Printing messages and asking questions

Applications, components, and libraries often print all sorts of messages. These include banners, logging, debugging, and computation results messages. But also, in some cases, user interaction messages. However, the authors of applications, components, and libraries often cannot anticipate the context where their software will be used and thus decide which and when messages should be displayed, suppressed, or diverted. Consider the different components in a Logtalk application development and deployment. At the base level, you have the Logtalk compiler and runtime. The compiler writes messages related to e.g. compiling and loading files, compiling entities, and compilation warnings and errors. The runtime may write banner messages or throw execution errors that may result in printing human-level messages. The development environment can be console-based, or you may be using a GUI tool such as PDT. In the latter case, PDT needs to intercept the Logtalk compiler and runtime messages to present the relevant information using its GUI. Then you have all the other components in a typical application. For example, your own libraries and third-party libraries. The libraries may want to print messages on their own, e.g. banners, debugging information, or logging information. As you assemble all your application components, you want to have the final word on which messages are printed, where, and when. Uncontrolled message printing by libraries could potentially disrupt application flow, expose implementation details, spam the user with irrelevant details, or break user interfaces.

The solution is to decouple the calls to print a message from the actual printing of the output text. The same is true for calls to read user input. By decoupling the call to input some data from the actual read of the data, we can easily switch from, for example, a command-line interface to a GUI input dialog or even automate providing the data (e.g., when automating testing of the user interaction).

Logtalk provides a solution based on the structured message printing mechanism that was introduced by Quintus Prolog, where it was apparently implemented by Dave Bowen (thanks to Richard O’Keefe for the historical bits). This mechanism gives the programmer full control of message printing, allowing it to filter, rewrite, or redirect any message. Variations of this mechanism can also be found in some Prolog systems, including SICStus Prolog, SWI-Prolog, and YAP. Based on this mechanism, Logtalk introduces an extension that also allows abstracting asking a user for input. Both mechanisms are implemented by the logtalk built-in object and described in this section. The message printing mechanism is extensively used by the Logtalk compiler itself and by the developer tools. The question-asking mechanism is used e.g. in the debugger tool.

Printing messages

The main predicate for printing a message is logtalk::print_message/3. A simple example, using the Logtalk runtime, is:

| ?- logtalk::print_message(banner, core, banner).

Logtalk 3.23.0
Copyright (c) 1998-2018 Paulo Moura
yes

The first argument of the predicate is the kind of message that we want to print. In this case, we use banner to indicate that we are printing a product name and copyright banner. An extensive list of message kinds is supported by default:

banner

banner messages (used e.g. when loading tools or main application components; can be suppressed by setting the report flag to warnings or off)

help

messages printed in reply to the user asking for help (mostly for helping port existing Prolog code)

information and information(Group)

messages usually printed in reply to a user’s request for information

silent and silent(Group)

not printed by default (but can be intercepted using the message_hook/4 predicate)

comment and comment(Group)

useful but usually not essential messages (can be suppressed by setting the report flag to warnings or off)

warning and warning(Group)

warning messages (generated e.g. by the compiler; can be suppressed by turning off the report flag)

error and error(Group)

error messages (generated e.g. by the compiler)

debug, debug(Group)

debugging messages (by default, only printed when the debug flag is turned on; the print_message/3 goals for these messages are suppressed by the compiler when the optimize flag is turned on)

question, question(Group)

questions to a user

Using a compound term allows easy partitioning of messages of the same kind in different groups. Note that you can define your own alternative message kind identifiers for your own components, together with suitable definitions for their associated prefixes and output streams.

The second argument of print_message/3 represents the component defining the message being printed. In this context, component is a generic term that can designate, e.g., a tool, a library, or some sub-system in a large application. In our example, the component name is core, identifying the Logtalk compiler/runtime. This argument was introduced to provide multiple namespaces for message terms and thus simplify programming-in-the-large by allowing easy filtering of all messages from a specific component and also avoiding conflicts when two components happen to define the same message term (e.g., banner). Users should choose and use a unique name for a component, which usually is the name of the component itself. For example, all messages from the lgtunit tool use lgtunit for the component argument. The compiler and runtime are interpreted as a single component designated as core.

The third argument of print_message/3 is the message itself, represented by a term. In the above example, the message term is banner. Using a term to represent a message instead of a string with the message text itself has significant advantages. Notably, it allows using a compound term for easy parameterization of the message text and simplifies machine processing, localization of applications, and message interception. For example:

| ?- logtalk::print_message(comment, core, redefining_entity(object, foo)).

