Holy Cow, ECLiPSe Prolog is somewhere between SWI-Prolog
and SICStus Prolog, concerning tolerating extra syntax.

SICStus Prolog, quite tolerant:

?- X = (- :- x0).
X = ((-):-x0) ?
?- X = (:- :- x0).
X = ((:-):-x0) ?

ECLIPSe Prolog, half way tolerant:

[eclipse 36]: X = (- :- x0).
X = ((-) :- x0)
[eclipse 37]: X = (:- :- x0).
syntax error: postfix/infix operator expected

SWI-Prolog, intolerant, good!!!

?- X = (- :- x0).
ERROR: Syntax error: Operator priority clash
?- X = (:- :- x0).
ERROR: Syntax error: Operator priority clash

My reader prototype sides with SWI-Prolog. Also
Dogelog Player and formerly Jekejeke Prolog side
with SWI-Prolog. Dogelog Player can be tried online:

?- X = (- :- x0).
error(syntax_error(operator_clash), [user:0])
error(syntax_error(cannot_start_term), [user:1])
?- X = (:- :- x0).
error(syntax_error(operator_clash), [user:1])
error(syntax_error(cannot_start_term), [user:2])
http://www.xlog.ch/izytab/moblet/docs/18_live/10_reference/example01/package.html

Unfortunately resyncing after an error is not yet
implemented, so there are two errors. But we got
already line numbers for Dogelog Player,

this is quite a gas!!!

Mostowski Collapse schrieb am Montag, 18. April 2022 um 23:27:20 UTC+2:
> Holy Cow, ECLiPSe Prolog is somewhere between SWI-Prolog
> and SICStus Prolog, concerning tolerating extra syntax.
>
> SICStus Prolog, quite tolerant:
>
> ?- X = (- :- x0).
> X = ((-):-x0) ?
> ?- X = (:- :- x0).
> X = ((:-):-x0) ?
> ECLIPSe Prolog, half way tolerant:
>
> [eclipse 36]: X = (- :- x0).
> X = ((-) :- x0)
> [eclipse 37]: X = (:- :- x0).
> syntax error: postfix/infix operator expected
>
> SWI-Prolog, intolerant, good!!!
>
> ?- X = (- :- x0).
> ERROR: Syntax error: Operator priority clash
> ?- X = (:- :- x0).
> ERROR: Syntax error: Operator priority clash

More proof that Logtalk is utter nonsense. What
does the Logtalk test suite even test concerning
the various "adapters" it has, like Trealla?

This is a nice gem:

$ ./tpl -v
Trealla Prolog (c) Infradig 2020-2022, v1.27.12-31-g3575df
$ ./tpl
?- X = (:- (:- x0)).
Error: operator clash, line 1
false.
?-

There is something wrong with Ulrich Neumerkels
compliance test suite, and maybe some QuickCheck
resp. Fuzzer approach is needed,

to chop of all heads of the hydra.

I am little bit behind schedule of a Fuzzer that also
does parenthesis. So this was my sixth sense. Was also
working on another front, resyncing after errors.

I found a solution today: Have a grammer that covers non-errorneous
sentences and errorneous sentences. How to do this simply?
Its more difficult to resync the tokenization, then to resync the

parsing. For parsing I found this useful, since Prolog has the
concept of terminating period. Take the very simplified
prototype posted elsewhere:

reader_secondary(H, X, L, C) --> [A],
{current_op(R, M, A), is_infix(M, D, E), L >= R}, !,
{R-D < C -> throw(error(syntax_error(operator_clash),_)); true},
{T is R-E}, reader(Z, T),
{J =.. [A,H,Z]},
reader_secondary(J, X, L, R).

It throws an exception inside an auxiliary action {}/1 of DCG. But
you can define throw//1, i.e. make throw itself a DCG non-terminal:

throw(T) --> read_sync, {throw(T)}.

And then replace the infix parsing, its only an example, works also
for all other parser errors:

reader_secondary(H, X, L, C) --> [A],
{current_op(R, M, A), is_infix(M, D, E), L >= R}, !,
({R-D < C} -> throw(error(syntax_error(operator_clash),_)); {true}),
{T is R-E}, reader(Z, T),
{J =.. [A,H,Z]},
reader_secondary(J, X, L, R).

