Error handling

Error handling is accomplished in Logtalk by using the standard catch/3 and throw/1 predicates [ISO95] together with a set of built-in methods that simplify generating errors decorated with expected context.

Errors thrown by Logtalk have, whenever possible, the following format:

error(Error, logtalk(Goal, ExecutionContext))

In this exception term, Goal is the goal that triggered the error Error and ExecutionContext is the context in which Goal is called. For example:

error(
    permission_error(modify,private_predicate,p/0),
    logtalk(foo::abolish(p/0), _)
)

Note, however, that Goal and ExecutionContext can be unbound or only partially instantiated when the corresponding information is not available (e.g. due to compiler optimizations that throw away the necessary error context information). The ExecutionContext argument is an opaque term that can be decoded using the logtalk::execution_context/7 predicate.

Raising Exceptions

The error handling section in the reference manual lists a set of convenient built-in methods that generate error/2 exception terms with the expected context argument. For example, instead of manually constructing a type error as in:

...,
context(Context),
throw(error(type_error(atom, 42), Context)).

we can simply write:

...,
type_error(atom, 42).

The provided error built-in methods cover all standard error types found in the ISO Prolog Core standard.

Type-checking

One of the most common case where errors may be generated is when type-checking predicate arguments and input data before processing it. The standard library includes a type object that defines an extensive set of types, together with predicates for validating and checking terms. The set of types is user extensible and new types can be defined by adding clauses for the type/1 and check/2 multifile predicates. For example, assume that we want to be able to check temperatures expressed in Celsius, Fahrenheit, or Kelvin scales. We start by declaring (in an object or category) the new type:

:- multifile(type::type/1).
type::type(temperature(_Unit)).

Next, we need to define the actual code that would verify that a temperature is valid. As the different scales use a different value for absolute zero, we can write:

:- multifile(type::check/2).
type::check(temperature(Unit), Term) :-
    check_temperature(Unit, Term).

% given that temperature has only a lower bound, we make use of the library
% property/2 type to define the necessary test expression for each unit
check_temperature(celsius, Term) :-
    type::check(property(float, [Temperature]>>(Temperature >= -273.15)), Term).
check_temperature(fahrenheit, Term) :-
    type::check(property(float, [Temperature]>>(Temperature >= -459.67)), Term).
check_temperature(kelvin, Term) :-
    type::check(property(float, [Temperature]>>(Temperature >= 0.0)), Term).

With this definition, a term is first checked that it is a float value before checking that it is in the expected open interval. But how do we use this new type? If we want just to test if a temperature is valid, we can write:

..., type::valid(temperature(celsius), 42.0), ...

The type::valid/2 predicate succeeds or fails depending on the second argument being of the type specified in the first argument. If instead of success or failure we want to generate an error for invalid values, we can use the type::check/2 predicate instead:

..., type::check(temperature(celsius), 42.0), ...

If we require an error/2 exception term with the error context, we can use instead the type::check/3 predicate:

...,
context(Context),
type::check(temperature(celsius), 42.0, Context),
...

Note that context/1 calls are inlined and messages to the library type object use static binding when compiling with the optimize flag turned on, thus enabling efficient type-checking.

Expected terms

Support for representing and handling expected terms is provided by the expecteds library. Expected terms allows defering errors to later stages of an application in alternative to raising an exception as soon as an error is detected.

Compiler warnings and errors

The current Logtalk compiler uses the standard read_term/3 built-in predicate to read and compile a Logtalk source file. This improves the compatibility with backend Prolog compilers and their proprietary syntax extensions and standard compliance quirks. But one consequence of this design choice is that invalid Prolog terms or syntax errors may abort the compilation process with limited information given to the user (due to the inherent limitations of the read_term/3 predicate).

Assuming that all the terms in a source file are valid, there is a set of errors and potential errors, described below, that the compiler will try to detect and report, depending on the used compiler flags (see the Compiler flags section of this manual on lint flags for details).

Unknown entities

The Logtalk compiler warns about any referenced entity that is not currently loaded. The warning may reveal a misspell entity name or just an entity that it will be loaded later. Out-of-oder loading should be avoided when possible as it prevents some code optimizations such as static binding of messages to methods.

