), one).
The first mode/2
directive specifies that the atom_concat/3
predicate
can be used to split an atom into a prefix and a suffix. The second mode/2
directive specifies that concatenating two atoms results in a new atom. There
are often several alternatives mode/2
directives that can be used to
specify a predicate. For example, an alternative to the second mode/2
directive above would be:
:- mode(atom_concat(+atom, +atom, ?atom), zero_or_one).
In this case, the same information is provided by both alternatives. But the first alternative is simpler and thus preferred.
Some old Prolog compilers supported some sort of mode directives to improve performance. To the best of my knowledge, there is no modern Prolog compiler supporting this kind of directive for that purpose. The current Logtalk version simply parses this directive for collecting its information for use in the reflection API (assuming the source_data flag is turned on). In any case, the use of mode directives is a good starting point for documenting your predicates.
Meta-predicate directiveï
Some predicates may have arguments that will be called as goals, interpreted as closures that will be used for constructing goals, or passing meta-arguments to calls to other meta-predicates. To ensure that these goals will be executed in the correct context (i.e. in the calling context, not in the meta-predicate definition context) we need to use the meta_predicate/1 directive (in the case of meta non-terminals, thereâs also a meta_non_terminal/1 directive). For example:
:- meta_predicate(findall(*, 0, *)).
:- meta_predicate(map(2, *, *)).
The meta-predicate mode arguments in this directive have the following meaning:
0
Meta-argument that will be called as a goal.
N
Meta-argument that will be a closure used to construct a call by extending it with
N
arguments. The value ofN
must be a positive integer.::
Argument that is context-aware but that will not be called as a goal or a closure. It can contain, however, sub-terms that will be called as goals or closures.
^
Goal that may be existentially quantified (
Vars^Goal
).*
Normal argument.
The following meta-predicate mode arguments are for use only when writing backend Prolog adapter files to deal with proprietary built-in meta-predicates and meta-directives:
/
Predicate indicator (
Name/Arity
), list of predicate indicators, or conjunction of predicate indicators.//
Non-terminal indicator (
Name//Arity
), list of predicate indicators, or conjunction of predicate indicators.[0]
List of goals.
[N]
List of closures.
[/]
List of predicate indicators.
[//]
List of non-terminal indicators.
To the best of my knowledge, the use of non-negative integers to specify closures has first introduced on Quintus Prolog for providing information for predicate cross-reference tools.
Note that Logtalk meta-predicate semantics are different from Prolog meta-predicate semantics (assuming a predicate-based module system as common):
Meta-arguments are always called in the meta-predicate calling context, independently of using explicit or implicit message-sending (to the object defining the meta-predicate when not local). Most Prolog systems have different semantics for explicit versus implicit module qualification.
Logtalk is not based on a predicate prefixing mechanism and thus the meta-predicate directive is required for any predicate with meta-arguments (including when simply passing the meta-arguments to a call to another meta-predicate). This is usually not required in Prolog systems due to the module-prefixing of meta-arguments.
Sending a message from a meta-predicate definition to call a meta-predicate defined in another object resets the calling context for any passed meta-argument to the object sending the message (including for messages to self). Meta-arguments behave differently in Prolog systems due to their module-prefixing.
Logtalk protects from common scenarios where specially crafted meta-predicate definitions are used to break object (and category) encapsulation by changing the meta-arguments passed by client code or trying to subvert the implicit calling context to call client predicates other than the predicates passed as meta-arguments.
Warning
As each Logtalk entity is independently compiled, this directive must be included in every object or category that contains a definition for the described meta-predicate, even if the meta-predicate declaration is inherited from another entity, to ensure proper compilation of meta-arguments.
Discontiguous directiveï
The clause of an object (or category) predicate may not be contiguous. In that case, we must declare the predicate discontiguous by using the discontiguous/1 directive:
:- discontiguous(foo/1).
This is a directive that we should avoid using: it makes your code harder to read and it is not supported by some Prolog backends.
Warning
As each Logtalk entity is compiled independently of other entities, this directive must be included in every object or category that contains a definition for the described predicate (even if the predicate declaration is inherited from other entity).
Dynamic directiveï
An object predicate can be static or dynamic. By default, all predicates (and non-terminals) of static objects defined in source files are static. To declare a dynamic predicate (or non-terminal) we use the dynamic/1 directive. For example:
:- dynamic(foo/1).
Predicates of objects dynamically created at runtime (using the create_object/4 built-in predicate) and predicates of dynamic objects defined in source files (using the dynamic/0 directive) are implicitly dynamic.
Dynamic predicates can be used to represent persistent mutable object state. Note that static objects may declare and define dynamic predicates. Categories can only declare dynamic predicates (with the importing objects holding the predicate definitions).
Warning
As each Logtalk entity is compiled independently from other entities, this directive must be included in every object that contains a definition for the described predicate (even if the predicate declaration is inherited from other object or imported from a category). If we omit the dynamic declaration then the predicate definition will be compiled static.
Operator directiveï
An object (or category) predicate can be declared as an operator using the familiar op/3 directive:
:- op(Priority, Specifier, Operator).
Operators are local to the object (or category) where they are declared. This means that, if you declare a public predicate as an operator, you cannot use operator notation when sending to an object (where the predicate is visible) the respective message (as this would imply visibility of the operator declaration in the context of the sender of the message). If you want to declare global operators and, at the same time, use them inside an entity, just write the corresponding directives at the top of your source file, before the entity opening directive.
Note that operators can also be declared using a scope directive. Only these operators are visible to the current_op/3 reflection method.
When the same operators are used on several entities within the same source file, the corresponding directives must either be repeated in each entity or appear before any entity that uses them. But in the later case, this results in a global scope for the operators. If you prefer the operators to be local to the source file, just undefine them at the end of the file. For example:
% before any entity that uses the operator
:- op(400, xfx, results).
...
% after all entities that used the operator
:- op(0, xfx, results).
Global operators can be declared in the application loader file.
Uses directiveï
When a predicate makes heavy use of predicates defined on other objects, its predicate clauses can be verbose due to all the necessary message sending goals. Consider the following example:
foo :-
...,
findall(X, list::member(X, L), A),
list::append(A, B, C),
list::select(Y, C, R),
...
Logtalk provides a directive, uses/2, which allows us to simplify the code above. One of the usage templates for this directive is:
:- uses(Object, [
Name1/Arity1, Name2/Arity2, ...
]).
Rewriting the code above using this directive results in a simplified and more readable predicate definition:
:- uses(list, [
append/3, member/2, select/3
]).
foo :-
...,
findall(X, member(X, L), A),
append(A, B, C),
select(Y, C, R),
...
Logtalk also supports an extended version of this directive that allows
the declaration of predicate aliases using the
notation Predicate as Alias
(or the alternative notation
Predicate::Alias
). For example:
:- uses(btrees, [new/1 as new_btree/1]).
:- uses(queues, [new/1 as new_queue/1]).
You may use this extended version for solving conflicts between predicates
declared on several uses/2
directives or just for giving new names to
the predicates that will be more meaningful on their using context.
Predicate aliases can also be used to define predicate shorthands, simplifying code maintenance. For example:
:- uses(pretty_printer, [
indent(4, Term) as indent(Term)
]).
