How to make this production more 'modular' - parsing

Question

I have the following expression group where everything is thrown into the same expr rule:
grammar MyGrammar;
expr
: '(' expr ')'
// BoolExressions -- cannot move these out or else get Left-Recursion
| expr ('=' | '!=') expr
| expr 'AND' expr
| expr 'OR' expr
| ATOM
;
ATOM: [a-z]+ | [0-9]+;
WHITESPACE: [ \t\r\n] -> skip;
It works, but I would like to extract the boolExpression stuff so that I can use that separately, as some other rules I have must use a boolean expression rather than any expression. However, as soon as I do that I get a left-recursion error.
What would be a good way to break this up, so that I can separate the BooleanExpression stuff? Ideally, I would like it to "look like this":
grammar MyGrammar;
expr
: '(' expr ')'
| boolExpr
| ATOM
;
boolExpr
: expr ('=' | '!=') expr
| expr 'AND' expr
| expr 'OR' expr
;
ATOM: [a-z]+ | [0-9]+;
WHITESPACE: [ \t\r\n] -> skip;
// error(119): The following sets of rules are
// mutually left-recursive [expr, boolExpr]

Answer 1

it doesn’t quite get you a single boolExpr, but you should consider labeled alternatives:
grammar MyGrammar;
expr
: '(' expr ')'
// BoolExressions -- cannot move these out or else get Left-Recursion
| expr ('=' | '!=') expr # compareExpr
| expr 'AND' expr # andExpr
| expr 'OR' expr # orExpr
| ATOM
;
ATOM: [a-z]+ | [0-9]+;
WHITESPACE: [ \t\r\n] -> skip;
This creates separate *Context classes for each alternative which significantly reduces the complexity of contexts your listeners and visitors will deal with (there will be more of them, though, obviously). Symbols are also scoped to each alternative so you can do something like:
grammar MyGrammar;
expr
: '(' expr ')'
// BoolExressions -- cannot move these out or else get Left-Recursion
| lhs=expr ('=' | '!=') rhs=expr # compareExpr
| lhs=expr 'AND' rhs=expr # andExpr
| lhs=expr 'OR' rhs=expr # orExpr
| ATOM
;
ATOM: [a-z]+ | [0-9]+;
WHITESPACE: [ \t\r\n] -> skip;