I want the result of parser to be passed to the next parser as an argument.

[Oualitsen]

Here is a pseudo example of what I want to achieve:

Parser<GQInlineFragmentDefinition> inlineFragment() {

    return seq3(
      ref1(token, "..."),
      ref1(token, "on"),
      
      
      seq3(identifier(), directiveValueList(), ref0(() => fragmentBlock(
        typeName: null // I want to pass the result of identifier() here 
      ))).map3(
        (typeName, directives, block) =>[typeName,  GQProjection(
          fragmentName: null,
          token: null,
          alias: null,
          block: block,
          directives: directives,
        ),]
      ),
    )
    
    
    .map3(( _, __, list) {
      var typeName = list.first as String;
      var projection = list.last as GQProjection;
      var def = GQInlineFragmentDefinition(
        typeName,
        projection.block!,
        projection.getDirectives(),
      );
      step1.addFragmentDefinition(def);
      return def;
    });
  }

[renggli]

Could you provide a reproducible minimal example? Do you want to do something like #112?

[Oualitsen]

The grammar is a bit too complicated to give you an example but here is my question as clear as I can make it:

suppose that you want a binary parser to parse degits (0 or 1)
the constraint is, when you read 0 you expect 1 as next token.
no constraint when you read one.

what I want to achieve is the following

  Parser<String> zeroParser() => '0'.toParser();
  Parser<String> oneParser() => '1'.toParser();
  Parser<String> digitPrser({
    String? previous
  }) {
    if(previous == "0") {
      // we must expect 1 because the previous is 0
      return oneParser();
    }
    return [zeroParser(), oneParser()].toChoiceParser().flatten();
    
  }

  Parser<String> binaryParser() {
    return [digitPrser(), digitPrser(
      previous: "0" // this is where I don't know how to get the previous value
    ) ].toSequenceParser().star().flatten();
  }

please note that this is the simplest example I could think of. Even chatgpt (even less google) could not help.

the main idea is:

parser1 parses a value and pass it to the next parser.

just like parser2(arg: parser1Result)

I hope this is clear enough.

[renggli]

Your simple example with 0s and 1s you could trivially implement as such:

final zero = '0'.toParser();
final one = '1'.toParser();
final parser = [seq2(zero, one), one].toChoiceParser().star().flatten();

For the a more generic case, where you build the next parser dynamically while parsing, you can use callCC like in the question I linked above:

final zero = '0'.toParser();
final one = '1'.toParser();
final parser = [zero, one]
    .toChoiceParser()
    .callCC<String>((continuation, context) {
      // Parse the original choice parser (zero or one).
      final result = continuation(context);
      // If this is a success and the result is '0', build a new parser
      // and continue with that:
      if (result case Success(value: '0')) {        
        return '1'.toParser().parseOn(result);
      }
      // Otherwise just return the result of the original choice parser. 
      return result;
    })
    .star()
    .flatten();

Keep in mind that this approach can quickly yield slow and hard to understand parsers (you are modifying the grammar on the fly).

[Oualitsen]

Thank you very much for your response.
I will try to use classCC and see if it works on my case.
Of course the parser for 0 and 1 was just a simple example to illustrate my need.
The first need I had is to be able to pass an argument to a parser for semantic checks.
This argument could be passed even further in the AST.
meaning: parserWithArg(String arg) ---will call ----> parser2WithArg(String arg) ----- will cal ---> parser3WithArgs ....etc.

I am using you amazing project to parse graphql schemas and generate Dart/Java code.
I want to make errors clearer in such a way when a developer reads where the error is, he will be able to fix it quickly because the error is thrown exactly at the right token.

Right now: I implemented the parser and save everything in memory in one step and then do the semantic checks once this is done.

I want to have two steps:

1st step: Parse and save important data in memory.
2nd step: use data previously saved and parse again to do the semantic checks.