Assuming multiple calls to the shorthand, a change to the indent value will require a change to a single line instead of changing every call.
Another common use of predicate aliases is changing the order of the predicate arguments without using lambda expressions. For example:
:- uses(meta, [
fold_left(Closure, Result0, List, Result) as foldl(Closure, List, Result0, Result)
]).
See the directive documentation for details and other examples.
The uses/2
directive allows simpler predicate definitions as long as
there are no conflicts between the predicates declared in the directive
and the predicates defined in the object (or category) containing the
directive. A predicate (or its alias if defined) cannot be listed in
more than one uses/2
directive. In addition, a uses/2
directive
cannot list a predicate (or its alias if defined) which is defined in
the object (or category) containing the directive. Any conflicts are
reported by Logtalk as compilation errors.
The object identifier argument can also be a parameter variable when using the directive in a parametric object or a parametric category. In this case, dynamic binding will necessarily be used for all listed predicates (and non-terminals). The parameter variable must be instantiated at runtime when the messages are sent. This feature simplifies experimenting with multiple implementations of the same protocol (for example, to evaluate the performance of each implementation for a particular case). It also simplifies writing tests that check multiple implementations of the same protocol.
An object (or category) can make a predicate listed in a uses/2
(or
use_module/2
) directive part of its protocol by simply adding a scope
directive for the predicate. For example, in the statistics
library
we have:
:- public(modes/2).
:- uses(numberlist, [modes/2]).
Therefore, a goal such as sample::modes(Sample, Modes)
implicitly calls
numberlist::modes(Sample, Modes)
without requiring an explicit local
definition for the modes/2
predicate (which would trigger a compilation
error).
Alias directiveï
Logtalk allows the definition of an alternative name for an inherited or imported predicate (or for an inherited or imported grammar rule non-terminal) through the use of the alias/2 directive:
:- alias(Entity, [
Predicate1 as Alias1,
Predicate2 as Alias2,
...
]).
This directive can be used in objects, protocols, or categories. The
first argument, Entity
, must be an entity referenced in the opening
directive of the entity containing the alias/2
directive. It can be
an extended or implemented protocol, an imported category, an extended
prototype, an instantiated class, or a specialized class. The second
argument is a list of pairs of predicate indicators (or grammar rule
non-terminal indicators) using the as
infix operator as connector.
A common use for the alias/2
directive is to give an alternative
name to an inherited predicate in order to improve readability. For
example:
:- object(square,
extends(rectangle)).
:- alias(rectangle, [width/1 as side/1]).
...
:- end_object.
The directive allows both width/1
and side/1
to be used as
messages to the object square
. Thus, using this directive, there is
no need to explicitly declare and define a ânewâ side/1
predicate.
Note that the alias/2
directive does not rename a predicate, only
provides an alternative, additional name; the original name continues to
be available (although it may be masked due to the default inheritance
conflict mechanism).
Another common use for this directive is to solve conflicts when two
inherited predicates have the same name and arity. We may want to
call the predicate which is masked out by the Logtalk lookup algorithm
(see the Inheritance section) or we may need to
call both predicates. This is simply accomplished by using the
alias/2
directive to give alternative names to masked out or
conflicting predicates. Consider the following example:
:- object(my_data_structure,
extends(list, set)).
:- alias(list, [member/2 as list_member/2]).
:- alias(set, [member/2 as set_member/2]).
...
:- end_object.
Assuming that both list
and set
objects define a member/2
predicate, without the alias/2
directives, only the definition of
member/2
predicate in the object list
would be visible on the
object my_data_structure
, as a result of the application of the
Logtalk predicate lookup algorithm. By using the alias/2
directives,
all the following messages would be valid (assuming a public scope for
the predicates):
% uses list member/2
| ?- my_data_structure::list_member(X, L).
% uses set member/2
| ?- my_data_structure::set_member(X, L).
% uses list member/2
| ?- my_data_structure::member(X, L).
When used this way, the alias/2
directive provides functionality
similar to programming constructs of other object-oriented languages
that support multi-inheritance (the most notable example probably being
the renaming of inherited features in Eiffel).
Note that the alias/2
directive never hides a predicate which is
visible on the entity containing the directive as a result of the
Logtalk lookup algorithm. However, it may be used to make visible a
predicate which otherwise would be masked by another predicate, as
illustrated in the above example.
The alias/2
directive may also be used to give access to an
inherited predicate, which otherwise would be masked by another
inherited predicate, while keeping the original name as follows:
:- object(my_data_structure,
extends(list, set)).
:- alias(list, [member/2 as list_member/2]).
:- alias(set, [member/2 as set_member/2]).
member(X, L) :-
^^set_member(X, L).
...
:- end_object.
Thus, when sending the message member/2
to my_data_structure
,
the predicate definition in set
will be used instead of the one
contained in list
.
Documenting directiveï
A predicate can be documented with arbitrary user-defined information by using the info/2 directive:
:- info(Name/Arity, List).
The second argument is a list of Key is Value
terms. See the
Documenting section for details.
Multifile directiveï
A predicate can be declared multifile by using the multifile/1 directive:
:- multifile(Name/Arity).
This allows clauses for a predicate to be defined in several objects and/or categories. This is a directive that should be used with care. Itâs commonly used in the definition of hook predicates. Multifile predicates (and non-terminals) may also be declared dynamic using the same predicate (or non-terminal) notation (multifile predicates are static by default).
Logtalk precludes using a multifile predicate for breaking object
encapsulation by checking that the object (or category) declaring the
predicate (using a scope directive) defines it also as multifile.
This entity is said to contain the primary declaration for the multifile
predicate. Entities containing primary multifile predicate declarations
must always be compiled before entities defining clauses for those multifile
predicates. The Logtalk compiler will print a warning if the scope
directive is missing. Note also that the multifile/1
directive
is mandatory when defining multifile predicates.
Consider the following simple example:
:- object(main).
:- public(a/1).
:- multifile(a/1).
a(1).
:- end_object.
After compiling and loading the main
object, we can define other
objects (or categories) that contribute with clauses for the multifile
predicate. For example:
:- object(other).
:- multifile(main::a/1).
main::a(2).
main::a(X) :-
b(X).
b(3).
b(4).
:- end_object.
After compiling and loading the above objects, you can use queries such as:
| ?- main::a(X).
X = 1 ;
X = 2 ;
X = 3 ;
X = 4
yes
Note that the order of multifile predicate clauses depend on several factors, including loading order and compiler implementation details. Therefore, your code should never assume or rely on a specific order of the multifile predicate clauses.
When a clause of a multifile predicate is a rule, its body is compiled
within the context of the object or category defining the clause. This
allows clauses for multifile predicates to call local object or category
predicates. But the values of the sender, this, and self in the
implicit execution context are passed from the clause head to the clause
body. This is necessary to ensure that these values are always valid and
to allow multifile predicate clauses to be defined in categories. A call
to the parameter/2
execution context methods, however, retrieves
parameters of the entity defining the clause, not from the entity for
which the clause is defined. The parameters of the entity for which the
clause is defined can be accessed by simple unification at the clause
head.
Multifile predicate rules should not contain cuts as these may prevent other clauses for the predicate for being used by callers. The compiler prints by default a warning when a cut is found in a multifile predicate definition.
