Open redruin1 opened 1 year ago
This isn't a problem with brackets. It might be an issue with ambiguous repetition.
Note that this grammar produces the same result:
program: instruction*
instruction: pneumonic parameter*
pneumonic: CNAME
parameter: [ESCAPED_STRING | INT]
And this grammar produces a similar result:
program: instruction*
instruction: pneumonic parameter*
pneumonic: CNAME
parameter: (ESCAPED_STRING | INT)?
Which means that repeating empty rules is probably creating this pattern.
Probably a duplicate of #1283. The solution is to make sure that you don't have ambiguites, and ideally you always want to use parser='lalr'
. In this case if you just use parameter*
(no brackets, since that is a duplication of the empty match possibility), it works and is even lalr compatible.
@MegaIng I tried disabling ChildFilterLALR and the ForestToParseTree cache, as that issue suggests, but this artifact still persists.
I'm writing a simple assembly-like language and I'm using Lark to parse it's AST, but I'm having trouble with ambiguity. Here's a boiled-down MinRe:
The above produces this output, which shows there's some phantom parameter between the pneumonic and the explicit strings and ints:
At first, I thought this was the parser matching the whitespace between the pneumonic and the first parameter, but removing this whitespace doesn't seem to help (especially since this whitespace is seemingly ignored anyway):
What did work however was to remove the brackets within the
instruction
rule:which resolves the ambiguity:
From what I understand the brackets indicate an "expected value" and the parser supplies
None
when nothing is found, but what is the parser actually matching in-between the pneumonic and the first parameter in this case?