A compact, efficient Prolog interpreter with ISO compliant aspirations.
MIT licensed
Integers are unbounded
Atoms are UTF-8 of unlimited length
The default double-quoted representation is *chars* list
Dynamic atoms are automatically garbage collected
Unlimited arity (system resources constrained)
Uses 1st & 2nd arg indexing
Atom garbage collection
DCGs
REPL with history
Compiles in <1s with *tcc*, or ~5s with *gcc* and *clang*
Runs on Linux, FreeBSD and macOS
Foreign function interface (FFI) for calling out to user C code
Access SQLITE databases using builtin module (uses FFI)
Attributed variables with SICStus interface (*WIP*)
Rational trees aka. cyclic terms (*WIP*)
Logtalk compatible (*WIP*)
Trealla is not WAM-based. It uses tree-walking, structure-sharing and deep-binding. Source is compiled to an AST that is interpreted at runtime.
The name Trealla comes from the Liaden Universe books by Lee & Miller. It is also a nod to the Trealla region of Western Australia.
tpl [options] [files] [-- args]
where options can be:
-O0, --noopt - no optimization
-f file - load file (*~/.tplrc* not loaded)
-l file - load file (*~/.tplrc* loaded)
file - load file (*~/.tplrc* loaded)
-g goal - query goal (only used once)
--library path - alt to TPL_LIBRARY_PATH env variable
-t, --trace - trace
-q, --quiet - quiet mode (no banner)
-v, --version - version
-h, --help - help
-d, --daemonize - daemonize
-w, --watchdog - create watchdog
--consult - consult from STDIN
For example:
tpl -g test2,halt samples/sieve
Invocation without any goal presents the REPL.
The default path to the library is relative to the executable location.
The file ~/.tplrc is consulted on startup unless the -f option is present.
When consulting, reconsulting and deconsulting files the .pl version of the filename is always preferred (if not specified) when looking for a file.
To run the Pereira benchmark suite:
tpl -g "bench_peirera,halt" -f samples/peirera.pl
Trealla uses UTF-8 internally and this works well with modern operating systems that are already [1], or moving to [2], native UTF-8.
It aligns well with standard C as functions like strcmp/memcmp that require no special handling to respect codepoint order. This also works seamlessly with the implementation of double-quoted strings (ie. chars-list), DCGs, and mmap'd files. Any code-point specific requirements, like get_char, get_code, sub_atom, atom_length, atom_codes, atom_chars & _upper/_lower are handled on the fly.
Unicode atoms do not need to be quoted unless they contain breaking characters...
?- [user].
是. % be: means, approximately, "True".
不是 :- \+ 是. % not be: means, approximately, "False".
<CTRL-D>
true.
?- 是.
true.
?- 不是.
false.
Trealla accepts as a variable any atom beginning with Unicode uppercase...
?- atom_upper(δ,C).
C = Δ.
?- Δ is 123456-123455.
Δ = 1.
?-
Written in plain-old C.
git clone https://github.com/infradig/trealla.git
cd trealla
On Debian+ systems you may need to install GNU readline, xxd & libffi
sudo apt install libreadline-dev xxd libffi-dev
Then...
make
To build without libffi:
make NOFFI=1
Other systems may vary. On Debian+ systems you may need to install OpenSSL:
sudo apt install libssl-dev
To build without OpenSSL:
make NOSSL=1
To build with the included ISOCLINE sources (default is to use GNU readline):
make ISOCLINE=1
Then...
make test
On BSD systems use gmake to build and do
pkg install editors/vim
to get the xxd utility.
For unbounded arithmetic Trealla uses a modified fork of the imath library, which is partially included in the source. Note, unbounded integers (aka. bigints) are for arithmetic purposes only and will give a type_error when used in places not expected. The imath library has a bug whereby printing large numbers becomes exponentially slower (100K+ digits) and will require a switch to libtomath at some point to remedy.
Contributions are welcome. Trealla coding style is snake-case (like original K&R). I consider camelCase to be an anti-pattern, probably because i'm dyslexic and it takes me twice as long to read and 4 times as long to write.