Local calls to the database methods from multifile predicate clauses
defined in an object take place in the object own database instead of
the database of the entity holding the multifile predicate primary
declaration. Similarly, local calls to the expand_term/2
and
expand_goal/2
methods from a multifile predicate clause look for
clauses of the term_expansion/2
and goal_expansion/2
hook
predicates starting from the entity defining the clause instead of the
entity holding the multifile predicate primary declaration. Local calls
to the current_predicate/1
, predicate_property/2
, and
current_op/3
methods from multifile predicate clauses defined in an
object also lookup predicates and their properties in the object own
database instead of the database of the entity holding the multifile
predicate primary declaration.
Coinductive directiveï
A predicate can be declared coinductive by using the coinductive/1 directive. For example:
:- coinductive(comember/2).
Logtalk support for coinductive predicates is experimental and requires a
backend Prolog compiler with minimal support for cyclic terms. The
value of the read-only coinduction flag is set to
supported
for the backend Prolog compilers providing that support.
Synchronized directiveï
A predicate can be declared synchronized by using the synchronized/1 directive. For example:
:- synchronized(write_log_entry/2).
:- synchronized([produce/1, consume/1]).
See the section on synchronized predicates for details.
Defining predicatesï
Object predicatesï
We define object predicates as we have always defined Prolog predicates,
the only difference be that we have four more control structures (the
three message sending operators plus the external call operator) to play
with. For example, if we wish to define an object containing common
utility list predicates like append/2
or member/2
we could write
something like:
:- object(list).
:- public(append/3).
append([], L, L).
append([H| T], L, [H| T2]) :-
append(T, L, T2).
:- public(member/2).
member(H, [H| _]).
member(H, [_| T]) :-
member(H, T).
:- end_object.
Note that, abstracting from the opening and closing object directives and the scope directives, what we have written is also valid Prolog code. Calls in a predicate definition body default to the local predicates, unless we use the message sending operators or the external call operator. This simplifies conversion from plain Prolog code to Logtalk objects: often we just need to add the necessary encapsulation and scope directives to the old code.
Category predicatesï
A category can only contain clauses for static predicates. But there are no restrictions in declaring and calling dynamic predicates from inside a category. Because a category can be imported by multiple objects, dynamic predicates must be called either in the context of self, using the message to self control structure, (::)/1, or in the context of this (i.e. in the context of the object importing the category). For example, if we want to define a category implementing attributes using the dynamic database of self we could write:
:- category(attributes).
:- public(get/2).
:- public(set/2).
:- private(attribute_/2).
:- dynamic(attribute_/2).
get(Var, Value) :-
::attribute_(Var, Value).
set(Var, Value) :-
::retractall(attribute_(Var, _)),
::asserta(attribute_(Var, Value).
:- end_category.
In this case, the get/2
and set/2
predicates will always
access/update the correct definition, contained in the object receiving
the messages.
In alternative, if we want a category implementing attributes using the dynamic database of this, we would write instead:
:- category(attributes).
:- public(get/2).
:- public(set/2).
:- private(attribute_/2).
:- dynamic(attribute_/2).
get(Var, Value) :-
attribute_(Var, Value).
set(Var, Value) :-
retractall(attribute_(Var, _)),
asserta(attribute_(Var, Value).
:- end_category.
In this case, each object importing the category will have its own clauses
for the attribute_/2
private dynamic predicate.
Meta-predicatesï
Meta-predicates may be defined inside objects and categories as any other predicate. A meta-predicate is declared using the meta_predicate/1 directive as described earlier on this section. When defining a meta-predicate, the arguments in the clause heads corresponding to the meta-arguments must be variables. All meta-arguments are called in the context of the object or category calling the meta-predicate. In particular, when sending a message that corresponds to a meta-predicate, the meta-arguments are called in the context of the object or category sending the message.
The most simple example is a meta-predicate with a meta-argument that is called as a goal. E.g. the ignore/1 built-in predicate could be defined as:
:- public(ignore/1).
:- meta_predicate(ignore(0)).
ignore(Goal) :-
(Goal -> true; true).
The 0
in the meta-predicate template tells us that the meta-argument is a
goal that will be called by the meta-predicate.
Some meta-predicates have meta-arguments which are not goals but closures. Logtalk supports the definition of meta-predicates that are called with closures instead of goals as long as the definition uses the call/1-N built-in predicate to call the closure with the additional arguments. A classical example is a list mapping predicate:
:- public(map/2).
:- meta_predicate(map(1, *)).
map(_, []).
map(Closure, [Arg| Args]) :-
call(Closure, Arg),
map(Closure, Args).
Note that in this case the meta-predicate directive specifies that the closure will be extended with exactly one additional argument. When calling a meta-predicate, a closure can correspond to a user-defined predicate, a built-in predicate, a lambda expression, or a control construct.
In some cases, is not a meta-argument but one of its sub-terms that is called as a goal or used as a closure. For example:
:- public(call_all/1).
:- meta_predicate(call_all(::)).
call_all([]).
call_all([Goal| Goals]) :-
call(Goal),
call_all(Goals).
The ::
mode indicator in the meta-predicate template allows the
corresponding argument in the meta-predicate definiton to be a
non-variable term and instructs the compiler to look into the argument
sub-terms for goal and closure meta-variables.
When a meta-predicate calls another meta-predicate, both predicates require
meta_predicate/1
directives. For example, the map/2
meta-predicate
defined above is usually implemented by exchanging the argument order to
take advantage of first-argument indexing:
:- meta_predicate(map(1, *)).
map(Closure, List) :-
map_(List, Closure).
:- meta_predicate(map_(*, 1)).
map_([], _).
map_([Head| Tail], Closure) :-
call(Closure, Head),
map_(Tail, Closure).
Note that Logtalk, unlike most Prolog module systems, is not based on a predicate prefixing mechanism. Thus, the meta-argument calling context is not part of the meta-argument itself.
Lambda expressionsï
The use of lambda expressions as meta-predicate goal and closure arguments often saves writing auxiliary predicates for the sole purpose of calling the meta-predicates. A simple example of a lambda expression is:
| ?- meta::map([X,Y]>>(Y is 2*X), [1,2,3], Ys).
Ys = [2,4,6]
yes
In this example, a lambda expression, [X,Y]>>(Y is 2*X)
, is used as
an argument to the map/3
list mapping predicate, defined in the
library object meta
, in order to double the elements of a list of
integers. Using a lambda expression avoids writing an auxiliary
predicate for the sole purpose of doubling the list elements. The lambda
parameters are represented by the list [X,Y]
, which is connected to
the lambda goal, (Y is 2*X)
, by the (>>)/2
operator. The map/3
predicate calls the lambda goal with fresh/unique variables, represented
by the X
and Y
parameters, for each pair of elements of the second
and third list arguments.
Currying is supported. I.e. it is possible to write a lambda expression whose goal is another lambda expression. The above example can be rewritten as:
| ?- meta::map([X]>>([Y]>>(Y is 2*X)), [1,2,3], Ys).
Ys = [2,4,6]
yes
Lambda expressions may also contain lambda free variables. I.e. variables that are global to the lambda expression and shared with the surrounding meta-call context. Consider the following variant of the previous example:
| ?- between(1, 3, N), meta::map({N}/[X,Y]>>(Y is N*X), [1,2,3], L).
