Troubles with repeating delimiter

[ctron]

I am trying to get a comma separated list going, where a trailing comma is allowed:

pattern bar = {
    field1: foo,
    field2: boolean,
    field3: {},  <1>
}

<1> The final command may be present or not

I am using the playground for experimenting with this. Here is the link to my original version: playground link

This results in the problem, that it always seems to require a comma, my expectation would be, that if the first section (fields += Field ',') no longer matches, the next one is processes, which doesn't require a comma. However, that doesn't seem to work.

If I change the rule of Object to be a bit more explicit:

Object:
    '{' ( (fields += Field) (',' fields += Field)* | (fields += Field ',')+ )'}'
    ;

My expectation would be that it tries either branch, on of them succeeding. However, what I get is a frozen playground, and the following error in the browser console:

Request initialize failed with message: Parser Definition Errors detected:
 Ambiguous Alternatives Detected: <1 ,2> in <OR1> inside <Object> Rule,
<ID, ?:KW, ::KW> may appears as a prefix path in all these alternatives.
See: https://chevrotain.io/docs/guide/resolving_grammar_errors.html#AMBIGUOUS_ALTERNATIVES
For Further details.
-------------------------------
Ambiguous Alternatives Detected: <1 ,2> in <OR1> inside <Object> Rule,
<ID, ::KW, ID> may appears as a prefix path in all these alternatives.
See: https://chevrotain.io/docs/guide/resolving_grammar_errors.html#AMBIGUOUS_ALTERNATIVES
For Further details.
-------------------------------
Ambiguous Alternatives Detected: <1 ,2> in <OR1> inside <Object> Rule,
<ID, ::KW, {:KW> may appears as a prefix path in all these alternatives.
See: https://chevrotain.io/docs/guide/resolving_grammar_errors.html#AMBIGUOUS_ALTERNATIVES
For Further details.
-------------------------------
Ambiguous Alternatives Detected: <1 ,2> in <OR1> inside <Object> Rule,
<ID, ::KW, string:KW> may appears as a prefix path in all these alternatives.
See: https://chevrotain.io/docs/guide/resolving_grammar_errors.html#AMBIGUOUS_ALTERNATIVES
For Further details.
-------------------------------
Ambiguous Alternatives Detected: <1 ,2> in <OR1> inside <Object> Rule,
<ID, ::KW, boolean:KW> may appears as a prefix path in all these alternatives.
See: https://chevrotain.io/docs/guide/resolving_grammar_errors.html#AMBIGUOUS_ALTERNATIVES
For Further details.

Playground link here

My expectation would be, if there are errors like this, they get shown to the user and modifying the grammar into an "ok" state, would revive the editor.

[msujew]

Hey @ctron,

as for the playground issue: We already have an issue on the langium-website for this.

Regarding your grammar: ( (fields += Field) (',' fields += Field)* | (fields += Field ',')+ ) is ambiguous, since both alternatives start with the Field subrule and the static lookahead isn't long enough to figure out how to disambiguate it. There's an easier way to accomplish your goal:

Object:
    '{' ( (fields += Field) (',' (fields += Field)?)* )? '}'
;

[ctron]

Awesome thanks! So I do understand that the second version didn't work. But shouldn't the first one have worked too:

Object:
    '{' (fields += Field ',')* (fields += Field)? '}'
    ;

[msujew]

Well, no, and it's due to our parsers lookahead algorithm. Basically it doesn't really decide whether to parse (fields += Field ',')* OR (fields += Field)?, but instead it tries to parse (fields += Field ',')* eagerly due to it's limited lookahead. The parser doesn't look ahead enough to figure out that , is the disambiguating character. While this makes grammar construction a bit less intuitive it has a dramatic impact on parsing performance. Effectively reducing the cost of parsing from O(n³) to O(n).

You should generally avoid constructs that have multiple optional groups right after another that start with the same tokens.