To cross-compile on Linux and produce a Windows executable...
sudo apt-get install mingw-w64
make CC=x86_64-w64-mingw32-gcc NOSSL=1 NOFFI=1 ISOCLINE=1
$ file tpl.exe
tpl.exe: PE32+ executable (console) x86-64, for MS Windows
$ wine tpl.exe -g test5,halt -f samples/sieve.pl
This project started in March 2020 and it would not be where it is today without help from these people:
Special thanks to Xin Wang for providing the testing framework, for the initial push to get serious and for being the first to take this work (in it's nascent form) seriously.
Special thanks to Paulo Moura for his patience and sleuthing in the quest for Trealla to run his Logtalk project.
Special thanks to Markus Triska for driving the use of packed UTF-8 strings for character-lists. For the idea of mmap()-ing files as strings. For his rigorous approach to types and for bug-checking. Also for use of his format_//2 library.
Special thanks to Jos De Roo for his testing against some classic Prolog examples and his EYE project.
Special thanks to Christian Thaeter for his work with code cleanup and development ideas.
Special thanks to Ulrich Neumerkel for his DCG reference library, for his drive towards ISO standardization and for being himself.
Double-quoted strings, when set_prolog_flag(double_quotes,chars) is set (which is the default) are stored as packed UTF-8 byte arrays. This is compact and efficient. Such strings emulate a list representation and from the programmer point of view are very much indistinguishable from lists.
A good use of such strings is open(filename,read,Str,[mmap(Ls)) which gives a memory-mapped view of a file as a string Ls. List operations on files are now essentially zero-overhead! DCG applications will gain greatly (phrase_from_file/[2-3] uses this).
Both strings and atoms make use of low-overhead reflist-counted byte slices where appropriate.
between/3
forall/2
samsort/2 # same as msort/2
msort/2
merge/3
format/[1-3]
predicate_property/2
numbervars/[1,3-4]
e/0
name/2
tab/[1,2]
maplist/[2-8] # auto-loaded from library(apply)
foldl/[4-7] # auto-loaded from library(apply)
include/3 # auto-loaded from library(apply)
exclude/3 # auto-loaded from library(apply)
get_unbuffered_code/1 # read a single unbuffered code
get_unbuffered_char/1 # read a single unbuffered character
read_term_from_atom/3 # read_term_from_atom(+atom,?term)
write_term_to_atom/3 # write_term_to_atom(?atom,?term,+list)
write_canonical_to_atom/3 # write_canonical_to_atom(?atom,?term,+list)
term_to_atom/2 # term_to_atom(?atom,?term)
setrand/1 # set_seed(+integer) set random number seed
srandom/1 # set_seed(+integer) set random number seed
set_seed/1 # set_seed(+integer) set random number seed
get_seed/1 # get_seed(-integer) get random number seed
rand/1 # rand(-integer) integer [0,RAND_MAX]
random/1 # random(-float) float [0.0,<1.0]
random_between/3 # random_between(+int,+int,-int) integer [arg1,<arg2]
random_float/0 # function returning float [0.0,<1.0]
random_integer/0 # function returning integer [0,RAND_MAX]
rand/0 # function returning integer [0,RAND_MAX]
bb_get/2 # SICStus-compatible
bb_put/2 # SICStus-compatible
bb_delete/2 # SICStus-compatible
bb_update/3 # SICStus-compatible
bb_del/1 # SICStus-compatible
bb_b_put/2 # SICStus-compatible
bb_b_del/1 # SICStus-compatible
b_getval/2 # SWI-compatible
b_setval/2 # SWI-compatible
b_setval0/2 # SWI-compatible
b_delete/1 # SWI-compatible
put_atts/2 # SICStus-compatible
get_atts/2 # SICStus-compatible
del_atts/1 # SICStus-compatible
put_attr/3 # SWI-compatible
get_attr/3 # SWI-compatible
del_attr/2 # SWI-compatible
freeze/2 # auto-loaded from library(freeze)
frozen/2 # auto-loaded from library(freeze)
when/2 # auto-loaded from library(when)
dif/2 # auto-loaded from library(dif)
must_be/4 # must_be(+term,+type,+goal,?arg)
can_be/4 # can_be(+term,+type,+goal,?arg)
must_be/2 # must_be(+type,+term)
can_be/2 # can_be(+type,+term)
expand_term/2 # expand_term(+rule,-Term)
memberchk/2 # memberchk(+rule,+list).