% Redefining object foo
yes

Message tokenization

The use of message terms requires a solution for generating the actual text of the messages. This is supported by defining grammar rules for the logtalk::message_tokens//2 multifile non-terminal, which translates a message term, for a given component, to a list of tokens. For example:

:- multifile(logtalk::message_tokens//2).
:- dynamic(logtalk::message_tokens//2).

logtalk::message_tokens(redefining_entity(Type, Entity), core) -->
    ['Redefining ~w ~q'-[Type, Entity], nl].

The following tokens can be used when translating a message:

at_same_line

Signals a following part to a multi-part message with no line break in between; this token is ignored when it’s not the first in the list of tokens

tab(Expression)

Evaluate the argument as an arithmetic expression and write the resulting number of spaces; this token is ignored when the number of spaces is not positive

nl

Change line in the output stream

flush

Flush the output stream (by calling the flush_output/1 standard predicate)

Format-Arguments

Format must be an atom and Arguments must be a list of format arguments (the token arguments are passed to a call to the format/3 de facto standard predicate)

term(Term, Options)

Term can be any term and Options must be a list of valid write_term/3 output options (the token arguments are passed to a call to the write_term/3 standard predicate)

ansi(Attributes, Format, Arguments)

Taken from SWI-Prolog; by default, do nothing; can be used for styled output

begin(Kind, Var)

Taken from SWI-Prolog; by default, do nothing; can be used together with end(Var) to wrap a sequence of message tokens

end(Var)

Taken from SWI-Prolog; by default, do nothing

The logtalk object also defines public predicates for printing a list of tokens, for hooking into printing an individual token, and for setting default output streams and message prefixes. For example, the SWI-Prolog adapter file uses the print message token hook predicate to enable coloring of messages printed on a console.

Meta-messages

Defining tokenization rules for every message is not always necessary, however. Logtalk defines several meta-messages that are handy for simple cases and temporary messages only used during application development, notably debugging messages. See the Debugging messages section and the logtalk built-in object remarks section for details.

Defining message prefixes and output streams

The logtalk::message_prefix_stream/4 hook predicate can be used to define a message line prefix and an output stream for printing messages of a given kind and component. For example:

:- multifile(logtalk::message_prefix_stream/4).
:- dynamic(logtalk::message_prefix_stream/4).

logtalk::message_prefix_stream(comment, my_app, '% ', user_output).
logtalk::message_prefix_stream(warning, my_app, '* ', user_error).

A single clause at most is expected per message kind and component pair. When this predicate is not defined for a given kind and component pair, the following defaults are used:

kind_prefix_stream(banner,         '',       user_output).
kind_prefix_stream(help,           '',       user_output).
kind_prefix_stream(question,       '',       user_output).
kind_prefix_stream(question(_),    '',       user_output).
kind_prefix_stream(information,    '% ',     user_output).
kind_prefix_stream(information(_), '% ',     user_output).
kind_prefix_stream(comment,        '% ',     user_output).
kind_prefix_stream(comment(_),     '% ',     user_output).
kind_prefix_stream(warning,        '*     ', user_error).
kind_prefix_stream(warning(_),     '*     ', user_error).
kind_prefix_stream(error,          '!     ', user_error).
kind_prefix_stream(error(_),       '!     ', user_error).
kind_prefix_stream(debug,          '>>> ',   user_error).
kind_prefix_stream(debug(_),       '>>> ',   user_error).

When the message kind is unknown, information is used instead.

Defining message prefixes and output files

Some applications require copying and saving messages without diverting them from their default stream. For simple cases, this can be accomplished by intercepting the messages using the logtalk::message_hook/4 multifile hook predicate (see next section). In more complex cases, where messages are already intercepted for a different purpose, it can be tricky to use multiple definitions of the message_hook/4 predicate as the order of the clauses of a multiple predicate cannot be assumed in general (for all message_hook/4 predicate definitions to run, all but the last one to be called must fail). Using a single master definition is also not ideal as it would result in strong coupling instead of a clean separation of concerns.

The experimental logtalk::message_prefix_file/6 hook predicate can be used to define a message line prefix and an output file for copying messages of a given kind and component pair. For example:

:- multifile(logtalk::message_prefix_file/6).
:- dynamic(logtalk::message_prefix_file/6).

logtalk::message_prefix_file(error,   app, '! ', 'log.txt', append, []).
logtalk::message_prefix_file(warning, app, '! ', 'log.txt', append, []).

A single clause at most is expected per message kind and component pair.

This predicate is called by default by the message printing mechanism. Definitions of the message_hook/4 hook predicate are free to decide if the logtalk::message_prefix_file/6 predicate should be called and acted upon.

Intercepting messages

Calls to the logtalk::print_message/3 predicate can be intercepted by defining clauses for the logtalk::message_hook/4 multifile hook predicate. This predicate can suppress, rewrite, and divert messages.

