Bug in handling ambiguity?

When running this code:

grammar = """
expression: "c" | "d" | "c" "d"
unit: expression "a"
    | "a" expression
    | "b" unit
    | "b" expression
start: unit*

%import common.WS
%ignore WS
"""

l = Lark(grammar, parser='earley', ambiguity='explicit')
print(l.parse('b c d a a c').pretty())

It is expected to have an ambiguous parse, but there is no ‘_ambig’ node.

At least these options are valid:

unit(
    b
    unit(
        expression(
            c
            d
        )
        a
    )
)
unit(
    a
    expression(
        c
    )
)

and this parse:

unit(
    b
    expression(
        c
    )
)
unit(
    expression(
        d
    )
    a
)
unit(
    a
    expression(
        c
    )
)

The only parse that comes back is the second one. When one removes the "b" expression option, you get the first one.

[erezsh]

Fixed! At the very least, the example you gave me now runs almost instantly. Please check if latest master fixes your issues.

Turns out it was a very simple bug, but it was a real flaw in my implementation. So thank you for reporting it and helping me weed it out.

See if you can find an example of name length affecting speed. I’ll be very happy to fix it if it’s true.

[erezsh]

In master.