[antlr-interest] Left recursion removal

[Johannes Luber]

Loring Craymer wrote:
> Unless I missed something, these two rules are
> identical, except that element_access is incorrectly
> defined (the three occurrences of primary_expression
> with suffixes are in fact just primary expression
> variants).  Take a closer look to be sure, then change
> occurrences of "element_access" to
> "primary_expression" and delete the element_access
> rule.
> 
> --Loring

Thanks for your idea! Using the primary_expression variant without left
recursion removal I saw that primary_expression has two additional
alternatives: "array_creation_expression" and "element_access '['
expression_list ']'". Simply replacing the element_access reference with
primary_expression would allow array_creation_expression directly in
front of "'[' expression_list ']'" instead having always either "'.'
identifier type_argument_list?", "'(' argument_list? ')'", "'++'" or
"'--'" between those subrules. Unfortunately every other alternative
requires that no tokens may be inserted in this place. That means that
the following rule isn't quite equivalent:

primary_expression
    :   array_creation_expression
    |   literal
    |   simple_name
    |   parenthesized_expression
    |   primary_expression '.' identifier type_argument_list?
    |   predefined_type '.' identifier type_argument_list?
    |   qualified_alias_member '.' identifier type_argument_list?
    |   primary_expression '(' argument_list? ')'
    |   primary_expression ('.' identifier type_argument_list? | '('
argument_list? ')' | '++' | '--')? '[' expression_list ']'
    |   this_access
    |   base_access
    |   primary_expression '++'
    |   primary_expression '--'
    |   object_creation_expression
    |   delegate_creation_expression
    |   typeof_expression
    |   checked_expression
    |   unchecked_expression
    |   default_value_expression
    |   anonymous_method_expression
    |   sizeof_expression
    |   {isUnsafeContext}?=> pointer_member_access
    |   {isUnsafeContext}?=> pointer_element_access
    |   {allowOrcas}?=> anonymous_object_creation_expression
    ;

How do I check best that array_creation_expression has to be used as
left-most token when the optional subrule "('.' identifier
type_argument_list? | '(' argument_list? ')' | '++' | '--')?" has been
used (and the opposite - no occurrence of the optional subrule requires
no array_creation_expression)? Can I do it with actions or do I have to
check it in another pass over the grammar (which would be doable but
require one pass more than already planned)?

Best regards,
Johannes Luber