fix: cyclic dependency panic

[Aditya-PS-05]

closes #21006

In commit 5038446, the query_with_diagnostics method was converted to a salsa tracked query but the cycle handler was not added during the conversion while other methods like type_for_type_alias_with_diagnostics, generic_predicates_for_param already have the cycle handlers.

So, I added a cycle handler for the query_with_diagnostics method.

And on testing, I found the issue was fixed.

[ShoyuVanilla]

Could you add a test for rust‑analyzer to address issue #21006?

Additionally, I suggest we investigate the root cause of the cycle reported in that issue:

• Is the cycle itself expected but simply missing a cycle handler?
• Or is the cycle unintended—meaning our lowering (or some other codes) is producing a cycle when it shouldn’t?

[Aditya-PS-05]

Is the cycle itself expected but simply missing a cycle handler?

the cycle is expected as it arises from rust code patterns that rustc handles correctly.

Our lowering is correct but it just needs proper cycle handling like other queries.

[ShoyuVanilla]

Which pattern? Could you point it out?

[Aditya-PS-05]

pattern when a trait's where clause references a type alias that depends on trait itself by its type parameter.

[ShoyuVanilla]

Well, your test passes even without the cycle handler

[Aditya-PS-05]

Well, your test passes even without the cycle handler

I am having some difficulty to write a proper test. The test still passes. Can you suggest me some ways to do it?

[ChayimFriedman2]

The best way is to start with the panicking code, and remove more and more code while keeping it panicking until you get the minimal snippet which still reproduces the problem.

However this query should never have a cycle (its parallels in rustc don't have either), and therefore the cycle means we're lowering things incorrectly. This fix is therefore also wrong.