nonmember/2 # \+ memberchk(+rule,+list)
atomic_concat/3 # atomic_concat(+atom,+list,-list)
atomic_list_concat/2 # atomic_list_concat(L,Atom)
atomic_list_concat/3 # atomic_list_concat(L,Sep,Atom)
read_term_from_chars/2 # read_term_from_chars(+chars,?term)
read_term_from_chars/3 # read_term_from_chars(+chars,?term,+list)
write_term_to_chars/3 # write_term_to_chars(?chars,?term,+list)
write_canonical_to_chars/3 # write_canonical_to_chars(?chars,?term,+list)
chars_base64/3 # currently options are ignored
chars_urlenc/3 # currently options are ignored
hex_chars/2 # as number_chars, but in hex
octal_chars/2 # as number_chars, but in octal
partial_string/2 # partial_string(+string,-String)
partial_string/3 # partial_string(+string,-String,-Var)
if/3, (*->)/2 # soft-cut
setup_call_cleanup/3 # setup_call_cleanup(+setup,+call,+cleanup)
call_cleanup/2 # call_cleanup(+call,+cleanup)
term_attvars/2 # term_attvars(+term,-Vs)
copy_term_nat/2 # doesn't copy attrs
copy_term/3 # copy_term(+term1,-term2,-Goals)
unifiable/3 # unifiable(+term1,+term2,-Goals)
?=/2 # ?=(+term1,+term2)
term_expansion/2
goal_expansion/2
cyclic_term/1
term_singletons/2
findall/4
sort/4
var_number/2
ignore/1
is_list/1
is_partial_list/1
is_list_or_partial_list/1
is_stream/1
term_hash/2
term_hash/3 # ignores arg2 (options)
time/1
inf/0
nan/0
\uXXXX and \UXXXXXXXX # quoted character escapes
gcd/2
char_type/2
code_type/2
uuid/1 # generates non-standard UUID
load_files/[1,2]
split_atom/4
plus/3
module/1
line_count/2
strip_module/3
atom_number/2
call_with_time_limit/2 # SWI-compatible
time_out/3 # SICStus-compatible
nb_setval(K,V)
nb_getval(K,V)
nb_delete(K)
nb_current(K,V)
b_setval(K,V)
b_getval(K,V)
b_delete(K)
'$kv_set'(+key,+value,[create(Bool)]) # use atomics only
'$kv_get'(+key,?value,[delete(Bool)]) # use atomics only
call_nth/2
offset/2
limit/2
getenv/2
setenv/2
unsetenv/1
directory_files/2
delete_file/1
exists_file/1 # also file_exists/1
rename_file/2
copy_file/2
time_file/2
size_file/2
exists_directory/1 # also directory_exists/1
make_directory/1
make_directory_path/1
working_directory/2
chdir/1
absolute_file_name/[2,3] # expand(Bool) & relative_to(file) options
access_file/2
current_key/1
recorda/2-3
recordz/2-3
recorded/2-3
instance/2
asserta/2
assertz/2
clause/3
erase/1
string_upper/2
string_lower/2
atom_upper/2
atom_lower/2
popcount/1 # function returning number of 1 bits
lsb/1 # function returning the least significant bit of a positive integer (count from zero)
msb/1 # function returning the most significant bit of a positive integer (count from zero)
log10/1 # function returning log10 of arg
now/0 # function returning C-time in secs as integer
now/1 # now (-integer) C-time in secs as integer
get_time/1 # get_time(-Var) elapsed wall time in secs as float
cpu_time/1 # cpu_time(-Var) elapsed CPU time in secs as float
sleep/1 # sleep time in secs
delay/1 # sleep time for ms
split/4 # split(+string,+sep,?left,?right)
pid/1
shell/1
shell/2
wall_time/1
date_time/6
date_time/7
loadfile/2 # loadfile(+filename,-string)
savefile/2 # savefile(+filename,+string)
getfile/2 # getfile(+filename,-strings)
getline/1 # getline(-string)
getline/2 # getline(+stream,-string)
getlines/1 # getlines(-strings)
getlines/2 # getlines(+stream,-strings)
read_line_to_codes/2 # removes terminator
read_line_to_string/2 # removes terminator
read_file_to_string/3
bread/3 # bread(+stream,?len,-string)
bwrite/2 # bwrite(+stream,+string)
replace/4 # replace(+string,+old,+new,-string)
persist/1 # directive 'persist funct/arity'
open(stream(Str),...) # with open/4 reopen a stream
open(F,M,S,[mmap(Ls)]) # with open/4 mmap() the file to Ls
Note: consult/1 and load_files/2 support lists of files as args. Also support loading into modules eg. consult(MOD:FILE-SPEC).