N = 1, L = [1,2,3] ;
N = 2, L = [2,4,6] ;
N = 3, L = [3,6,9]
yes
In this case, the lambda free variable, N
, bound by the between/3
goal, is fixed across all implicit calls made by the map/3
goal.
A second example of free variables in a lambda expression using GNU Prolog as the backend compiler:
| ?- meta::map({Z}/[X,Y]>>(Z#=X+Y), [1,2,3], Zs).
Z = _#22(3..268435455)
Zs = [_#3(2..268435454),_#66(1..268435453),_#110(0..268435452)]
yes
The ISO Prolog construct {}/1
for representing the lambda free
variables as this representation is often associated with set
representation. Note that the order of the free variables is of no
consequence (on the other hand, a list is used for the lambda parameters
as their order does matter).
Both lambda free variables and lambda parameters can be any Prolog term. Consider the following example by Markus Triska:
| ?- meta::map([A-B,B-A]>>true, [1-a,2-b,3-c], Zs).
Zs = [a-1,b-2,c-3]
yes
Lambda expressions can be used, as expected, in non-deterministic queries as in the following example using SWI-Prolog as the backend compiler and Markus Triskaâs CLP(FD) library:
| ?- meta::map({Z}/[X,Y]>>(clpfd:(Z#=X+Y)), Xs, Ys).
Xs = [],
Ys = [] ;
Xs = [_G1369],
Ys = [_G1378],
_G1369+_G1378#=Z ;
Xs = [_G1579, _G1582],
Ys = [_G1591, _G1594],
_G1582+_G1594#=Z,
_G1579+_G1591#=Z ;
Xs = [_G1789, _G1792, _G1795],
Ys = [_G1804, _G1807, _G1810],
_G1795+_G1810#=Z,
_G1792+_G1807#=Z,
_G1789+_G1804#=Z ;
...
As illustrated by the above examples, lambda expression syntax reuses
the ISO Prolog construct {}/1
and the standard operators (/)/2
and (>>)/2
, thus avoiding defining new operators, which is always
tricky for a portable system such as Logtalk. The operator (>>)/2
was chosen as it suggests an arrow, similar to the syntax used in other
languages such as OCaml and Haskell to connect lambda parameters with
lambda functions. This syntax was also chosen in order to simplify
parsing, error checking, and compilation of lambda expressions. The
full specification of the lambda expression syntax can be found in
the the language grammar.
The compiler checks whenever possible that all variables in a lambda expression are either classified as free variables or as lambda parameters. Non-classified variables in a lambda goal (including any anonymous variables) should be regarded as a programming error. The compiler also checks whenever possible if a variable is classified as both a free variable and a lambda parameter. There are a few cases where a variable playing a dual role is intended but, in general, this also results from a programming error. A third check verifies that no lambda parameter variable is used elsewhere in a clause. Such cases are either programming errors, when the variable appears before the lambda expression, or bad programming style, when the variable is used after the lambda expression. These linter warnings are controlled by the lambda_variables flag. Note that the dynamic features of the language and lack of sufficient information at compile time may prevent the compiler of checking all uses of lambda expressions. To improve linter coverage, compile code using lambda expressions with the optimize flag turned on as that will result in additional cases of meta-arguments being evaluated for possible optimizations.
Warning
Variables listed in lambda parameters must not be shared with other goals in a clause.
An optimizing meta-predicate and lambda expression compiler, based on the term-expansion mechanism, is provided as a standard library for practical performance.
A common use of lambda expressions as closure meta-arguments is to workaround
closures always being extended by appending additional argument to construct
a goal. For example, assume that we want to filer a list of atoms by a given
length. We can use the standard atom_length/2
predicate despite the
argument order by writing:
filter(Length, Atoms, Filtered) :-
meta::include({Length}/[Atom]>>atom_length(Atom,Length), Atoms, Filtered).
But Logtalk supports a faster alternative by using predicate aliases to change the argument order when calling library or built-in predicates:
:- uses(user, [
atom_length(Atom, Length) as length_atom(Length, Atom)
]).
filter(Length, Atoms, Filtered) :-
meta::include(length_atom(Length), Atoms, Filtered).
In this case, the performance is no longer dependent on compiling away lambda
expressions. The resulting code is also easier to read (and thus debug and
maintain). But the uses/2
directive is implicitly defining an auxiliary
predicate, which is exactly what we wanted to avoid in the first place by
using a lambda expression.
Redefining built-in predicatesï
Logtalk built-in predicates and Prolog built-in predicates can be redefined inside objects and categories. Although the redefinition of Logtalk built-in predicates should be avoided, the support for redefining Prolog built-in predicates is a practical requirement given the different sets of proprietary built-in predicates provided by backend Prolog systems.
The compiler supports a redefined_built_ins flag, whose default value is silent, that can be set to warning to alert the user of any redefined Logtalk or Prolog built-in predicate.
The redefinition of Prolog built-in predicates can be combined with the
conditional compilation directives
when writing portable applications where some of the supported backends
donât provide a built-in predicate found in the other backends. As an example,
consider the de facto standard msort/2
predicate (which sorts a list while
keeping duplicates). This predicate is provided as a built-in predicate in most
but not all backends. The list
library object includes the code:
:- if(predicate_property(msort(_, _), built_in)).
msort(List, Sorted) :-
{msort(List, Sorted)}.
:- else.
length(List, Length) :-
...
:- endif.
I.e. the object will use the built-in predicate when available. Otherwise,
it will use the predicate definition provided by the list
object.
The redefinition of built-in predicates can also be accomplished using
predicate shorthands. This can be useful
when porting code while minimizing the changes. For example, assume
that existing code uses the format/2
de facto standard predicate
for writing messages. To convert the code to use the
message printing mechanism we could write:
:- uses(logtalk, [
print_message(comment, core, Format+Arguments) as format(Format, Arguments)
]).
process(Crate, Contents) :-
format('Processing crate ~w...', [Crate]),
...,
format('Filing with ~w...', [Contents]),
....
The predicate shorthand instructs the compiler to rewrite all format/2
goals as logtalk::print_message/3
goals, thus allowing us to reuse
the code without changes.
Definite clause grammar rulesï
Definite clause grammar rules (DCGs) provide a convenient notation to represent the parsing and rewrite rules common of most grammars in Prolog. In Logtalk, definite clause grammar rules can be encapsulated in objects and categories. Currently, the ISO/IEC WG17 group is working on a draft specification for a definite clause grammars Prolog standard. Therefore, in the mean time, Logtalk follows the common practice of Prolog compilers supporting definite clause grammars, extending it to support calling grammar rules contained in categories and objects. A common example of a definite clause grammar is the definition of a set of rules for parsing simple arithmetic expressions:
:- object(calculator).
:- public(parse/2).
parse(Expression, Value) :-
phrase(expr(Value), Expression).
expr(Z) --> term(X), "+", expr(Y), {Z is X + Y}.
expr(Z) --> term(X), "-", expr(Y), {Z is X - Y}.
expr(X) --> term(X).
term(Z) --> number(X), "*", term(Y), {Z is X * Y}.
term(Z) --> number(X), "/", term(Y), {Z is X / Y}.
term(Z) --> number(Z).
number(C) --> "+", number(C).
number(C) --> "-", number(X), {C is -X}.
number(X) --> [C], {0'0 =< C, C =< 0'9, X is C - 0'0}.
