Fix(grammar): update sizeof operator precedence

[hsqStephenZhang]

problem

i was testing my c compiler on a corner case for sizeof expression, and i found that the output of tree-sitter is different from that of clang.

the case is:

int foo(void) {
    return 0;
}

int test(void) {
    int a = sizeof(foo)();
    return a;
}

clang's output

`-FunctionDecl 0x124950c00 <line:5:1, line:8:1> line:5:5 test 'int (void)'
  `-CompoundStmt 0x124950e78 <col:16, line:8:1>
    |-DeclStmt 0x124950e18 <line:6:5, col:26>
    | `-VarDecl 0x124950cc0 <col:5, col:25> col:9 used a 'int' cinit
    |   `-ImplicitCastExpr 0x124950e00 <col:13, col:25> 'int' <IntegralCast>
    |     `-UnaryExprOrTypeTraitExpr 0x124950de0 <col:13, col:25> 'unsigned long' sizeof
    |       `-CallExpr 0x124950db8 <col:19, col:25> 'int'
    |         `-ImplicitCastExpr 0x124950da0 <col:19, col:23> 'int (*)(void)' <FunctionToPointerDecay>
    |           `-ParenExpr 0x124950d48 <col:19, col:23> 'int (void)'
    |             `-DeclRefExpr 0x124950d28 <col:20> 'int (void)' Function 0x124935428 'foo' 'int (void)' non_odr_use_unevaluated

tree-sitter's output (part of)

body: compound_statement [4, 15] - [7, 1]
  declaration [5, 4] - [5, 26]
    type: primitive_type [5, 4] - [5, 7]
    declarator: init_declarator [5, 8] - [5, 25]
      declarator: identifier [5, 8] - [5, 9]
      value: call_expression [5, 12] - [5, 25]
        function: sizeof_expression [5, 12] - [5, 23]
          type: type_descriptor [5, 19] - [5, 22]
            type: type_identifier [5, 19] - [5, 22]
        arguments: argument_list [5, 23] - [5, 25]

the reason that cause this difference is that, clang parses the remaining tokens after sizeof as a whole postfix-expression, while tree-sitter allows a more flat grammar and tried to address this by precedence.

solution

after analyzing the output of tree-sitter parse --debug <<< 'int main() { sizeof(foo)(); }', i found out that there indeed are two options, and tree-sitter chose the wrong one due to the precedence of call and sizeof

so i use prec.dynamic(1, $.expression)) instead of $.expression, and the problem is solved.

test and verification

for the case tree-sitter parse --debug <<< 'int main() { sizeof(foo)(); }' mentioned before, here is the comparison

logs before this fix

reduce sym:type_specifier, child_count:1
reduce sym:expression, child_count:1
shift state:664
process version:1, version_count:2, state:1077, row:0, col:23
reduce sym:type_descriptor, child_count:1
shift state:396
process version:0, version_count:2, state:664, row:0, col:24
lex_internal state:50, row:0, column:24
  consume character:'('
lexed_lookahead sym:(, size:1
reduce sym:parenthesized_expression, child_count:3
reduce sym:expression, child_count:1
shift state:451
process version:1, version_count:2, state:396, row:0, col:24
reduce sym:sizeof_expression, child_count:4
reduce sym:expression, child_count:1
shift state:451

...

(translation_unit [0, 0] - [1, 0]
  (function_definition [0, 0] - [0, 29]
    type: (primitive_type [0, 0] - [0, 3])
    declarator: (function_declarator [0, 4] - [0, 10]
      declarator: (identifier [0, 4] - [0, 8])
      parameters: (parameter_list [0, 8] - [0, 10]))
    body: (compound_statement [0, 11] - [0, 29]
      (expression_statement [0, 13] - [0, 27]
        (call_expression [0, 13] - [0, 26]
          function: (sizeof_expression [0, 13] - [0, 24]
            type: (type_descriptor [0, 20] - [0, 23]
              type: (type_identifier [0, 20] - [0, 23])))
          arguments: (argument_list [0, 24] - [0, 26]))))))

after the fix

process version:0, version_count:1, state:527, row:0, col:23
lex_internal state:49, row:0, column:23
  consume character:')'
lexed_lookahead sym:), size:1
reduce sym:type_specifier, child_count:1
reduce sym:expression, child_count:1
shift state:664
process version:1, version_count:2, state:1077, row:0, col:23
reduce sym:type_descriptor, child_count:1
shift state:396
process version:0, version_count:2, state:664, row:0, col:24
lex_internal state:50, row:0, column:24

(translation_unit [0, 0] - [1, 0]
  (function_definition [0, 0] - [0, 29]
    type: (primitive_type [0, 0] - [0, 3])
    declarator: (function_declarator [0, 4] - [0, 10]
      declarator: (identifier [0, 4] - [0, 8])
      parameters: (parameter_list [0, 8] - [0, 10]))
    body: (compound_statement [0, 11] - [0, 29]
      (expression_statement [0, 13] - [0, 27]
        (sizeof_expression [0, 13] - [0, 26]
          value: (call_expression [0, 19] - [0, 26]
            function: (parenthesized_expression [0, 19] - [0, 24]
              (identifier [0, 20] - [0, 23]))
            arguments: (argument_list [0, 24] - [0, 26])))))))

and i added it to the corpus, the other test cases remain unaffected.