Scryer Prolog

Scryer Prolog aims to become to ISO Prolog what GHC is to Haskell: an open source industrial strength production environment that is also a testbed for bleeding edge research in logic and constraint programming, which is itself written in a high-level language.

The homepage of the project is: https://www.scryer.pl

Phase 1

Produce an implementation of the Warren Abstract Machine in Rust, done according to the progression of languages in Warren's Abstract Machine: A Tutorial Reconstruction.

Phase 1 has been completed in that Scryer Prolog implements in some form all of the WAM book, including lists, cuts, Debray allocation, first argument indexing, last call optimization and conjunctive queries.

Phase 2

Extend Scryer Prolog to include the following, among other features:

Phase 3

Use the WAM code produced by the completed code generator to get JIT-compiled and -executed Prolog programs. The question of how to get assembly from WAM code is something I'm still considering.

It's my hope to use Scryer Prolog as the logic engine of a low level (and ideally, very fast) Shen implementation.

Nice to have features

There are no current plans to implement any of these, but they might be nice to have in the future. They'd make a good project for anyone wanting to contribute code to Scryer Prolog.

Implement the global analysis techniques described in Peter van Roy's thesis, "Can Logic Programming Execute as Fast as Imperative Programming?"
Add unum representation and arithmetic, using either an existing unum implementation or an ad hoc one. Unums are described in Gustafson's book "The End of Error."
Add concurrent tables to manage shared references to atoms and strings.
Add some form of JIT predicate indexing.

Installing Scryer Prolog

Native Install

First, install the latest stable version of Rust using your preferred method. Scryer tends to use features from newer Rust releases, whereas Rust packages in Linux distributions, Macports, etc. tend to lag behind. rustup will keep your Rust updated to the latest stable release; any existing Rust distribution should be uninstalled from your system before rustup is used.

Currently the only way to install the latest version of Scryer is to clone directly from this git repository, and compile the system. This can be done as follows:

$> git clone https://github.com/mthom/scryer-prolog
$> cd scryer-prolog
$> cargo build --release

The --release flag performs various optimizations, producing a faster executable.

After compilation, the executable scryer-prolog is available in the directory target/release and can be invoked to run the system.

On Windows, Scryer Prolog is easier to build inside a MSYS2 environment as some crates may require native C compilation. However, the resulting binary does not need MSYS2 to run. When executing Scryer in a shell, it is recommended to use a more advanced shell than mintty (the default MSYS2 shell). The Windows Terminal works correctly.

To build a Windows Installer, you'll need first Scryer Prolog compiled in release mode, then, with WiX Toolset installed, execute:

candle.exe scryer-prolog.wxs
light.exe scryer-prolog.wixobj

It will generate a very basic MSI file which installs the main executable and a shortcut in the Start Menu. It can be installed with a double-click. To uninstall, go to the Control Panel and uninstall as usual.

Scryer Prolog must be built with Rust 1.63 and up.

Docker Install

First, install Docker on Linux, Windows, or Mac.

Once Docker is installed, you can download and run Scryer Prolog with a single command:

$> docker run -it mjt128/scryer-prolog

To consult your Prolog files, bind mount your programs folder as a Docker volume:

$> docker run -v /home/user/prolog:/mnt -it mjt128/scryer-prolog
?- consult('/mnt/program.pl').
true.

This works on Windows too:

$> docker run -v C:\Users\user\Documents\prolog:/mnt -it mjt128/scryer-prolog
?- consult('/mnt/program.pl').
true.

Tutorial

Prolog files are loaded by specifying them as arguments on the command line. For example, to load program.pl, use:

$> scryer-prolog program.pl

Loading a Prolog file is also called “consulting” it. The built-in predicate consult/1 can be used to consult a file from within Prolog:

?- consult('program.pl').

As an abbreviation for consult/1, you can specify a list of program files, given as atoms:

?- ['program.pl'].

The special notation [user] is used to read Prolog text from standard input. For example,

?- [user].
hello(declarative_world).
hello(pure_world).

Pressing RETURN followed by Ctrl-d stops reading from standard input and consults the entered Prolog text.

After a program is consulted, you can ask queries about the predicates it defines. For example, with the program shown above:

?- hello(What).
   What = declarative_world
;  What = pure_world.

Press SPACE to show further answers, if any exist. Press RETURN or . to abort the search and return to the toplevel prompt. Press f to see up to the next multiple of 5 answers, and a to see all answers. Press h to show a help message.

Use TAB to complete atoms and predicate names in queries. For instance, after consulting the program above, typing decl followed by TAB yields declarative_world. Press TAB repeatedly to cycle through alternative completions.