The parser is now calling throw//1 instead of throw/1. And throw//1
can do some syncing.

Mostowski Collapse schrieb am Freitag, 29. April 2022 um 12:42:21 UTC+2:
> More proof that Logtalk is utter nonsense. What
> does the Logtalk test suite even test concerning
> the various "adapters" it has, like Trealla?
>
> This is a nice gem:
>
> $ ./tpl -v
> Trealla Prolog (c) Infradig 2020-2022, v1.27.12-31-g3575df
> $ ./tpl
> ?- X = (:- (:- x0)).
> Error: operator clash, line 1
> false.
> ?-
>
> There is something wrong with Ulrich Neumerkels
> compliance test suite, and maybe some QuickCheck
> resp. Fuzzer approach is needed,
>
> to chop of all heads of the hydra.

I implemented this already for the full fledged parser. But the simplified
version doesn’t have it yet, because the simplified version didn’t parse a
list of period terminated terms anyway yet. If I have time I will do the

simplified version as well. Currently I get nice syncing, like for example:

?- X = (:- :- x0).
error(syntax_error(operator_clash), [user:1])
?- X = (:- x0.
error(syntax_error(parenthesis_balance), [user:2])
?- X = :- x0).
error(syntax_error(operator_clash), [user:3])
?- X = (:- x0).
X = (:- x0).
http://www.xlog.ch/izytab/doclet/docs/18_live/10_reference/example01/package.html

It can parse all 4 queries, and doesn’t get out of sync. read_sync//0 is
currently implemented by skipping tokens until the terminating period
is reached or end of file. A more mature implementation of throw//1

can do much more, like fetching line number before it does read_sync//0.

Mostowski Collapse schrieb am Freitag, 29. April 2022 um 12:45:50 UTC+2:
> I am little bit behind schedule of a Fuzzer that also
> does parenthesis. So this was my sixth sense. Was also
> working on another front, resyncing after errors.
>
> I found a solution today: Have a grammer that covers non-errorneous
> sentences and errorneous sentences. How to do this simply?
> Its more difficult to resync the tokenization, then to resync the
>
> parsing. For parsing I found this useful, since Prolog has the
> concept of terminating period. Take the very simplified
> prototype posted elsewhere:
> reader_secondary(H, X, L, C) --> [A],
> {current_op(R, M, A), is_infix(M, D, E), L >= R}, !,
> {R-D < C -> throw(error(syntax_error(operator_clash),_)); true},
> {T is R-E}, reader(Z, T),
> {J =.. [A,H,Z]},
> reader_secondary(J, X, L, R).
> It throws an exception inside an auxiliary action {}/1 of DCG. But
> you can define throw//1, i.e. make throw itself a DCG non-terminal:
>
> throw(T) --> read_sync, {throw(T)}.
>
> And then replace the infix parsing, its only an example, works also
> for all other parser errors:
> reader_secondary(H, X, L, C) --> [A],
> {current_op(R, M, A), is_infix(M, D, E), L >= R}, !,
> ({R-D < C} -> throw(error(syntax_error(operator_clash),_)); {true}),
> {T is R-E}, reader(Z, T),
> {J =.. [A,H,Z]},
> reader_secondary(J, X, L, R).
> The parser is now calling throw//1 instead of throw/1. And throw//1
> can do some syncing.
> Mostowski Collapse schrieb am Freitag, 29. April 2022 um 12:42:21 UTC+2:
> > More proof that Logtalk is utter nonsense. What
> > does the Logtalk test suite even test concerning
> > the various "adapters" it has, like Trealla?
> >
> > This is a nice gem:
> >
> > $ ./tpl -v
> > Trealla Prolog (c) Infradig 2020-2022, v1.27.12-31-g3575df
> > $ ./tpl
> > ?- X = (:- (:- x0)).
> > Error: operator clash, line 1
> > false.
> > ?-
> >
> > There is something wrong with Ulrich Neumerkels
> > compliance test suite, and maybe some QuickCheck
> > resp. Fuzzer approach is needed,
> >
> > to chop of all heads of the hydra.

