Clang-query silent failures are extremely misleading

[higher-performance]

Example:

int main() { return 0; }

clang-query> m expr(hasType(anyOf(type(), type())))
0 matches.

These should really produce errors -- it's too easy to write a complicated query thinking that anyOf or allOf are working correctly when they aren't.

[llvmbot]

@llvm/issue-subscribers-clang-query

Author: None (higher-performance)

[Example](https://godbolt.org/z/n7Eddabrd):

int main() { return 0; }

clang-query> m expr(hasType(anyOf(type(), type())))
0 matches.

These should really produce errors -- it's too easy to write a complicated query thinking that anyOf or allOf are working correctly when they aren't.

[AaronBallman]

I'm really not certain what the heuristic would be for determining what is a mistake and what isn't. What's more, we don't have any support for matcher-level diagnostic reasoning because clang-query is basically a thin wrapper around the C++ DSL for AST matchers.

[higher-performance]

I may have been unclear on the question, but this isn't related to heuristics at all. The problem I'm referring to is that there is a silent bug somewhere in either clang-query or the AST matchers, because the examples in the link should behave identically (but don't):

m expr(hasType(type()))
m expr(hasType(anyOf(type(), type())))
m expr(hasType(allOf(type(), type())))

The last two lines wouldn't even compile with static AST matchers (due to an ambiguous call to hasType), but somehow the dynamic version just fails silently instead of producing an error.

[AaronBallman]

I may have been unclear on the question, but this isn't related to heuristics at all. The problem I'm referring to is that there is a silent bug somewhere in either clang-query or the AST matchers, because the examples in the link should behave identically (but don't):

m expr(hasType(type()))
m expr(hasType(anyOf(type(), type())))
m expr(hasType(allOf(type(), type())))

Ah, thank you for clarifying!

The last two lines wouldn't even compile with static AST matchers (due to an ambiguous call to hasType), but somehow the dynamic version just fails silently instead of producing an error.

Yeah, this is a product of the dynamic AST matchers being more lax than the C++ DSL due to marshaling. It would be nice to address this (though my intuition is that it's a fair amount of effort).