"Joel E. Denny" <[EMAIL PROTECTED]> writes: > I'm also thinking it's strange that the user can declare %type, > %destructor, and %printer for the error token. I'm thinking it's even > stranger that the parser copies the semantic value for the error token > from the lookahead (yylval), for which there may be a separate %destructor > invocation.
Obviously it's bogus to copy the value from the lookahead token. But what about "error: ...;" rules? For them, it sort of makes sense to have a type for the error token. Perhaps if we had some way to specify the semantic value for an error token that is generated internally by the parser? Just thinking out loud here.