To quit Scryer Prolog, use the standard predicate halt/0:

?- halt.

Dynamic operators

Scryer supports dynamic operators. Using the built-in arithmetic operators with the usual precedences,

?- write_canonical(-5 + 3 - (2 * 4) // 8), nl.
-(+(-5,3),//(*(2,4),8))
   true.

New operators can be defined using the op declaration.

First instantiated argument indexing

Scryer Prolog indexes on the leftmost argument that is not a variable in all clauses of a predicate's definition. We call this strategy first instantiated argument indexing.

A key motivation for first instantiated argument indexing is to enable indexing for meta-predicates such as maplist/N and foldl/N, whose first argument is a partial goal that is a variable in the definition of these predicates and therefore cannot be used for indexing.

For example, a natural definition of maplist/2 reads:

maplist(_, []).
maplist(Goal_1, [L|Ls]) :-
        call(Goal_1, L),
        maplist(Goal_1, Ls).

In this case, first instantiated argument indexing automatically uses the second argument for indexing, and thus prevents choicepoints for calls with lists of fixed lengths (and deterministic goals). Conveniently, no auxiliary predicates with reordered arguments are needed to benefit from indexing in such cases.

Conventional first argument indexing naturally arises as a special case of this strategy, if the first argument is instantiated in any clause of a predicate's definition.

Strings and partial strings

A very compact internal representation of strings is one of the key innovations of Scryer Prolog. This means that terms which appear as lists of characters to Prolog programs are stored in packed UTF-8 encoding by the engine.

Without this innovation, storing a list of characters in memory would use one memory cell per character, one memory cell per list constructor, and one memory cell for each tail that occurs in the list. Since one memory cell takes 8 bytes on 64-bit machines, the packed representation used by Scryer Prolog yields an up to 24-fold reduction of memory usage, and corresponding reduction of memory accesses when creating and processing strings.

Scryer Prolog's compact internal string representation makes it ideally suited for the use case Prolog was originally developed for: efficient and convenient text processing, especially with definite clause grammars (DCGs) as provided by library(dcgs) and library(pio) to transparently apply DCGs to files.

In Scryer Prolog, the default value of the Prolog flag double_quotes is chars, which is also the recommended setting. This means that lists of characters can be written as double-quoted strings, in the tradition of Marseille Prolog.

For example, the following query succeeds:

?- "abc" = [a,b,c].
   true.

This shows that the string "abc", which is represented as a sequence of 3 bytes internally, appears to Prolog programs as a list of characters.

Scryer Prolog uses the same efficient encoding for partial strings, which appear to Prolog code as partial lists of characters. The predicate partial_string/3 from library(iso_ext) lets you construct partial strings explicitly. For example:

?- partial_string("abc", Ls0, Ls).
   Ls0 = [a,b,c|Ls].

In this case, and as the answer illustrates, Ls0 is indistinguishable from a partial list with tail Ls, while the efficient packed representation is used internally.

An important design goal of Scryer Prolog is to automatically use the efficient string representation whenever possible. Therefore, it is only very rarely necessary to use partial_string/3 explicitly. In the above example, posting Ls0 = [a,b,c|Ls] yields the exact same internal representation, and has the advantage that only the standard predicate (=)/2 is used.

The efficient internal representation of strings and partial strings was first proposed and explained by Ulrich Neumerkel in issues #24 and #95, and Scryer Prolog is the first Prolog system that implements it.

Occurs check and cyclic terms

The occurs check is an element of algorithms that perform syntactic unification, causing the unification to fail if a variable is unified with a term that contains that variable as a proper subterm. For efficiency, the occurs check is omitted by default in Scryer Prolog and many other Prolog systems.

In Scryer Prolog, performing unifications which succeed only if the occurs check is omitted yield cyclic terms, also called rational trees. For example:

?- X = f(X), Y = g(X,Y).
   X = f(X), Y = g(f(X),Y).

The creation of cyclic terms often indicates a programming mistake in the formulation of Prolog predicates, and to obtain logically sound results it is desirable to either perform all unifications with occurs check enabled, or let Prolog throw an error if enabling the occurs check is necessary to prevent a unification.

Scryer Prolog supports this via the Prolog flag occurs_check. It can be set to one of the following values to obtain the desired behaviour:

false Do not perform the occurs check. This is the default.
true Perform all unifications with the occurs check enabled.
error Yield an error if a unification is performed that the occurs check would have prevented.

Especially when starting with Prolog, we recommend to add the following directive to the ~/.scryerrc configuration file so that programming mistakes in predicates that lead to the creation of cyclic terms are indicated by errors:

:- set_prolog_flag(occurs_check, error).

