Consider an infix operator like subset (⊂). The subset operator is not associative, because its result (a boolean) is not itself a set and so cannot be fed into one or another side of the subset operator. Consider:
S ⊂ T ⊂ M
Ideally this would be a parse failure, but tree-sitter does not seem to allow parse failures based on operator conflict; instead, it requires you unambiguously resolve the conflict at parser generation time by specifying associativity or precedence. Is there any way to indicate to tree-sitter this should be a parse conflict? Not only for non-associative operators of the same kind, but also between different operators with equivalent precedence which are not associative, like:
S ⊂ T ⊆ M
Or is the only solution to specify an unambiguous parse, then handle this at the semantic level?
You are correct this should be handled at the semantic level. So ⊂ should be marked left-associative in the grammar for parsing purposes, even though it is not. For the string S ⊂ T ⊂ M it will then be parsed as:
(op ⊂
(op ⊂
(id S)
(id T)
)
(id M)
)
At the semantic level you can then add a rule checking whether ⊂ has any child nodes which are also ⊂ (or any other operator of equal precedence), which you can surface as an associativity/precedence conflict error.