:- end_object.
After compiling and loading this object, we can test the grammar rules
using the parse/2
message:
| ?- calculator::parse("1+2-3*4", Result).
Result = -9
yes
The non-terminals can be called from predicates using the private built-in
methods phrase/2 and phrase/3 as shown in the
example above. When we want to use the built-in methods phrase/2
and
phrase/3
, the non-terminal used as first argument must be within the
scope of the sender. For the above example, assuming that we want the
predicate corresponding to the expr//1
non-terminal to be public,
the corresponding scope directive would be:
:- public(expr//1).
The //
infix operator used above tells the Logtalk compiler that the
scope directive refers to a grammar rule non-terminal, not to a predicate.
The idea is that the predicate corresponding to the translation of the
expr//1
non-terminal will have a number of arguments equal to one plus
the number of additional arguments necessary for processing the implicit
difference list of tokens.
In the body of a grammar rule, we can call rules that are inherited from ancestor objects, imported from categories, or contained in other objects. This is accomplished by using non-terminals as messages. Using a non-terminal as a message to self allows us to call grammar rules in categories and ancestor objects. To call grammar rules encapsulated in other objects, we use a non-terminal as a message to those objects. Consider the following example, containing grammar rules for parsing natural language sentences:
:- object(sentence,
imports(determiners, nouns, verbs)).
:- public(parse/2).
parse(List, true) :-
phrase(sentence, List).
parse(_, false).
sentence --> noun_phrase, verb_phrase.
noun_phrase --> ::determiner, ::noun.
noun_phrase --> ::noun.
verb_phrase --> ::verb.
verb_phrase --> ::verb, noun_phrase.
:- end_object.
The categories imported by the object would contain the necessary grammar rules for parsing determiners, nouns, and verbs. For example:
:- category(determiners).
:- private(determiner//0).
determiner --> [the].
determiner --> [a].
:- end_category.
Along with the message sending operators ((::)/1
, (::)/2
, and (^^)/1
),
we may also use other control constructs such as (<<)/2
, (\+)/1
, !/0
,
(;)/2
, (->)/2
, {}/1
, call//1-N
, and catch/3
in the body of a
grammar rule. When using a backend Prolog compiler that supports modules, we may
also use the (:)/2
control construct.
Warning
The semantics of (\+)/1
and (->)/2
control constructs in grammar rules
with a terminal or a non-terminal in the first argument are problematic due
to unrestricted look ahead that may or may not be valid depending on the grammar
rule implicit arguments. By default, the linter will print warnings for such
calls (controlled by the grammar_rules flag).
Preferably restrit the use of the (\+)/1
control construct to {}/1
arguments and the use of the (->)/2
control construct to {}/1
test
arguments.
In addition, grammar rules may contain meta-calls (a variable taking the place
of a non-terminal), which are translated to calls of the built-in method
phrase//1
. The meta_non_terminal/1 directive allows the
declaration of non-terminals that have arguments that are meta-called from
grammar rules. For example:
:- meta_non_terminal(zero_or_more(1, *)).
zero_or_more(Closure, [Terminal| Terminals]) -->
call(Closure, Terminal), !, zero_or_more(Closure, Terminals).
zero_or_more(_, []) -->
[].
You may have noticed that Logtalk defines {}/1 as a control construct for bypassing the compiler when compiling a clause body goal. As exemplified above, this is the same control construct that is used in grammar rules for bypassing the expansion of rule body goals when a rule is converted into a clause. Both control constructs can be combined in order to call a goal from a grammar rule body, while bypassing at the same time the Logtalk compiler. Consider the following example:
bar :-
write('bar predicate called'), nl.
:- object(bypass).
:- public(foo//0).
foo --> {{bar}}.
:- end_object.
After compiling and loading this code, we may try the following query:
| ?- logtalk << phrase(bypass::foo, _, _).
bar predicate called
yes
This is the expected result as the expansion of the grammar rule into a
clause leaves the {bar}
goal untouched, which, in turn, is converted
into the goal bar
when the clause is compiled. Note that we tested
the bypass::foo//0
non-terminal by calling the phrase/3
built-in
method in the context of the logtalk
built-in object. This workaround
is necessary due to the Prolog backend implementation of the phrase/3
predicate no being aware of the Logtalk (::)/2
message-sending control
construct semantics.
A grammar rule non-terminal may be declared as dynamic or discontiguous,
as any object predicate, using the same Name//Arity
notation
illustrated above for the scope directives. In addition, grammar rule
non-terminals can be documented using the info/2
directive, as in the following example:
:- public(sentence//0).
:- info(sentence//0, [
comment is 'Rewrites sentence into noun and verb phrases.'
]).
Note
Future Logtalk versions may compile grammar rules differently from Prolog
traditional compilation to prevent name clases between non-terminals and
predicates. Therefore, you should always call non-terminals from predicates
using the phrase/2-3
built-in methods and always call predicates from
grammar rules using the call//1
built-in method. This recommended
practice, besides making your code forward compatible with future Logtalk
versions, also make the code more clear. The linter prints warnings when
these guidelines are not followed (notably, when a predicate is called as
a non-terminal or a non-terminal is called as a predicate).
Built-in methodsï
Built-in methods are built-in object and category predicates. These include methods to access message execution context, to find sets of solutions, to inspect objects, for database handling, for term and goal expansion, and for printing messages. Some of them are counterparts to standard Prolog built-in predicates that take into account Logtalk semantics. Similar to Prolog built-in predicates, built-in methods cannot not be redefined.
Logic and control methodsï
The !/0, true/0, fail/0, false/0, and repeat/0 standard control constructs and logic predicates are interpreted as built-in public methods and thus can be used as messages to any object. In practice, they are only used as messages when sending multiple messages to the same object (see the section on message broadcasting).
Execution context methodsï
Logtalk defines five built-in private methods to access an object execution context. These methods are in the common usage scenarios translated to a single unification performed at compile time with a clause head context argument. Therefore, they can be freely used without worrying about performance penalties. When called from inside a category, these methods refer to the execution context of the object importing the category. These methods are private and cannot be used as messages to objects.
To find the object that received the message under execution we may use the self/1 method. We may also retrieve the object that has sent the message under execution using the sender/1 method.
The method this/1 enables us to retrieve the name of the object for which the predicate clause whose body is being executed is defined instead of using the name directly. This helps to avoid breaking the code if we decide to change the object name and forget to change the name references. This method may also be used from within a category. In this case, the method returns the object importing the category on whose behalf the predicate clause is being executed.
Here is a short example including calls to these three object execution context methods:
:- object(test).
:- public(test/0).
test :-
this(This),
write('Calling predicate definition in '),
writeq(This), nl,
self(Self),
write('to answer a message received by '),
writeq(Self), nl,
sender(Sender),
write('that was sent by '),
writeq(Sender), nl, nl.
:- end_object.
:- object(descendant,
extends(test)).
:- end_object.