Scryer Prolog implements specialized reasoning to make unifications fast in many frequently occurring situations also if the occurs check is enabled.

Tabling (SLG resolution)

One of the foremost attractions of Prolog is that logical consequences of pure programs can be derived by various execution strategies that differ regarding essential properties such as termination, completeness and efficiency.

The default execution strategy of Prolog is depth-first search with chronological backtracking. This strategy is very efficient. Its main drawback is that it is incomplete: It may fail to find any solution even if one exists.

Scryer Prolog supports an alternative execution strategy which is called tabling and also known as tabled execution and SLG resolution. To enable tabled execution for a predicate, use library(tabling) and add a (table)/1 directive for the desired predicate indicator. For example, if we write:

:- use_module(library(tabling)).
:- table a/0.

a :- a.

Then the query ?- a. terminates (and fails), whereas it does not terminate with the default execution strategy.

Scryer Prolog implements tabling via delimited continuations as described in Tabling as a Library with Delimited Control by Desouter et. al.

Constraint Logic Programming (CLP)

Scryer Prolog provides excellent support for Constraint Logic Programming (CLP), which is the amalgamation of Logic Programming (LP) and Constraints.

In addition to built-in support for dif/2, freeze/2, CLP(B) and CLP(ℤ), Scryer provides a convenient way to implement new user-defined constraints: Attributed variables are available via library(atts) as in SICStus Prolog, which is one of the most sophisticated and fastest constraint systems in existence. In library(iso_ext), Scryer provides predicates for backtrackable (bb_b_put/2) and non-backtrackable (bb_put/2) global variables, which are needed to implement certain types of constraint solvers.

These features make Scryer Prolog an ideal platform for teaching, learning and developing portable CLP applications.

Modules

Scryer has a simple predicate-based module system. It provides a way to separate units of code into distinct namespaces, for both predicates and operators. See the files src/lib/*.pl for examples.

At the time of this writing, many predicates reside in their own modules that need to be imported before they can be used. The modules that ship with Scryer Prolog are also called library modules or libraries, and include:

lists providing length/2, member/2, select/3, append/[2,3], foldl/[4,5], maplist/[2-9], same_length/2, transpose/2 etc.
dcgs Definite Clause Grammars (DCGs), a built-in grammar mechanism that uses the operator (-->)/2 to define grammar rules, and the predicates phrase/[2,3] to invoke them.
dif The predicate dif/2 provides declarative disequality: It is true if and only if its arguments are different, and delays the test until a sound decision can be made.
reif providing if_/3, tfilter/3 and related predicates as described in Indexing dif/2.
clpz CLP(ℤ): Constraint Logic Programming over Integers, providing declarative integer arithmetic via (#=)/2, (#\=)/2, (#>=)/2 etc., and various global constraints and enumeration predicates for solving combinatorial tasks.
pairs By convention, pairs are Prolog terms with principal functor (-)/2, written as Key-Value. This library provides pairs_keys_values/3, pairs_keys/2, and other predicates to reason about pairs.
si The predicates atom_si/1, integer_si/1, atomic_si/1 and list_si/1 implement sound type checks. They raise instantiation errors if no decision can be made. They are declarative replacements for logically flawed lower-level type tests. For instance, instead of integer(X), write integer_si(X) to ensure soundness of your programs. "si" stands for sufficiently instantiated, and also for sound inference.
debug Various predicates that allow for declarative debugging.
pio phrase_from_file/2 applies a DCG nonterminal to the contents of a file, reading lazily only as much as is needed. Due to the compact internal string representation, also extremely large files can be efficiently processed with Scryer Prolog in this way. phrase_to_file/2 and phrase_to_stream/2 write lists of characters described by DCGs to files and streams, respectively.
lambda Lambda expressions to simplify higher order programming.
charsio Various predicates that are useful for parsing and reasoning about characters, notably char_type/2 to classify characters according to their type, and conversion predicates for different encodings of strings.
error must_be/2 and can_be/2 complement the type checks provided by library(si), and are especially useful for Prolog library authors.
tabling The operator (table)/1 is used in directives that prepare predicates for tabled execution (SLG resolution).
format The nonterminal format_//2 is used to describe formatted output, arranging arguments according to a given format string. The predicates format/[2,3], portray_clause/[1,2] and listing/1 provide formatted impure output.
assoc providing empty_assoc/1, get_assoc/3, put_assoc/4 etc. to manage elements in AVL trees which ensure O(log(N)) access.
ordsets represents ordered sets as lists.
clpb CLP(B): Constraint Logic Programming over Boolean variables, a BDD-based SAT solver provided via the predicates sat/1, taut/2, labeling/1 etc.
arithmetic Arithmetic predicates such as lsb/2, msb/2 and number_to_rational/2.
time Predicates for reasoning about time, including time/1 to measure the CPU time of a goal, current_time/1 to obtain the current system time, the nonterminal format_time//2 to describe strings with dates and times, and sleep/1 to slow down a computation.
files Predicates for reasoning about files and directories, such as directory_files/2, file_exists/1 and file_size/2.
cont Provides delimited continuations via reset/3 and shift/1.
random Probabilistic predicates and random number generators.
http/http_open Open a stream to read answers from web servers. HTTPS is also supported.
http/http_server Runs a HTTP/1.1 and HTTP/2.0 web server. Uses Hyper as a backend. Supports some query and form handling.
sgml load_html/3 and load_xml/3 represent HTML and XML documents as Prolog terms for convenient and efficient reasoning. Use library(xpath) to extract information from parsed documents.
csv parse_csv//1 and parse_csv//2 can be used with phrase_from_file/2 or phrase/2 to parse csv
serialization/abnf DCGs describing the ABNF grammar core (RFC 5234), which is used to describe many IETF syntaxes, such as HTTP v1.1, SMTP, iCalendar, and more.
serialization/json json_chars//1 can be used with phrase_from_file/2 or phrase/2 to parse and generate JSON.
xpath The predicate xpath/3 is used for convenient reasoning about HTML and XML documents, inspired by the XPath language. This library is often used together with library(sgml).
sockets Predicates for opening and accepting TCP connections as streams.
os Predicates for reasoning about environment variables.
iso_ext Conforming extensions to and candidates for inclusion in the Prolog ISO standard, such as setup_call_cleanup/3, call_nth/2 and call_with_inference_limit/3.
crypto Cryptographically secure random numbers and hashes, HMAC-based key derivation (HKDF), password-based key derivation (PBKDF2), public key signatures and signature verification with Ed25519, ECDH key exchange over Curve25519 (X25519), authenticated symmetric encryption with ChaCha20-Poly1305, and reasoning about elliptic curves.
uuid UUIDv4 generation and hex representation
tls Predicates for negotiating TLS connections explicitly.
ugraphs Graph manipulation library
simplex Providing assignment/2, transportation/4 and other predicates for solving linear programming problems.