popen/3 # popen(+cmd,+mode,-stream)
popen/4 # popen(+cmd,+mode,-stream,+opts)
Note: popen/[3,4] use the Unix popen() system call:
tpl -g "use_module(library(apply)),popen('ps -a',read,S,[]),getlines(S,Ls),close(S),maplist(print,Ls),halt"
PID TTY TIME CMD
2806 tty2 00:00:00 gnome-session-b
31645 pts/0 00:00:00 tpl
31646 pts/0 00:00:00 sh
31647 pts/0 00:00:00 ps
Note: read_term/[2,3] supports the positions(Start,End) and the line_counts(Start,End) property options to report file information. This is analogous to stream_property/2 use of position(Pos) and line_count(Line) options.
Uses Ulrich Neumerkel's standard reference library. DCG rules are translated automatically as this library is auto-included.
:- use_module(library(dcgs)).
Hash a plain-text data string to a hexadecimal byte string representing the cryptographic strength hashed value. The options are algorithm(Name) where Name can be sha256, sha384 or sha512 at the moment. If it is a variable it will be unified with the default sha256 algorithm. This predicate is only available when compiled with OpenSSL...
crypto_data_hash/3 # crypto_data_hash(+data,-hash,+options)
Convert a hexadecimal string to a byte-list. At least one arg must be instantiated...
hex_bytes/2 # hex_bytes(?hash,?bytes)
:- use_module(library(http)).
http_get/3 # http_get(Url, Data, Opts)
http_post/4 # http_post(Url, Data, Opts)
http_patch/4 # http_patch(Url, Data, Opts)
http_put/4 # http_put(Url, Data, Opts)
http_delete/3 # http_delete(Url, Data, Opts)
http_server/2 # http_server(Goal,Opts),
http_request/5 # http_request(S, Method, Path, Ver, Hdrs)
A server Goal takes a single arg, the connection stream.
server/2 # server(+host,-stream)
server/3 # server(+host,-stream,+list)
accept/2 # accept(+stream,-stream)
client/4 # client(+url,-host,-path,-stream)
client/5 # client(+url,-host,-path,-stream,+list)
The options list can include udp(bool) (default is false), nodelay(bool) (default is true), ssl(bool) (default is false) and certfile(filespec).
The additional server options can include keyfile(filespec) and certfile(filespec). If just one concatenated file is supplied, use keyfile(filespec) only.
The optional schemes 'unix://', 'http://' (the default) and 'https://' can be provided in the client URL.
With bread/3 the 'len' arg can be an integer > 0 meaning return that many bytes, = 0 meaning return what is there (if non-blocking) or a variable meaning return all bytes until end end of file,
Allows the loading of dynamic libraries and calling of foreign functions written in C from within Prolog...
'$dlopen'/3 # '$dlopen(+name, +flag, -handle)
These predicates register a foreign function as a builtin and use a wrapper to validate arg types at call/runtime...
'$register_function'/4 # '$ffi_reg'(+handle,+symbol,+types,+ret_type)
'$register_predicate'/4 # '$ffi_reg'(+handle,+symbol,+types,+ret_type)
The allowed types are int8, int16, int32, int64, uint8, uint16, uint32, uint64, fp32, fp64, cstr, const_cstr and ptr (for arbitrary pointers/handles).