This is an interesting pPEG example bug. Actually I wanted to hunt
parenthesis parsing bugs, so I was up to generating some test cases that
contain parenthesis. But this test case doesn’t contain any parenthesis,

but it seems a binary infix operator is parsed as an unary functor?

?- string_termList(":- x0 =:= :- x1 - x2 .", [Y]), write_canonical(Y), nl.
:-(:-(=:=(x0),-(x1,x2)))
Y = (:- (=:=(x0):-x1-x2)).

I am still using pPEG from 14.04.2022. How do I do a package update?

Concerning parenthesis I only found this bug, but its not
really parenthesis related. So unlike Trealla it could be
that parenthesis are not a problem. This here gives the same

associativity hickup with and without parenthesis:

?- string_termList("- ( x0 * x1 ) ** x2 =:= x3 .", [Y]),
write_canonical(Y), nl.
-(=:=(**(*(x0,x1),x2),x3))
Y = - ((x0*x1)**x2=:=x3).

?- string_termList("- x4 ** x2 =:= x3 .", [Y]), write_canonical(Y), nl.
-(=:=(**(x4,x2),x3))
Y = - (x4**x2=:=x3).

?- Y = (- x4 ** x2 =:= x3), write_canonical(Y), nl.
=:=(-(**(x4,x2)),x3)
Y = (-x4**x2=:=x3).

Mostowski Collapse schrieb:
> This is an interesting pPEG example bug. Actually I wanted to hunt
> parenthesis parsing bugs, so I was up to generating some test cases that
> contain parenthesis. But this test case doesn’t contain any parenthesis,
>
> but it seems a binary infix operator is parsed as an unary functor?
>
> ?- string_termList(":- x0 =:= :- x1 - x2 .", [Y]), write_canonical(Y), nl.
> :-(:-(=:=(x0),-(x1,x2)))
> Y =  (:- (=:=(x0):-x1-x2)).
>
> I am still using pPEG from 14.04.2022. How do I do a package update?

I didn’t start a new testing campaign with pPEG yet, where I get
a new type of error, which I do not yet know how to silence.
But meanwhile I have a funny test case where

SWI-Prolog and SICStus disagree:

/* SWI-Prolog */
?- X = (- - - * - - - x0), write_canonical(X), nl.
*(-(-(-)),-(-(-(x0))))
X = - - (-)* - - -x0.

/* SICStus Prolog */
?- X = (- - - * - - - x0), write_canonical(X), nl.
-(-(-(-(*))),-(-(x0)))
X = - - - (*)- - -x0 ?

The SWI-Prolog built-in with_output_to/2 does also not work.
Would need something like a with_error_to/2 ?
Does this exist? I only get:

?- with_output_to(atom(_), (write('I am chatty'), nl)).
true.

?- with_output_to(atom(_), string_termList("- x4 ** x2 =:= .", [Y])).
% pPEG Error: Prolog.expr failed, expected Prolog.expr at line 1.16:
% 1 | - x4 ** x2 =:= .
% ^
false.

Mostowski Collapse schrieb am Montag, 2. Mai 2022 um 16:48:56 UTC+2:
> I didn’t start a new testing campaign with pPEG yet, where I get
> a new type of error, which I do not yet know how to silence.
> But meanwhile I have a funny test case where
>
> SWI-Prolog and SICStus disagree:
>
> /* SWI-Prolog */
> ?- X = (- - - * - - - x0), write_canonical(X), nl.
> *(-(-(-)),-(-(-(x0))))
> X = - - (-)* - - -x0.
>
> /* SICStus Prolog */
> ?- X = (- - - * - - - x0), write_canonical(X), nl.
> -(-(-(-(*))),-(-(x0)))
> X = - - - (*)- - -x0 ?

My parser research has arrived at this eclectic challenge.
Here is an example where I can coerce both SWI-Prolog and
pPEG SWIPL Example into masking the real error.

Here is what ECLiPSe Prolog does:

[eclipse 1]: X = foo = bar = baz = 'ab\qc' .
syntax error: bracket necessary
| X = foo = bar = baz = 'ab\qc' .
| ^ here

And here is SWI-Prolog and pPEG SWIPL Example:

?- X = foo = bar = baz = 'ab\qc' .
ERROR: Syntax error: Unknown character escape in quoted atom or string: `\q'
ERROR: X = foo = bar = baz = 'ab
ERROR: ** here **
ERROR: \qc' .

