Auto type binary operators

[Ten0]

This adds support for type inference of e.g. some_timestamptz_column - some_pginterval_expression.

Also adds support for update(...) expression (I found myself needing it for the test).

[Ten0]

The ubuntu mysql fails seem unrelated.

[weiznich]

Yes the test failures are unrelated. I hadn't the time to investigate them yet. What seems to be really strange is that the test itself passes, but nextest then reports a sigsev

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 52 filtered out; finished in 0.00s

[weiznich]

Overall this change looks good to me. Thanks for submitting.

I think PartialEq is not handled correctly yet, we might just need to mark it as unsupported for now?

[weiznich]

I think that case won't work. PartialEq is neither a member of std::ops nor has it an associated type Output. Maybe we can just mark it as unsupported for now?

[Ten0]

Ah right this one should have been among the "always boolean". 👍
fixed.

[weiznich]

Thanks for the update, I just added a correction to remove the now wrong comment about the fact that we implement all the ops as PartialEq is now missing.

[Ten0]

Thanks for the update, I just added a correction to remove the now wrong comment about the fact that we implement all the ops as PartialEq is now missing.

It's not missing: there is no PartialEq BinOp: there's only an Eq BinOp (which Rust calls PartialEq for, which is mapped, to bool, and was incorrectly mapped to <Lhs as std::ops::PartialEq<Rhs>>::Output before).
What I mean by this comment is that our match is in fact exhaustive at the time of writing, and will likely stay so unless Rust adds new operators, despite what the _ => branch might suggest.

[weiznich]

You are correct, I just search for PartialEq and missed that it's only Eq 🤦
I've reverted the removal, so everything should be fine now.