After compiling and loading these two objects, we can try the following goal:
| ?- descendant::test.
Calling predicate definition in test
to answer a message received by descendant
that was sent by user
yes
Note that the goals self(Self)
, sender(Sender)
, and
this(This)
, being translated to unifications with the clause head
context arguments at compile time, are effectively removed from the
clause body. Therefore, a clause such as:
predicate(Arg) :-
self(Self),
atom(Arg),
... .
is compiled with the goal atom(Arg)
as the first condition on the
clause body. As such, the use of these context execution methods do not
interfere with the optimizations that some Prolog compilers perform when
the first clause body condition is a call to a built-in type-test
predicate or a comparison operator.
For parametric objects and categories, the method parameter/2 enables us to retrieve current parameter values (see the section on parametric objects for a detailed description). For example:
:- object(block(_Color)).
:- public(test/0).
test :-
parameter(1, Color),
write('Color parameter value is '),
writeq(Color), nl.
:- end_object.
An alternative to the parameter/2
predicate is to use
parameter variables:
:- object(block(_Color_)).
:- public(test/0).
test :-
write('Color parameter value is '),
writeq(_Color_), nl.
:- end_object.
After compiling and loading either version of the object, we can try the following goal:
| ?- block(blue)::test.
Color parameter value is blue
yes
Calls to the parameter/2
method are translated to a compile time
unification when the second argument is a variable. When the second
argument is bound, the calls are translated to a call to the built-in
predicate arg/3
.
When type-checking predicate arguments, it is often useful to include
the predicate execution context when reporting an argument error. The
context/1 method provides access to that context. For
example, assume a predicate foo/2
that takes an atom and an integer
as arguments. We could type-check the arguments by writing (using the
library type
object):
foo(A, N) :-
% type-check arguments
context(Context),
type::check(atom, A, Context),
type::check(integer, N, Context),
% arguments are fine; go ahead
... .
Error handling and throwing methodsï
Besides the catch/3 and throw/1 methods inherited from
Prolog, Logtalk also provides a set of convenience methods to throw
standard error/2
exception terms:
instantiation_error/0,
uninstantiation_error/1,
type_error/2,
domain_error/2,
existence_error/2,
permission_error/3,
representation_error/1,
evaluation_error/1,
resource_error/1,
syntax_error/1, and
system_error/0. When using these methods, the second argument
of the error/2
exception term is bound to the execution context (as it
would be provided by the context/1 method).
Database methodsï
Logtalk provides a set of built-in methods for object database handling
similar to the usual database Prolog predicates:
abolish/1,
asserta/1,
assertz/1,
clause/2,
retract/1, and
retractall/1. These
methods always operate on the database of the object receiving the corresponding
message. When called locally, these predicates take into account any
uses/2 or use_module/2 directives that refer
to the dynamic predicate being handled. For example, in the following object, the
clauses for the data/1
predicate are retracted and asserted in user
due to
the uses/2
directive:
:- object(an_object).
:- uses(user, [data/1]).
:- public(some_predicate/1).
some_predicate(Arg) :-
retractall(data(_)),
assertz(data(Arg)).
:- end_object.
When working with dynamic grammar rule non-terminals, you may use the built-in method expand_term/2 convert a grammar rule into a clause that can then be used with the database methods.
Logtalk also supports asserta/2
, assertz/2
, clause/3
, and
erase/1
built-in methods when run with a backend that supports the
corresponding legacy built-in predicates that work with
clause references.
Meta-call methodsï
Logtalk supports the generalized call/1-N meta-predicate. This built-in private meta-predicate must be used in the implementation of meta-predicates which work with closures instead of goals. In addition, Logtalk supports the built-in private meta-predicates ignore/1, once/1, and (\+)/1. These methods cannot be used as messages to objects.
All solutions methodsï
The usual all solutions meta-predicates are built-in private methods in Logtalk: bagof/3, findall/3, findall/4, and setof/3. There is also a forall/2 method that implements generate-and-test loops. These methods cannot be used as messages to objects.
Reflection methodsï
Logtalk provides a comprehensive set of built-in predicates and built-in methods for querying about entities and predicates. Some of the information, however, requires that the source files are compiled with the source_data flag turned on.
The reflection API supports two different views on entities and their contents, which we may call the transparent box view and the black box view. In the transparent box view, we look into an entity disregarding how it will be used and returning all information available on it, including predicate declarations and predicate definitions. This view is supported by the entity property built-in predicates. In the black box view, we look into an entity from a usage point-of-view using built-in methods for inspecting object operators and predicates that are within scope from where we are making the call: current_op/3, which returns operator specifications, predicate_property/2, which returns predicate properties, and current_predicate/1, which enables us to query about user-defined predicate definitions. See below for a more detailed description of these methods.
Definite clause grammar parsing methods and non-terminalsï
Logtalk supports two definite clause grammar parsing built-in private methods, phrase/2 and phrase/3, with definitions similar to the predicates with the same name found on most Prolog compilers that support definite clause grammars. These methods cannot be used as messages to objects.
Logtalk also supports phrase//1, call//1-N, and
eos//0 built-in non-terminals.
The call//1-N
non-terminals takes a closure (which can be a lambda
expression) plus zero or more additional arguments and are processed by
appending the input list of tokens and the list of remaining tokens to
the arguments.
Predicate propertiesï
We can find the properties of visible predicates by calling the predicate_property/2 built-in method. For example:
| ?- bar::predicate_property(foo(_), Property).
Note that this method takes into account the predicateâs scope declarations. In the above example, the call will only return properties for public predicates.
An objectâs set of visible predicates is the union of all the predicates declared for the object with all the built-in methods and all the Logtalk and Prolog built-in predicates.
