- Reference manual
- Built-in Predicates
- Notation of Predicate Descriptions
- Character representation
- Loading Prolog source files
- Editor Interface
- Verify Type of a Term
- Comparison and Unification of Terms
- Control Predicates
- Meta-Call Predicates
- Delimited continuations
- Exception handling
- Printing messages
- Handling signals
- DCG Grammar rules
- Declaring predicate properties
- Examining the program
- Input and output
- Status of streams
- Primitive character I/O
- Term reading and writing
- Analysing and Constructing Terms
- Analysing and Constructing Atoms
- Localization (locale) support
- Character properties
- Character Conversion
- Misc arithmetic support predicates
- Built-in list operations
- Finding all Solutions to a Goal
- Formatted Write
- Global variables
- Terminal Control
- Operating System Interaction
- File System Interaction
- User Top-level Manipulation
- Creating a Protocol of the User Interaction
- Debugging and Tracing Programs
- Obtaining Runtime Statistics
- Execution profiling
- Memory Management
- Windows DDE interface
- Built-in Predicates
- Reference manual
- [ISO]catch(:Goal, +Catcher, :Recover)
- Behaves as call/1
if no exception is raised when executing Goal. If an
exception is raised using throw/1
while Goal executes, and the Goal is the innermost
goal for which Catcher unifies with the argument of throw/1,
all choice points generated by Goal are cut, the system
backtracks to the start of catch/3
while preserving the thrown exception term, and Recover is
called as in call/1.
The overhead of calling a goal through catch/3 is comparable to call/1. Recovery from an exception is much slower, especially if the exception term is large due to the copying thereof or is decorated with a stack trace using, e.g., the library
library(prolog_stack)based on the prolog_exception_hook/4 hook predicate to rewrite exceptions.
- Raise an exception. The system looks for the innermost catch/3
ancestor for which Exception unifies with the Catcher
argument of the catch/3
call. See catch/3
ISO demands that throw/1 make a copy of Exception, walk up the stack to a catch/3 call, backtrack and try to unify the copy of Exception with Catcher. SWI-Prolog delays backtracking until it actually finds a matching catch/3 goal. The advantage is that we can start the debugger at the first possible location while preserving the entire exception context if there is no matching catch/3 goal. This approach can lead to different behaviour if Goal and Catcher of catch/3 call shared variables. We assume this to be highly unlikely and could not think of a scenario where this is useful.76I'd like to acknowledge Bart Demoen for his clarifications on these matters.
In addition to explicit calls to throw/1, many built-in predicates throw exceptions directly from C. If the Exception term cannot be copied due to lack of stack space, the following actions are tried in order:
- If the exception is of the form
error(Formal, ImplementationDefined), try to raise the exception without the ImplementationDefined part.
- Try to raise
- Abort (see abort/0).
- If the exception is of the form
- catch_with_backtrace(:Goal, +Catcher, :Recover)
- As catch/3,
but if library
library(prolog_stack)is loaded and an exception of the shape
error(Format, Context)is raised Context is extended with a backtrace. To catch an error and print its message including a backtrace, use the following template:
:- use_module(library(prolog_stack)). ..., catch_with_backtrace(Goal, Error, print_message(error, Error)), ...,
This is good practice for a catch-all wrapper around an application. See also main/0 from library
Under some conditions an exception may be raised as a result of handling another exception. Below are some of the scenarios:
- The predicate setup_call_cleanup/3 calls the cleanup handler as a result of an exception and the cleanup handler raises an exception itself. In this case the most urgent exception is propagated into the environment.
- Raising an exception fails due to lack of resources, e.g., lack of
stack space to store the exception. In this case a resource exception is
raised. If that too fails the system tries to raise a resource exception
without (stack) context. If that fails it will raise the exception
'$aborted', also raised by abort/0. As no stack space is required for processing this atomic exception, this should always succeed.
- Certain callback operations raise an exception while processing another exception or a previous callback already raised an exception before there was an opportunity to process the exception. The most notable callback subject to this issue are prolog_event_hook/1 (supporting e.g., the graphical debugger), prolog_exception_hook/4 (rewriting exceptions, e.g., by adding context) and print_message/2 when called from the core facilities such as the internal debugger. As with setup_call_cleanup/3, the most urgent exception is preserved.
If the most urgent exceptions needs to be preserved, the following exception ordering is respected, preserving the topmost matching error.