?- string_termList("X = foo = bar = baz = 'ab\\qc' .", [T]).
% pPEG Error: _esc failed, ...
% 1 | X = foo = bar = baz = 'ab\qc' .
% ^
false.

It seems Scryer Prolog is not that advanced like ECLiPSe Prolog:

$ target/release/scryer-prolog
?- X = foo = bar = baz = 'ab\qc' .
error(syntax_error(invalid_single_quoted_character),read_term/3).

Initially my thought was my monadic parser from Dogelog player
will automatically be able to deal with this. But its non-trivial.
But my Dogelog player got a scanner/parser booster yesterday:

?- X = foo = bar = baz = 'ab\qc' .
error(syntax_error(operator_clash), [user:1])
Mostowski Collapse schrieb am Dienstag, 17. Mai 2022 um 23:34:51 UTC+2:
> My parser research has arrived at this eclectic challenge.
> Here is an example where I can coerce both SWI-Prolog and
> pPEG SWIPL Example into masking the real error.
>
> Here is what ECLiPSe Prolog does:
>
> [eclipse 1]: X = foo = bar = baz = 'ab\qc' .
> syntax error: bracket necessary
> | X = foo = bar = baz = 'ab\qc' .
> | ^ here
>
> And here is SWI-Prolog and pPEG SWIPL Example:
>
> ?- X = foo = bar = baz = 'ab\qc' .
> ERROR: Syntax error: Unknown character escape in quoted atom or string: `\q'
> ERROR: X = foo = bar = baz = 'ab
> ERROR: ** here **
> ERROR: \qc' .
>
> ?- string_termList("X = foo = bar = baz = 'ab\\qc' .", [T]).
> % pPEG Error: _esc failed, ...
> % 1 | X = foo = bar = baz = 'ab\qc' .
> % ^
> false.

Now I have a new project going. I need a better microscope
to find differences in parsing of the different Prolog systems.
So that I might generate a phylogenic tree.

I see a trend against ECliPSe Prolog concering fuzz3, now also
results available for Scryer Prolog and Trealla Prolog. Unfortunately
no data yet available for Tau Prolog, still waiting

for charsio from Tau Prolog:

ECLiPSe Scryer Trealla
fuzz3 292 0 0
fuzz4 382 801 532

My suspicion, for fuzz4, Scryer Prolog behaves like GNU Prolog,
not shown here, but GNU Prolog has same figures. But this would
need a distance matrice, to get some support for this hypothesis,

that they are really the same.

Mostowski Collapse schrieb am Dienstag, 17. Mai 2022 um 23:37:12 UTC+2:
> It seems Scryer Prolog is not that advanced like ECLiPSe Prolog:
>
> $ target/release/scryer-prolog
> ?- X = foo = bar = baz = 'ab\qc' .
> error(syntax_error(invalid_single_quoted_character),read_term/3).
>
> Initially my thought was my monadic parser from Dogelog player
> will automatically be able to deal with this. But its non-trivial.
> But my Dogelog player got a scanner/parser booster yesterday:
> ?- X = foo = bar = baz = 'ab\qc' .
> error(syntax_error(operator_clash), [user:1])
> Mostowski Collapse schrieb am Dienstag, 17. Mai 2022 um 23:34:51 UTC+2:
> > My parser research has arrived at this eclectic challenge.
> > Here is an example where I can coerce both SWI-Prolog and
> > pPEG SWIPL Example into masking the real error.
> >
> > Here is what ECLiPSe Prolog does:
> >
> > [eclipse 1]: X = foo = bar = baz = 'ab\qc' .
> > syntax error: bracket necessary
> > | X = foo = bar = baz = 'ab\qc' .
> > | ^ here
> >
> > And here is SWI-Prolog and pPEG SWIPL Example:
> >
> > ?- X = foo = bar = baz = 'ab\qc' .
> > ERROR: Syntax error: Unknown character escape in quoted atom or string: `\q'
> > ERROR: X = foo = bar = baz = 'ab
> > ERROR: ** here **
> > ERROR: \qc' .
> >
> > ?- string_termList("X = foo = bar = baz = 'ab\\qc' .", [T]).
> > % pPEG Error: _esc failed, ...
> > % 1 | X = foo = bar = baz = 'ab\qc' .
> > % ^
> > false.

