How to make ANLTR report unknown identifiers as syntax errors

Question

I want to define a grammar where the lexer rule for an identifier ID should only successfully match if is valid. Such identifiers are provided in a set to the lexer.

Based on this answer here I came up with the following solution. I add another lexer constructor, passing in the set of valid identifiers. Then I added a predicate after the ID token rule to disable it if the matched text is not found. The lexer would then match the UNKNOWN token (same rule) and I was hoping to force a NoViableAltException, as there is no rule for expr containing UNKNOWN. Instead I get MismatchedInputExceptions which may be fine, but all the following syntax errors are no longer reported.

grammar T;
@lexer::members {
private java.util.Set<String> identifiers = new java.util.HashSet<>();
  public TLexer(CharStream input, java.util.Set<String> s) {
    this(input);
    this.identifiers = s;
  }

  public boolean hasIdentifier(String s) {
    return identifiers.contains(s);
  }
}

prog: expr;

expr
  : expr ('AND' | 'OR') expr
  | ID '=' STRING
  | ID ('=' | '<' | '>') INT
  ;
  
INT: ...
STRING: ...
ID: [a-z][a-zA-Z0-9]* {hasIdentifier(getText())}?;
UNKNOWN: [a-z][a-zA-Z0-9]*;
WS: [ \r\n\t] -> skip();

Is there a better way so that any invalid identifiers are reported as actual syntax errors?

Answer 1

You will get other syntax errors only if the parser can recover from the first error. That ability depends heavily on the grammar. If no re-sync is possible the parser will eventually give up and never report more errors.

Therefore I wouldn't try to get more errors. Instead report one to the user and let him fix that, before continuing with further errors.