To use predicates provided by the lists library, write:

?- use_module(library(lists)).

To load modules contained in files, the library functor can be omitted, prompting Scryer to search for the file (specified as an atom) from its working directory:

?- use_module('file.pl').

use_module directives can be qualified by adding a list of imports:

?- use_module(library(lists), [member/2]).

A qualified use_module can be used to remove imports from the toplevel by calling it with an empty import list.

The (:)/2 operator resolves calls to predicates that might not be imported to the current working namespace:

?- lists:member(X, Xs).

The [user] prompt can also be used to define modules inline at the REPL:

?- [user].
:- module(test, [local_member/2]).
:- use_module(library(lists)).

local_member(X, Xs) :- member(X, Xs).

The user listing can also be terminated by placing end_of_file. at the end of the stream.

Configuration file

At startup, Scryer Prolog consults the file ~/.scryerrc, if the file exists. This file is useful to automatically load libraries and define predicates that you need often.

For example, a sensible starting point for ~/.scryerrc is:

:- use_module(library(lists)).
:- use_module(library(dcgs)).
:- use_module(library(reif)).

Development environment

To write and edit Prolog programs, we recommend GNU Emacs with the Prolog mode maintained by Stefan Bruda.

Use ediprolog to consult Prolog code and evaluate Prolog queries in arbitrary Emacs buffers.

Emacs definitions that show Prolog terms as trees are available in tools.

To debug Prolog code, we recommend the predicates from library(debug), most notably:

(*)/1 to "generalize away" a Prolog goal. Use it to debug unexpected failures by generalizing your definitions until they succeed. Simply place * in front of a goal to generalize it away.
($)/1 to emit a trace of the execution, showing when a goal is invoked, and when it has succeeded. Place $ in front of a goal to emit this information for that goal.

This way of debugging Prolog code has several major benefits, such as: It stays close to the actual Prolog code under consideration, it does not need additional tools and formalisms for its application, and further, it encourages declarative reasoning that can in principle also be performed automatically.

Support and discussions

If Scryer Prolog crashes or yields unexpected errors, consider filing an issue.

To get in touch with the Scryer Prolog community, participate in discussions or visit our #scryer IRC channel on Libera!

mthom/scryer-prolog

mthom

Reviews

Repository Details