- All other exceptions
Note The above resolution is not described in the ISO
standard. This is not needed either because ISO does not specify
and does not deal with environment management issues such as (debugger)
callbacks. Neither does it define abort/0
or timeout handling. Notably abort/0
and timeout are non-logical control structures. They are implemented on
top of exceptions as they need to unwind the stack, destroy choice
points and call cleanup handlers in the same way. However, the pending
exception should not be replaced by another one before the intended
handler is reached. The abort exception cannot be caught, something
which is achieved by wrapping the cleanup handler of catch/3
Before the introduction of exceptions in SWI-Prolog a runtime error was handled by printing an error message, after which the predicate failed. If the Prolog flag debug_on_error was in effect (default), the tracer was switched on. The combination of the error message and trace information is generally sufficient to locate the error.
With exception handling, things are different. A programmer may wish to trap an exception using catch/3 to avoid it reaching the user. If the exception is not handled by user code, the interactive top level will trap it to prevent termination.
If we do not take special precautions, the context information associated with an unexpected exception (i.e., a programming error) is lost. Therefore, if an exception is raised which is not caught using catch/3 and the top level is running, the error will be printed, and the system will enter trace mode.
If the system is in a non-interactive call-back from foreign code and there is no catch/3 active in the current context, it cannot determine whether or not the exception will be caught by the external routine calling Prolog. It will then base its behaviour on the Prolog flag debug_on_error:
- current_prolog_flag(debug_on_error, false)
The exception does not trap the debugger and is returned to the foreign routine calling Prolog, where it can be accessed using PL_exception(). This is the default.
- current_prolog_flag(debug_on_error, true)
If the exception is not caught by Prolog in the current context, it will trap the tracer to help analyse the context of the error.
While looking for the context in which an exception takes place, it
is advised to switch on debug mode using the predicate debug/0.
can be used to add more debugging facilities to exceptions. An example
is the library
library(http/http_error), generating a full
stack trace on errors in the HTTP server library.
The predicate throw/1
takes a single argument, the exception term, and the ISO
standard stipulates that the exception term be of the form
the‘formal' description of the error, as listed in chapter 7.12.2 pp. 62-63 ("Error classification") of the ISO standard. It indicates the error class and possibly relevant error context information. It may be a compound term of arity 1,2 or 3 - or simply an atom if there is no relevant error context information.
additional context information beyond the one in Formal. If may be unset, i.e. a fresh variable, or set to something that hopefully will help the programmer in debugging. The structure of Context is left unspecified by the ISO Standard, so SWI-Prolog creates it own convention (see below).
Thus, constructing an error term and throwing it might take this form (although you would not use the illustrative explicit naming given here; instead composing the exception term directly in a one-liner):
Exception = error(Formal, Context), Context = ... some local convention ..., Formal = type_error(ValidType, Culprit), % for "type error" for example ValidType = integer, % valid atoms are listed in the ISO standard Culprit = ... some value ..., throw(Exception)
Note that the ISO standard formal term expresses what should be the case or what is the expected correct state, and not what is the problem. For example:
- If a variable is found to be uninstantiated but should be
instantiated, the error term is
instantiation_error: The problem is not that there is an unwanted instantiation, but that the correct state is the one with an instantiated variable.
- In case a variable is found to be instantiated but should be
uninstantiated (because it will be used for output), the error term is
uninstantiation_error(Culprit): The problem is not that there is lack of instantiation, but that the correct state is the one which Culprit (or one of its subterms) is more uninstantiated than is the case.
- If you try to disassemble an empty list with compound_name_arguments/3,
the error term is
type_error(compound,). The problem is not that
is (erroneously) a compound term, but that a compound term is expected and
does not belong to that class.
User predicates are free to choose the structure of their exception terms (i.e., they can define their own conventions) but should adhere to the ISO standard if possible, in particular for libraries.
Notably, exceptions of the shape
are recognised by the development tools and therefore expressing
unexpected situations using these exceptions improves the debugging
In SWI-Prolog, the second argument of the exception term, i.e., the
Context argument, is generally of the form
context(Location, Message), where:
describes the execution context in which the exception occurred. While the Location argument may be specified as a predicate indicator (Name/Arity), it is typically filled by the
library(prolog_stack)library. This library recognises uncaught errors or errors caught by catch_with_backtrace/3 and fills the Location argument with a backtrace.
provides an additional description of the error or can be left as a fresh variable if there is nothing appropriate to fill in.
ISO standard exceptions can be thrown via the predicates exported
library(error). Termwise, these predicates look
exactly like the
Formal of the ISO standard error term they throw:
- instantiation_error/1 (the argument is not used: ISO specifies no argument)
- existence_error/3 (a SWI-Prolog extension that is not ISO)