DCG interesting example

[mingodad]

Looking through github to find an interesting example for DCG's to test with trealla I found this https://github.com/litwr2/dcg-based-parser but id doesn't work out of the box with trealla.

The docs are here https://litwr2.github.io/dcg-based-parser.html . Could this be adapted to trealla and added to samples ?

[infradig]

There's some non-standard stuff in there, it would be of dubious value to support them. Not even SWI Prolog has them... unget_char/1 for instance.

On Tue, Nov 1, 2022 at 9:13 PM Domingo Alvarez Duarte < @.***> wrote:

Looking through github to find an interesting example for DCG's to test with trealla I found this https://github.com/litwr2/dcg-based-parser but id doesn't work out of the box with trealla.

The docs are here https://litwr2.github.io/dcg-based-parser.html . Could this be adapted to trealla and added to samples ?

— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/71, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSETMM4645C4V6RVF6IDWGD3PFANCNFSM6AAAAAART6T5ME . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[mingodad]

I'm asking this here because it's short and it's a topic of my practical interest, but my actual knowledge of Prolog isn't that great at the moment. It doesn't need to be exactly like that, but the overall functionality yes.

[mingodad]

Trealla also has several non standard stuff.

[guregu]

The JSON library included in Trealla (borrowed from Scryer) is a nice modern DCG: https://github.com/trealla-prolog/trealla/blob/main/library/json.pl although it can get rather slow.

This Markdown library by @aarroyoc supports Trealla: https://github.com/aarroyoc/marquete

[guregu]

I don't think it's Trealla's fault the JSON library is slow, btw. Scryer has the same problem (https://github.com/mthom/scryer-prolog/issues/1566). The code is very nice though.

[mingodad]

Thank you @guregu for the links ! Have you done any profiling to see where the problem is (slow) ?

[guregu]

For the WebAssembly ports, I needed fast JSON so I attempted to fix it. Unfortunately, it's beyond my ability at the moment. I would imagine that an application of a cut or two in the right place could help it, but it might be an issue with it using dif/2 instead, honestly no idea. The issue on Scryer's repo has some measurements. We could maybe try to borrow SWI's JSON DCG instead but I haven't looked at it closely.

I ended up making a hacky module that abuses Trealla's term reading/writing to do JSON stuff, taking advantage of the fact that JSON is a valid Prolog term. It does the job and it's very fast but it's not pretty :-)

[infradig]

Trealla and Scryer use the same reference implementation of DCGs in Prolog. Others probably use a native version, which even if transcribed, are probably faster.

On Tue, Nov 1, 2022 at 10:37 PM guregu @.***> wrote:

For the WebAssembly ports, I needed fast JSON so I attempted to fix it. Unfortunately, it's beyond my ability at the moment. I would imagine that an application of a cut or two in the right place could help it, but it might be an issue with it using dif/2 instead, honestly no idea. The issue on Scryer's repo has some measurements. We could maybe try to borrow SWI's JSON DCG instead but I haven't looked at it closely.

I ended up making a hacky module https://github.com/guregu/trealla/blob/main/library/pseudojson.pl that abuses Trealla's term reading/writing to do JSON stuff, taking advantage of the fact that JSON is a valid Prolog term. It does the job and it's very fast but it's not pretty :-)

— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/71#issuecomment-1298449334, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSEXGDAUNDBOYQRPDVX3WGEFKNANCNFSM6AAAAAART6T5ME . You are receiving this because you commented.Message ID: @.***>

[pmoura]

There's some non-standard stuff in there, it would be of dubious value to support them. Not even SWI Prolog has them... unget_char/1 for instance. …

Possibly calls to get_char/1 + unget_char/1 can be replaced with calls to peek_char/2.

Another issue is the non-standard syntax for the dynamic directive. For example (in file hoc3x.pro):

:- dynamic(names/2, ids/1, eoi/0).

This should be either:

:- dynamic((names/2, ids/1, eoi/0)).

or:

:- dynamic([names/2, ids/1, eoi/0]).

There's also the use of a non-standard Prolog flag: syntax_error.

[pmoura]

The JSON library included in Trealla (borrowed from Scryer) is a nice modern DCG: https://github.com/trealla-prolog/trealla/blob/main/library/json.pl although it can get rather slow.

There's also in alternative Logtalk's json library. It should be faster.

[mingodad]

Thank you @pmoura for discovering https://github.com/trealla-prolog/trealla/issues/73 when trying to test https://github.com/litwr2/dcg-based-parser !

Another interesting project that I'm looking now is https://github.com/souffle-lang/souffle and it has a nice way to declare relations and read input data in a more secure less verbose way (see here https://souffle-lang.github.io/simple).

.decl edge(x:number, y:number)
.input edge

.decl path(x:number, y:number)
.output path

path(x, y) :- edge(x, y).
path(x, y) :- path(x, z), edge(z, y).

edge.facts:

1   2
2   3

Could trealla also support this ?

[pmoura]

Without trying to address parsing of the declarations, assuming edge.facts is a TSV file, using Logtalk's csv library with Trealla:

$ tplgt
...

?- {csv(loader)}.
...
   true.
?- assertz(edge(_,_)), retractall(edge(_,_)).
   true.
?- csv(keep, tab, true)::read_file('edge.facts', user, edge/2).
   true
;  ... .
?- edge(A, B).
   A = 1, B = 2
;  A = 2, B = 3.
?- 

[mingodad]

That's the idea but as builtin and a lot less verbose !

[mingodad]

And also with a cheery on top, naming the arguments and using types.

[pmoura]

That's the idea but as builtin and a lot less verbose !

This functionality could be best added as a library, not as a built-in feature.

[infradig]

Not sure what you're asking for here. You can easily parse a CSV or TSV...

$ tpl ?- split("1\t2",'\t',L,R). L = "1", R = "2". ?- split_string("1\t2",'\t',' ',L). L = ["1","2"].

?- help(split/4). split/4: split(+string,+string,?left,?right) true. ?- help(split_string/4). split_string/4: split_string(+string,+sep,+pad,-list) true.

Where split_string/4 is compatible with SWI Prolog.

It doesn't nicely convert to numbers.

I would second Paulo in writing such applications in Logtalk, you gain greatly in portability. I cannot imaging adding extensions like this Souffle to Trealla.