Lexer negation in ANTLR - why doesn't double-negation work? - antlr

Question

I have this token I would like to match any string of non-special characters except for the literal string "TODAY".
ANTLR makes this a pain in the butt:
UNQUOTED :
( ~('T'|'t'|~UnquotedStartChar) UnquotedChar*
( ~('O'|'o'|~UnquotedChar) UnquotedChar*
| ('O'|'o')
( ~('D'|'d'|~UnquotedChar) UnquotedChar*
| ('D'|'d')
( ~('A'|'a'|~UnquotedChar) UnquotedChar*
| ('A'|'a')
( ~('Y'|'y'|~UnquotedChar) UnquotedChar*
)?
)?
)?
)?
)
;
I figured I was being clever by double-negating in the ~('T'|'t'|~UnquotedStartChar) - it should match everything that isn't 'T', 't', or non-permitted in the start char normally... which intuitively seems like it should work, and when I think about it a bit, seems like it should work, but when I actually try to compile it, I get this error message:
error(100): antlr/QueryParser.g:0:1: syntax error: buildnfa: MismatchedTreeNodeException(16!=32)
UnquotedStartChar is, in itself, quite a mess... but I'm not entirely sure it's relevant yet. Here's the first couple of levels anyway.
fragment
UnquotedStartChar
: EscapeSequence
| ~( ProhibitedAtStartOfUnquoted )
;
fragment
ProhibitedAtStartOfUnquoted
: ProhibitedInUnquoted
| Slash | PLUS | MINUS | DOLLAR;
fragment
EscapeSequence
: Backslash
( 'u' HexDigit HexDigit HexDigit HexDigit
| ~( 'u' )
)
;

Answer 1

In lexer grammar, rule (token) that is defined earlier takes precedence in case of a conflict, so you could just define a token for 'TODAY' above UNQUOTED and it should match the string instead of the general UNQUOTED token.