As a first example, assume that you want to make Logtalk startup less verbose by suppressing printing of the default compiler flag values. This can be easily accomplished by defining the following category in a settings file:

:- category(my_terse_logtalk_startup_settings).

    :- multifile(logtalk::message_hook/4).
    :- dynamic(logtalk::message_hook/4).

    logtalk::message_hook(default_flags, comment(settings), core, _).

:- end_category.

The printing message mechanism automatically calls the message_hook/4 hook predicate. When this call succeeds, the mechanism assumes that the message has been successfully handled.

As another example, assume that you want to print all otherwise silent compiler messages:

:- category(my_verbose_logtalk_message_settings).

    :- multifile(logtalk::message_hook/4).
    :- dynamic(logtalk::message_hook/4).

    logtalk::message_hook(_Message, silent, core, Tokens) :-
        logtalk::message_prefix_stream(comment, core, Prefix, Stream),
        logtalk::print_message_tokens(Stream, Prefix, Tokens).

    logtalk::message_hook(_Message, silent(Key), core, Tokens) :-
        logtalk::message_prefix_stream(comment(Key), core, Prefix, Stream),
        logtalk::print_message_tokens(Stream, Prefix, Tokens).

:- end_category.

Asking questions

Logtalk structured question-asking mechanism complements the message printing mechanism. It provides an abstraction for the common task of asking a user a question and reading back its reply. By default, this mechanism writes the question, writes a prompt, and reads the answer using the current user input and output streams but allows all steps to be intercepted, filtered, rewritten, and redirected. Two typical examples are using a GUI dialog for asking questions and automatically providing answers to specific questions.

The question-asking mechanism works in tandem with the message printing mechanism, using it to print the question text and a prompt. It provides an asking predicate and a hook predicate, both declared and defined in the logtalk built-in object. The asking predicate, logtalk::ask_question/5, is used for asking a question and reading the answer. Assume that we defined the following message tokenization and question prompt and stream:

:- category(hitchhikers_guide_to_the_galaxy).

    :- multifile(logtalk::message_tokens//2).
    :- dynamic(logtalk::message_tokens//2).

    % abstract the question text using the atom ultimate_question;
    % the second argument, hitchhikers, is the application component
    logtalk::message_tokens(ultimate_question, hitchhikers) -->
        ['The answer to the ultimate question of life, the universe and everything is?'-[], nl].

   :- multifile(logtalk::question_prompt_stream/4).
   :- dynamic(logtalk::question_prompt_stream/4).

   % the prompt is specified here instead of being part of the question text
   % as it will be repeated if the answer doesn't satisfy the question closure
   logtalk::question_prompt_stream(question, hitchhikers, '> ', user_input).

:- end_category.

After compiling and loading this category, we can now ask the ultimate question:

| ?- logtalk::ask_question(question, hitchhikers, ultimate_question, '=='(42), N).

The answer to the ultimate question of life, the universe and everything is?
> 42.

N = 42
yes

Note that the fourth argument, '=='(42) in our example, is a closure that is used to check the answers provided by the user. The question is repeated until the goal constructed by extending the closure with the user answer succeeds. For example:

| ?- logtalk::ask_question(question, hitchhikers, ultimate_question, '=='(42), N).
The answer to the ultimate question of life, the universe and everything is?
> icecream.
> tea.
> 42.

N = 42
yes

Practical usage examples of this mechanism can be found, e.g., in the debugger tool where it’s used to abstract the user interaction when tracing a goal execution in debug mode.

Intercepting questions

Calls to the logtalk::ask_question/5 predicate can be intercepted by defining clauses for the logtalk::question_hook/6 multifile hook predicate. This predicate can suppress, rewrite, and divert questions. For example, assume that we want to automate testing and thus cannot rely on someone manually providing answers:

:- category(hitchhikers_fixed_answers).

    :- multifile(logtalk::question_hook/6).
    :- dynamic(logtalk::question_hook/6).

    logtalk::question_hook(ultimate_question, question, hitchhikers, _, _, 42).

:- end_category.

After compiling and loading this category, trying the question again will now skip asking the user:

| ?- logtalk::ask_question(question, hitchhikers, ultimate_question, '=='(42), N).

N = 42
yes

In a practical case, the fixed answer would be used for follow-up goals being tested. The question-answer read loop (which calls the question check closure) is not used when a fixed answer is provided using the logtalk::question_hook/6 predicate thus preventing the creation of endless loops. For example, the following query succeeds:

| ?- logtalk::ask_question(question, hitchhikers, ultimate_question, '=='(41), N).

N = 42
yes

Note that the logtalk::question_hook/6 predicate takes as argument the closure specified in the logtalk::ask_question/5 call, allowing a fixed answer to be checked before being returned.