Assuming the following C-code in samples/foo.c:
double foo(double x, int64_t y)
{
return pow(x, (double)y);
}
int bar(double x, int64_t y, double *result)
{
*result = pow(x, (double)y);
return 0;
}
char *baz(const char *x, const char *y)
{
char *s = malloc(strlen(x) + strlen(y) + 1);
strcpy(s, x);
strcat(s, y);
return s;
}
$ gcc -fPIC -c foo.c
$ gcc -shared -o libfoo.so foo.o
Register a builtin function...
?- '$dlopen'('samples/libfoo.so', 0, H),
'$register_function'(H, foo, [fp64, int64], fp64).
H = 94051868794416.
?- R is foo(2.0, 3).
R = 8.0.
?- R is foo(abc,3).
error(type_error(float,abc),foo/2).
Register a builtin predicate...
?- '$dlopen'('samples/libfoo.so', 0, H),
'$register_predicate'(H, bar, [fp64, int64, -fp64], int64),
'$register_predicate'(H, baz, [cstr, cstr], cstr),
H = 94051868794416.
?- bar(2.0, 3, X, Return).
X = 8.0, Return = 0.
?- baz('abc', '123', Return).
Return = abc123.
Note: the foreign function return value is passed as an extra argument to the predicate call.
There is an example using SQLITE. First make sure SQLITE is installed on your system, for example...
$ sudo apt install sqlite3
Then, given the code in samples/sqlite3.pl...
:- use_module(library(sqlite3)).
run :-
test('samples/sqlite3.db', 'SELECT * FROM company').
test(Database, Query) :-
flag('SQLITE_OK', SQLITE_OK),
sqlite3_open(Database, Connection, Ret), Ret =:= SQLITE_OK,
bagof(Row, sqlite3_query(Connection, Query, Row, _), Results),
writeq(Results), nl.
Run...
$ tpl -g run,halt samples/sqlite3.pl
[[1,'Paul',32,'California',20000.0],[2,'Allen',25,'Texas',15000.0],[3,'Teddy',23,'Norway',20000.0],[4,'Mark',25,'Rich-Mond ',65000.0],[5,'David',27,'Texas',85000.0],[6,'Kim',22,'South-Hall',45000.0]]
Declaring something dynamic with the persist directive:
:- persist :predindicator
causes that clause to be saved to a per-module database on update (asserta/assertz/retract). Maybe this should be an option to dynamic/2?
Trealla is single-threaded internally but cooperative multitasking is available in the form of light-weight coroutines that run until they yield control, either explicitly or implicitly (when waiting on input or a timer)...
task/[1-n] # concurrent form of call/1-n
tasklist/[2-8] # concurrent form of maplist/1-n
Note: tasklist limits the number of concurrent tasks to a small pool (4?) of tasks active at one time. New tasks are scheduled as prior ones complete.
An example:
:-use_module(library(http)).
geturl(Url) :-
http_get(Url,_Data,[status_code(Code),final_url(Location)]),
format("Job [~w] ~w ==> ~w done~n",[Url,Code,Location]).
% Fetch each URL in list sequentially...
test54 :-
L = ['www.google.com','www.bing.com','www.duckduckgo.com'],
maplist(geturl,L),
writeln('Finished').
```console
$ tpl samples/test -g "time(test54),halt"
Job [www.google.com] 200 ==> www.google.com done
Job [www.bing.com] 200 ==> www.bing.com done
Job [www.duckduckgo.com] 200 ==> https://duckduckgo.com done
Finished
Time elapsed 0.663 secs
% Fetch each URL in list concurrently...
test56 :-
L = ['www.google.com','www.bing.com','www.duckduckgo.com'],
tasklist(geturl,L),
writeln('Finished').
$ tpl samples/test -g "time(test56),halt" Job [www.duckduckgo.com] 200 ==> https://duckduckgo.com done Job [www.bing.com] 200 ==> www.bing.com done Job [www.google.com] 200 ==> www.google.com done Finished Time elapsed 0.33 secs
Multiple* high level *prolog* objects can be created and assigned to
operating system threads in a C-wrapper program by calling
```c
prolog *pl = pl_create()
pl_consult(pl, filename)
pl_eval(pl, expr)
etc.
Each such prolog instance is thread-safe. Such instances could use Unix domain sockets for IPC.