I made a kind of phylogenetic tree, the numbers in it indicate
how many test cases the various Prolog systems agree.
The results for fuzz4 are as follows:

?- pairing([eclipse4,gnu4,jekejeke4,swi4,scryer4,sicstus4,
ppeg4,trealla4], R), show(R, 0).
+--- 8713
+--- 9509
+--- trealla4
+--- 9997
+--- scryer4
+--- gnu4
+--- 9288
+--- 9586
+--- 9860
+--- eclipse4
+--- swi4
+--- jekejeke4
+--- 9642
+--- ppeg4
+--- sicstus4

SWI-Prolog is close to ECLiPSe Prolog in fuzz4 test cases. But pPEG
SWIPL Example is close to SICStus Prolog in fuzz4 test cases.

And my suspicion got confirmed, Scryer and GNU
do the same in fuzz4 test cases, except for a

few amusing strange results:

$ target/release/scryer-prolog -v
"v0.9.0-146-g25418db2-modified"
$ target/release/scryer-prolog
?- X = (- ( ( ) =:= x0).
X = (- '('=:=x0).
?- X = (- ( ( ) , x0).
X = (- '(',x0).
?- X = ( ( ( ) :- x0).
X = ('(':-x0).

LoL

Mostowski Collapse schrieb:
> I made a kind of phylogenetic tree, the numbers in it indicate
> how many test cases the various Prolog systems agree.
> The results for fuzz4 are as follows:
>
> ?- pairing([eclipse4,gnu4,jekejeke4,swi4,scryer4,sicstus4,
> ppeg4,trealla4], R), show(R, 0).
> +--- 8713
> +--- 9509
> +--- trealla4
> +--- 9997
> +--- scryer4
> +--- gnu4
> +--- 9288
> +--- 9586
> +--- 9860
> +--- eclipse4
> +--- swi4
> +--- jekejeke4
> +--- 9642
> +--- ppeg4
> +--- sicstus4
>
> SWI-Prolog is close to ECLiPSe Prolog in fuzz4 test cases. But pPEG
> SWIPL Example is close to SICStus Prolog in fuzz4 test cases.
>

Ok, this was fun:

Phylogenetic Trees for Prolog Operator Parsers
https://twitter.com/dogelogch/status/1528118780890603522

Phylogenetic Trees for Prolog Operator Parsers
https://www.facebook.com/groups/dogelog

Mostowski Collapse schrieb am Samstag, 21. Mai 2022 um 15:21:49 UTC+2:
> And my suspicion got confirmed, Scryer and GNU
> do the same in fuzz4 test cases, except for a
>
> few amusing strange results:
>
> $ target/release/scryer-prolog -v
> "v0.9.0-146-g25418db2-modified"
> $ target/release/scryer-prolog
> ?- X = (- ( ( ) =:= x0).
> X = (- '('=:=x0).
> ?- X = (- ( ( ) , x0).
> X = (- '(',x0).
> ?- X = ( ( ( ) :- x0).
> X = ('(':-x0).
>
> LoL
>
> Mostowski Collapse schrieb:
> > I made a kind of phylogenetic tree, the numbers in it indicate
> > how many test cases the various Prolog systems agree.
> > The results for fuzz4 are as follows:
> >
> > ?- pairing([eclipse4,gnu4,jekejeke4,swi4,scryer4,sicstus4,
> > ppeg4,trealla4], R), show(R, 0).
> > +--- 8713
> > +--- 9509
> > +--- trealla4
> > +--- 9997
> > +--- scryer4
> > +--- gnu4
> > +--- 9288
> > +--- 9586
> > +--- 9860
> > +--- eclipse4
> > +--- swi4
> > +--- jekejeke4
> > +--- 9642
> > +--- ppeg4
> > +--- sicstus4
> >
> > SWI-Prolog is close to ECLiPSe Prolog in fuzz4 test cases. But pPEG
> > SWIPL Example is close to SICStus Prolog in fuzz4 test cases.
> >