This is way in some parser I needed to arguments such the (The graphql Type on which a given projected has been made)

[Oualitsen]

Hello again.
Your answer did not really work for my particular case.
I spent sometime to give another simple example to explain my exact need hoping that you will understand it better and give me hint.

I have the following parser that parses
["odd", "even"] ":" [0-9] +

This means that the parser will parse things like

'odd:3', 'even:4', 'odd:10', 'even:11'

but I have one condition (for my graphql semantic validations)

the integer parser should take a boolean value to tell it what to expect (even = true and odd = false)
after it is done parsing the integer it does the check.

now here is my simple implemenation:

Parser smartIntegerParser(bool isEvenExpected) {
  return digit().plus().flatten().trim().map(int.parse).map((value) {
    final isEven = value % 2 == 0;
    var label = isEvenExpected ? "Even" : "Odd";
    if (isEven == isEvenExpected) {
      return "$value OK!";
    } else {
      throw "$value is not $label";
    }
  });
}

final labelParser = ["odd", "even"].map((e) => e.toParser()).toChoiceParser();

final colon = char(':').trim();

final parser =
    [labelParser.map((value) {
      bool isEventExpected = value == "even";
      // I want to pass this the the smartIntegerParser parser
    },), colon, smartIntegerParser(false)].toSequenceParser();

as you can see in the comment, I can get the isEventExpected from the labelParser. And I really need to do the semantic check on the smartIntegerParser parser as implemented above!
I know it is trivial on this example but this is just an illustration of what I need.

If someone can give me a hint or just say it is not possible I would really appreciate!

[renggli]

Right now: I implemented the parser and save everything in memory in one step and then do the semantic checks once this is done.

This is how pretty much any compiler / code transformator works. For separation of concerns / easy testability / easy extensibility you want to keep parsing separate from everything else (semantic analysis, code generation, ...).

I have the following parser that parses
["odd", "even"] ":" [0-9] +

It looks to me like you are mixing in semantic analysis and the parsing. It would be much simpler, if you would check the even-odd thing after parsing on the resulting parse tree:

final parser = seq3(
  ['even'.toParser(), 'odd'.toParser()].toChoiceParser(),
  char(':'),
  digit().plus().flatten(),
).map3((key, _, value) => MapEntry(key, int.parse(value)));

// 1. Parse
final result = parser.parse('even:1').value;

// 2. Validation
if (entry.key == 'even' && !entry.value.isEven)) throw ValidationError('Even entry is odd');
if (entry.key == 'odd' && !entry.value.isOdd)) throw ValidationError('Odd entry is even');

If you really wanted, you can fold this even-odd check into the grammar itself. But again, none of the example you provided so far require that:

final parser =
    seq3(
          ['even'.toParser(), 'odd'.toParser()].toChoiceParser(),
          char(':'),
          digit().plus().flatten(),
        )
        .map3((key, _, value) => MapEntry(key, int.parse(value)))
        .where(
          (entry) =>
              (entry.key == 'even' && entry.value.isEven) ||
              (entry.key == 'odd' && entry.value.isOdd),
          message: 'Value does not match key',
        );

now here is my simple implementation:

This example fits perfectly into the example I gave with #191 (comment). Just replace return '1'.toParser().parseOn(result); with return smartIntegerParser(false).parseOn(result).

[Oualitsen]

Yes I guess there is no way to do what I wanted to achieve.

What I wanted to do is to throw an exception on the right spot! meaning I do parse a give file and save data to memory then parse again (second time with data in memory) to check the semantics. AKA multipasse compiling.

So all the example I was giving is an over simplification of the grammar I am working on. grammar

I figured out another way:
While parsing, I will save the position (line, column, fileName) of each token and I believe this should work without having to parse the file a seond time.

Thank you for your support

[renggli]

While parsing, I will save the position (line, column, fileName) of each token and I believe this should work without having to parse the file a second time.

That's the way to go. There is the token() parser that simplifies that and wraps input with this information "automatically".