The `varsym` lexing rules are defined as follows: ``` <0> { @varsym / { precededByClosingToken `alexAndPred` followedByOpeningToken } { varsym_tight_infix } @varsym / { followedByOpeningToken } { varsym_prefix } @varsym / { precededByClosingToken } { varsym_suffix } @varsym { varsym_loose_infix } } ``` Unfortunately, this means that the predicates `precededByClosingToken` and `followedByOpeningToken` will be recomputed several times before we figure out the whitespace context. Rather than check for closing/opening tokens using lexer predicates, I'd rather do it in the lexer action: ``` <0> { varsym { varsym } } varsym :: Action varsym span buf len = ... -- check for opening/closing tokens here ``` But this isn't currently possible because the lexer action doesn't give access to the buffer state *after* the token. Only the predicates have this information. This is fixable by passing around an additional parameter to every lexer action: ```diff - type Action = PsSpan -> StringBuffer -> Int -> P (PsLocated Token) + type Action = PsSpan -> StringBuffer -> Int -> StringBuffer -> P (PsLocated Token) ``` Hopefully, this will improve the lexer performance ever so slightly. But even if not, it should make the handling of operator whitespace a bit cleaner.