lark-parser indented DSL and multiline documentation strings

Question

I'm trying to implement a record definition DSL using lark. It is based on indentation, which makes things a bit more complex.

Lark is a great tool, but I'm facing some dificulteis.

Here is a snippet of the DSL I'm implementing:

record Order :
    """Order record documentation
    should have arbitrary size"""

    field1 Int
    field2 Datetime:
        """Attributes should also have
        multiline documentation"""

    field3 String "inline documentation also works"

and here is the grammar used:

?start: (_NEWLINE | redorddef)*

simple_type: NAME

multiline_doc:  MULTILINE_STRING _NEWLINE
inline_doc: INLINE_STRING

?element_doc:  ":" _NEWLINE _INDENT multiline_doc _DEDENT | inline_doc

attribute_name: NAME
attribute_simple_type: attribute_name simple_type [element_doc] _NEWLINE
attributes: attribute_simple_type+
_recordbody: _NEWLINE _INDENT [multiline_doc] attributes _DEDENT
redorddef: "record" NAME ":" _recordbody



MULTILINE_STRING: /"""([^"\\]*(\\.[^"\\]*)*)"""/
INLINE_STRING: /"([^"\\]*(\\.[^"\\]*)*)"/

_WS_INLINE: (" "|/\t/)+
COMMENT: /#[^\n]*/
_NEWLINE: ( /\r?\n[\t ]*/ | COMMENT )+

%import common.CNAME -> NAME
%import common.INT

%ignore /[\t \f]+/  // WS
%ignore /\\[\t \f]*\r?\n/   // LINE_CONT
%ignore COMMENT
%declare _INDENT _DEDENT

It works fine for multiline string docs for the record definition, works fine for inline attribute definition, but doesn't work for attribute multiline string doc.

The code I use to execute is this:

import sys
import pprint

from pathlib import Path

from lark import Lark, UnexpectedInput
from lark.indenter import Indenter

scheman_data_works = '''
record Order :
        """Order record documentation
        should have arbitrary size"""

        field1 Int
        # field2 Datetime:
        #   """Attributes should also have
        #   multiline documentation"""

        field3 String "inline documentation also works"
'''

scheman_data_wrong = '''
record Order :
        """Order record documentation
        should have arbitrary size"""

        field1 Int
        field2 Datetime:
                """Attributes should also have
                multiline documentation"""

        field3 String "inline documentation also works"
'''
grammar = r'''

?start: (_NEWLINE | redorddef)*

simple_type: NAME

multiline_doc:  MULTILINE_STRING _NEWLINE
inline_doc: INLINE_STRING

?element_doc:  ":" _NEWLINE _INDENT multiline_doc _DEDENT | inline_doc

attribute_name: NAME
attribute_simple_type: attribute_name simple_type [element_doc] _NEWLINE
attributes: attribute_simple_type+
_recordbody: _NEWLINE _INDENT [multiline_doc] attributes _DEDENT
redorddef: "record" NAME ":" _recordbody



MULTILINE_STRING: /"""([^"\\]*(\\.[^"\\]*)*)"""/
INLINE_STRING: /"([^"\\]*(\\.[^"\\]*)*)"/

_WS_INLINE: (" "|/\t/)+
COMMENT: /#[^\n]*/
_NEWLINE: ( /\r?\n[\t ]*/ | COMMENT )+

%import common.CNAME -> NAME
%import common.INT

%ignore /[\t \f]+/  // WS
%ignore /\\[\t \f]*\r?\n/   // LINE_CONT
%ignore COMMENT
%declare _INDENT _DEDENT

'''

class SchemanIndenter(Indenter):
    NL_type = '_NEWLINE'
    OPEN_PAREN_types = ['LPAR', 'LSQB', 'LBRACE']
    CLOSE_PAREN_types = ['RPAR', 'RSQB', 'RBRACE']
    INDENT_type = '_INDENT'
    DEDENT_type = '_DEDENT'
    tab_len = 4

scheman_parser = Lark(grammar, parser='lalr', postlex=SchemanIndenter())
print(scheman_parser.parse(scheman_data_works).pretty())
print("\n\n")
print(scheman_parser.parse(scheman_data_wrong).pretty())

and the result is:

redorddef
Order
multiline_doc """Order record documentation
        should have arbitrary size"""
attributes
    attribute_simple_type
    attribute_name    field1
    simple_type       Int
    attribute_simple_type
    attribute_name    field3
    simple_type       String
    inline_doc        "inline documentation also works"




Traceback (most recent call last):
File "schema_parser.py", line 83, in <module>
    print(scheman_parser.parse(scheman_data_wrong).pretty())
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/lark.py", line 228, in parse
    return self.parser.parse(text)
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/parser_frontends.py", line 38, in parse
    return self.parser.parse(token_stream, *[sps] if sps is not NotImplemented else [])
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/parsers/lalr_parser.py", line 68, in parse
    for token in stream:
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/indenter.py", line 31, in process
    for token in stream:
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/lexer.py", line 319, in lex
    for x in l.lex(stream, self.root_lexer.newline_types, self.root_lexer.ignore_types):
File "/Users/branquif/Dropbox/swf_projects/schema-manager/.venv/lib/python3.7/site-packages/lark/lexer.py", line 167, in lex
    raise UnexpectedCharacters(stream, line_ctr.char_pos, line_ctr.line, line_ctr.column, state=self.state)
lark.exceptions.UnexpectedCharacters: No terminal defined for 'f' at line 11 col 2

        field3 String "inline documentation also
^

I undestand indented grammars are more complex, and lark seems to make it easier, but cannot find the mistake here.

PS: I also tried pyparsing, without success wit this same scenario, and would be hard for me to move to PLY, given the amount of code that will probably be needed.

Answer 1

The bug comes from misplaced _NEWLINE terminals. Generally, it's recommended to make sure rules are balanced, in terms of their role in the grammar. So here's how you should have defined element_doc:

?element_doc:  ":" _NEWLINE _INDENT multiline_doc _DEDENT
            | inline_doc _NEWLINE

Notice the added newline, which means that no matter which of the two options the parser takes, it ends in a similar state, syntax-wise (_DEDENT also matches a newline).

The second change, as a consequence of the first one, is:

attribute_simple_type: attribute_name simple_type (element_doc|_NEWLINE)

As element_doc already handles newlines, we shouldn't try to match it twice.