Singleton variables

Singleton variables in a clause are often misspell variables and, as such, one of the most common errors when programming in Prolog. Assuming that the backend Prolog compiler implementation of the read_term/3 predicate supports the standard singletons/1 option, the compiler warns about any singleton variable found while compiling a source file.

Redefinition of Prolog built-in predicates

The Logtalk compiler will warn us of any redefinition of a Prolog built-in predicate inside an object or category. Sometimes the redefinition is intended. In other cases, the user may not be aware that a particular backend Prolog compiler may already provide the predicate as a built-in predicate or may want to ensure code portability among several Prolog compilers with different sets of built-in predicates.

Redefinition of Logtalk built-in predicates

Similar to the redefinition of Prolog built-in predicates, the Logtalk compiler will warn us if we try to redefine a Logtalk built-in. But the redefinition will probably be an error in most (if not all) cases.

Redefinition of Logtalk built-in methods

An error will be thrown if we attempt to redefine a Logtalk built-in method inside an entity. The default behavior is to report the error and abort the compilation of the offending entity.

Misspell calls of local predicates

A warning will be reported if Logtalk finds (in the body of a predicate definition) a call to a local predicate that is not defined, built-in (either in Prolog or in Logtalk) or declared dynamic. In most cases these calls are simple misspell errors.

Portability warnings

A warning will be reported if a predicate clause contains a call to a non-standard built-in predicate or arithmetic function, Portability warnings are also reported for non-standard flags or flag values. These warnings often cannot be avoided due to the limited scope of the ISO Prolog standard.

Deprecated elements

A warning will be reported if a deprecated directive, control construct, or predicate is used. These warnings should be fixed as soon as possible as support for any deprecated features will likely be discontinued in future versions.

Missing directives

A warning will be reported for any missing dynamic, discontiguous, meta-predicate, and public predicate directive.

Duplicated directives

A warning will be reported for any duplicated scope, multifile, dynamic, discontiguous, meta-predicate, and meta-non-terminal directives. Note that conflicting directives for the same predicate are handled as errors, not as duplicated directive warnings.

Duplicated clauses

A warning will be reported for any duplicated entity clauses. This check is computationally heavy, however, and usually turned off by default.

Goals that are always true or false

A warning will be reported for any goal that is always true or false. This is usually caused by typos in the code. For example, writing X == y instead of X == Y.

Trivial fails

A warning will be reported for any call to a local static predicate with no matching clause.

Suspicious calls

A warning will be reported for calls that are syntactically correct but most likely a semantic error. An example is (::)/1 calls in clauses that apparently are meant to implement recursive predicate definitions where the user intention is to call the local predicate definition.

Lambda variables

A warning will be reported for lambda expressions with unclassified variables (not listed as either lambda free or lambda parameter variables), for variables playing a dual role (as both lambda free and lambda parameter variables), and for lambda parameters used elsewhere in a clause.

Redefinition of predicates declared in uses/2 or use_module/2 directives

A error will be reported for any attempt to define locally a predicate that is already declared in an uses/2 or use_module/2 directive.

Other warnings and errors

The Logtalk compiler will throw an error if it finds a predicate clause or a directive that cannot be parsed. The default behavior is to report the error and abort the compilation.

Runtime errors

This section briefly describes runtime errors that result from misuse of Logtalk built-in predicates, built-in methods or from message sending. For a complete and detailed description of runtime errors please consult the Reference Manual.

Logtalk built-in predicates

Most Logtalk built-in predicates checks the type and mode of the calling arguments, throwing an exception in case of misuse.

Logtalk built-in methods

Most Logtalk built-in method checks the type and mode of the calling arguments, throwing an exception in case of misuse.

Message sending

The message sending mechanisms always check if the receiver of a message is a defined object and if the message corresponds to a declared predicate within the scope of the sender. The built-in protocol forwarding declares a predicate, forward/1, which is automatically called (if defined) by the runtime for any message that the receiving object does not understand. The usual definition for this error handler is to delegate or forward the message to another object that might be able to answer it:

forward(Message) :-
    % forward the message while preserving the sender
    [Object::Message].

If preserving the original sender is not required, this definition can be simplified to:

forward(Message) :-
    Object::Message.

More sophisticated definitions are, of course, possible.