The following predicate properties are supported:
scope(Scope)
The predicate scope (useful for finding the predicate scope with a single call to
predicate_property/2
)public
,protected
,private
The predicate scope (useful for testing if a predicate have a specific scope)
static
,dynamic
All predicates are either static or dynamic (note, however, that a dynamic predicate can only be abolished if it was dynamically declared)
logtalk
,prolog
,foreign
A predicate can be defined in Logtalk source code, Prolog code, or in foreign code (e.g. in C)
built_in
The predicate is a built-in predicate
multifile
The predicate is declared multifile (i.e. it can have clauses defined in multiple files or entities)
meta_predicate(Template)
The predicate is declared as a meta-predicate with the specified template
coinductive(Template)
The predicate is declared as a coinductive predicate with the specified template
declared_in(Entity)
The predicate is declared (using a scope directive) in the specified entity
defined_in(Entity)
The predicate definition is looked up in the specified entity (note that this property does not necessarily imply that clauses for the predicate exist in
Entity
; the predicate can simply be false as per the closed-world assumption)redefined_from(Entity)
The predicate is a redefinition of a predicate definition inherited from the specified entity
non_terminal(NonTerminal//Arity)
The predicate resulted from the compilation of the specified grammar rule non-terminal
alias_of(Predicate)
The predicate (name) is an alias for the specified predicate
alias_declared_in(Entity)
The predicate alias is declared in the specified entity
synchronized
The predicate is declared as synchronized (i.e. itâs a deterministic predicate synchronized using a mutex when using a backend Prolog compiler supporting a compatible multi-threading implementation)
Some properties are only available when the entities are defined in source files and when those source files are compiled with the source_data flag turned on:
recursive
The predicate definition includes at least one recursive rule
inline
The predicate definition is inlined
auxiliary
The predicate is not user-defined but rather automatically generated by the compiler or the term-expansion mechanism
mode(Mode, Solutions)
Instantiation, type, and determinism mode for the predicate (which can have multiple modes)
info(ListOfPairs)
Documentation key-value pairs as specified in the user-defined
info/2
directivenumber_of_clauses(N)
The number of clauses for the predicate existing at compilation time (note that this property is not updated at runtime when asserting and retracting clauses for dynamic predicates)
number_of_rules(N)
The number of rules for the predicate existing at compilation time (note that this property is not updated at runtime when asserting and retracting clauses for dynamic predicates)
declared_in(Entity, Line)
The predicate is declared (using a scope directive) in the specified entity in a source file at the specified line (if applicable)
defined_in(Entity, Line)
The predicate is defined in the specified entity in a source file at the specified line (if applicable)
redefined_from(Entity, Line)
The predicate is a redefinition of a predicate definition inherited from the specified entity, which is defined in a source file at the specified line (if applicable)
alias_declared_in(Entity, Line)
The predicate alias is declared in the specified entity in a source file at the specified line (if applicable)
The properties declared_in/1-2
, defined_in/1-2
, and
redefined_from/1-2
do not apply to built-in methods and Logtalk or
Prolog built-in predicates. Note that if a predicate is declared in a
category imported by the object, it will be the category name â not the
object name â that will be returned by the property declared_in/1
.
The same is true for protocol declared predicates.
Some properties such as line numbers are only available when the entity
holding the predicates is defined in a source file compiled with the
source_data flag turned on. Moreover, line
numbers are only supported in backend Prolog compilers
that provide access to the start line of a read term. When such support is
not available, the value -1
is returned for the start and end lines.
Finding declared predicatesï
We can find, by backtracking, all visible user predicates by calling the current_predicate/1 built-in method. This method takes into account predicate scope declarations. For example, the following call will only return user predicates that are declared public:
| ?- some_object::current_predicate(Name/Arity).
The predicate property non_terminal/1
may be used to retrieve all
grammar rule non-terminals declared for an object. For example:
current_non_terminal(Object, Name//Args) :-
Object::current_predicate(Name/Arity),
functor(Predicate, Functor, Arity),
Object::predicate_property(Predicate, non_terminal(Name//Args)).
Usually, the non-terminal and the corresponding predicate share the same functor but users should not rely on this always being true.
Calling Prolog predicatesï
Logtalk is designed for both robustness and portability. In the context
of calling Prolog predicates, robustness requires that the compilation of
Logtalk source code must not have accidental dependencies on Prolog code that
happens to be loaded at the time of the compilation. One immediate consequence
is that only Prolog built-in predicates are visible from within objects and
categories. But Prolog systems provide a widely diverse set of built-in
predicates, easily rising portability issues. Relying on non-standard
predicates is often unavoidable, however, due to the narrow scope of Prolog
standards. Logtalk applications may also require calling user-defined Prolog
predicates, either in user
or in Prolog modules.
Calling Prolog built-in predicatesï
In predicate clauses and object initialization/1
directives, predicate
calls that are not prefixed with a message sending, super call, or module
qualification operator (::
, ^^
, or :
), are compiled to either
calls to local predicates or as calls to Logtalk/Prolog built-in predicates.
A predicate call is compiled as a call to a local predicate if the object
(or category) contains a scope directive, a multifile directive, a dynamic
directive, or a definition for the called predicate. When that is not the
case, the compiler checks if the call corresponds to a Logtalk or Prolog
built-in predicate. Consider the following example:
foo :-
...,
write(bar),
...
The call to the write/1
predicate will be compiled as a call to the
corresponding Prolog standard built-in predicate unless the object (or
category) containing the above definition also contains a predicate
named write/1
or a directive for the predicate.
When calling non-standard Prolog built-in predicates or using non-standard
Prolog arithmetic functions, we may run into portability problems while
trying your applications with different backend Prolog compilers. We can
use the compiler portability flag to generate
warnings for calls to non-standard predicates and arithmetic functions.
We can also help document those calls using the uses/2
directive. For example, a few Prolog systems provide an atom_string/2
non-standard predicate. We can write (in the object or category calling the
predicate):
:- uses(user, [atom_string/2])
This directive is based on the fact that built-in predicates are visible in
plain Prolog (i.e. in user
). Besides helping to document the dependency
on a non-standard built-in predicate, this directive will also silence the
compiler portability warning.
Calling Prolog non-standard built-in meta-predicatesï
Prolog built-in meta-predicates may only be called locally within
objects or categories, i.e. they cannot be used as messages. Compiling
calls to non-standard, Prolog built-in meta-predicates can be tricky,
however, as there is no standard way of checking if a built-in predicate
is also a meta-predicate and finding out which are its meta-arguments.
But Logtalk supports overriding the original meta-predicate template
when not programmatically available or usable. For example, assume a
det_call/1
Prolog built-in meta-predicate that takes a goal as
argument. We can add to the object (or category) calling it the
directive:
:- meta_predicate(user::det_call(0)).
Another solution is to explicitly declare all non-standard built-in Prolog
meta-predicates in the corresponding adapter file using the internal
predicate '$lgt_prolog_meta_predicate'/3
. For example:
'$lgt_prolog_meta_predicate'(det_call(_), det_call(0), predicate).
The third argument can be either the atom predicate
or the atom
control_construct
, a distinction that is useful when compiling in
debug mode.
Calling Prolog foreign predicatesï
Prolog systems often support defining foreign predicates, i.e. predicates defined using languages other than Prolog using a foreign language interface. There isnât, however, any standard for defining, making available, and recognizing foreign predicates. From a Logtalk perspective, the two most common scenarios are calling a foreign predicate (from within an object or a category) and making a set of foreign predicates available as part of an object (or category) protocol. Assuming, as this is the most common case, that foreign predicates are globally visible once made available (using a Prolog system specific loading or linking procedure), we can simply call them as user-defined plain predicates, as explained in the next section. When defining an object (or category) that makes available foreign predicates, the advisable solution is to name the predicates after the object (or category) and then define object (or category) predicates that call the foreign predicates. Most backend adapter files include support for recognizing foreign predicates that allows the Logtalk compiler to inline calls to the predicates (thus avoiding call indirection overheads).
Calling Prolog user-defined plain predicatesï
User-defined Prolog plain predicates (i.e. predicates that are not defined
in a Prolog module) can be called from within objects or categories by
sending the corresponding message to user
. For example:
foo :-
...,
user::bar,
...
In alternative, we can use the uses/2 directive and write:
:- uses(user, [bar/0]).
foo :-
...,
bar,
...
Note that user
is a pseudo-object in Logtalk containing all predicate
definitions that are not encapsulated (either in a Logtalk entity or a
Prolog module).
When the Prolog predicate is not a meta-predicate, we can also use the {}/1 compiler bypass control construct. For example:
foo :-
...,
{bar},
...
But note that in this case the reflection API
will not record the dependency of the foo/0
predicate on the Prolog
bar/0
predicate as we are effectively bypassing the compiler.
Calling Prolog module predicatesï
Prolog module predicates can be called from within objects or categories by using explicit qualification. For example:
foo :-
...,
module:bar,
...
You can also use in alternative the use_module/2 directive to call the module predicates using implicit qualification:
:- use_module(module, [bar/0]).
foo :-
...,
bar,
...
Note that the first argument of the use_module/2
directive, when used
within an object or a category, is a module name, not a file specification
(also be aware that Prolog modules are sometimes defined in files with names
that differ from the module names).
As loading a Prolog module varies between Prolog systems, the actual loading directive or goal is preferably done from the application loader file. An advantage of this approach is that it contributes to a clean separation between loading and using a resource with the loader file being the central point that loads all application resources (complex applications often use a hierarchy of loader files but the main idea remains the same).
As an example, assume that we need to call predicates defined in a CLP(FD)
Prolog library, which can be loaded using library(clpfd)
as the file
specification. In the loader file, we would add:
:- use_module(library(clpfd), []).
Specifying an empty import list is often used to avoid adding the module exported predicates to plain Prolog. In the objects and categories we can then call the library predicates, using implicit or explicit qualification, as explained. For example:
:- object(puzzle).
:- public(puzzle/1).
:- use_module(clpfd, [
all_different/1, ins/2, label/1,
(#=)/2, (#\=)/2,
op(700, xfx, #=), op(700, xfx, #\=)
]).
puzzle([S,E,N,D] + [M,O,R,E] = [M,O,N,E,Y]) :-
Vars = [S,E,N,D,M,O,R,Y],
Vars ins 0..9,
all_different(Vars),
S*1000 + E*100 + N*10 + D +
M*1000 + O*100 + R*10 + E #=
M*10000 + O*1000 + N*100 + E*10 + Y,
M #\= 0, S #\= 0,
label([M,O,N,E,Y]).
:- end_object.
Warning
The actual module code must be loaded prior to compilation of Logtalk source code that uses it. In particular, programmers should not expect that the module be auto-loaded (including when using a backend Prolog compiler that supports an auto-loading mechanism).
The module identifier argument can also be a parameter variable when using the directive in a parametric object or a parametric category. In this case, dynamic binding will necessarily be used for all listed predicates (and non-terminals). The parameter variable must be instantiated at runtime when the calls are made.
Logtalk supports the declaration of predicate aliases
and predicate shorthands in use_module/2
directives used within object and categories. For example, the ECLiPSe IC
Constraint Solvers define a (::)/2
variable domain operator that clashes
with the Logtalk (::)/2
message sending operator. We can solve the conflict
by writing:
:- use_module(ic, [(::)/2 as ins/2]).
With this directive, calls to the ins/2
predicate alias will be
automatically compiled by Logtalk to calls to the (::)/2
predicate in
the ic
module.
Calling Prolog module meta-predicatesï
The Logtalk library provides implementations of common meta-predicates,
which can be used in place of module meta-predicates (e.g. list mapping
meta-predicates). If that is not the case, the Logtalk compiler may need
help to understand the module meta-predicate templates. Despite some recent
progress in standardization of the syntax of meta_predicate/1
directives
and of the meta_predicate/1
property returned by the predicate_property/2
reflection predicate, portability is still a major problem. Thus, Logtalk
allows the original meta_predicate/1
directive to be overridden
with a local directive that Logtalk can make sense of. Note that Logtalk
is not based on a predicate prefixing mechanism as found in module systems.
This fundamental difference precludes an automated solution at the Logtalk
compiler level.
As an example, assume that you want to call from an object (or a category) a module meta-predicate with the following meta-predicate directive:
:- module(foo, [bar/2]).
:- meta_predicate(bar(*, :)).
The :
meta-argument specifier is ambiguous. It tell us that the second
argument of the meta-predicate is module sensitive but it does not tell us
how. Some legacy module libraries and some Prolog systems use :
to
mean 0
(i.e. a meta-argument that will be meta-called). Some others
use :
for meta-arguments that are not meta-called but that still need
to be augmented with module information. Whichever the case, the Logtalk
compiler doesnât have enough information to unambiguously parse the
directive and correctly compile the meta-arguments in the meta-predicate
call. Therefore, the Logtalk compiler will generate an error stating that
:
is not a valid meta-argument specifier when trying to compile a
foo:bar/2
goal. There are two alternative solutions for this problem.
The advised solution is to override the meta-predicate directive by writing,
inside the object (or category) where the meta-predicate is called:
:- meta_predicate(foo:bar(*, *)).
or:
:- meta_predicate(foo:bar(*, 0)).
depending on the true meaning of the second meta-argument. The second alternative, only usable when the meta-argument can be handled as a normal argument, is to simply use the {}/1 compiler bypass control construct to call the meta-predicate as-is:
... :- {foo:bar(..., ...)}, ...
The downside of this alternative is that it hides the dependency on the module library from the reflection API and thus from the developer tools.
Defining Prolog multifile predicatesï
Some Prolog module libraries, e.g. constraint packages, expect clauses for some library predicates to be defined in other modules. This is accomplished by declaring the library predicate multifile and by explicitly prefixing predicate clause heads with the library module identifier. For example:
:- multifile(clpfd:run_propagator/2).
clpfd:run_propagator(..., ...) :-
...
Logtalk supports the definition of Prolog module multifile predicates in objects and categories. While the clause head is compiled as-is, the clause body is compiled in the same way as a regular object or category predicate, thus allowing calls to local object or category predicates. For example:
:- object(...).
:- multifile(clpfd:run_propagator/2).
clpfd:run_propagator(..., ...) :-
% calls to local object predicates
...
:- end_object.
The Logtalk compiler will print a warning if the multifile/1
directive is missing. These multifile predicates may also be declared
dynamic using the same Module:Name/Arity
notation.
Asserting and retracting Prolog predicatesï
To assert and retract clauses for Prolog dynamic predicates, we can use an explicitly qualified module argument. For example:
:- object(...).
:- dynamic(m:bar/1).
foo(X) :-
retractall(m:bar(_)),
assertz(m:bar(X)),
...
:- end_object.
In alternative, we can use use_module/2 directives to declare the module predicates. For example:
:- object(...).
:- use_module(m, [bar/1]).
:- dynamic(m:bar/1).
foo(X) :-
% retract and assert bar/1 clauses in module m
retractall(bar(_)),
assertz(bar(X)),
...
:- end_object.
When the Prolog dynamic predicates are defined in user
, the recommended
and most portable practice (as not all backends support a module system) is
to use a uses/2 directive:
:- object(...).
:- uses(user, [bar/1]).
:- dynamic(user::bar/1).
foo(X) :-
% retract and assert bar/1 clauses in user
retractall(bar(_)),
assertz(bar(X)),
...
:- end_object.
Note that in the alternatives using uses/2
or use_module/2
directives,
the argument of the database handling predicates must be know at compile time.
If that is not the case, you must use instead either an explicitly-qualified
argument or the {}/1 control construct. For example:
:- object(...).
add(X) :-
% assert clause X in module m
assertz(m:X),
...
remove(Y) :-
% retract all clauses in user whose head unifies with Y
{retractall(Y)},
...
:- end_object.