Separate callable and class checks to prepare for checking
class scopes. Extract common checks to separate methods.
Also, it is found that in IR checker annotations on type parameters also checked,
because they stored in `matchedExpectToActual`. But it is expected that
only classes and callables are checked. This started to fail because of
added input parameter type checks inside `areAnnotationsCompatible`.
That's why `expectSymbol is IrTypeParameterSymbol` early-return is added.
^KT-60668
^KT-60936
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
Motivation:
- Functions with prefix "are" must return Boolean. And
AbstractExpectActualCompatibilityChecker even already contains some
functions with prefix "are" that return Boolean (e.g.
`areCompatibleCallableVisibilities`,
`areCompatibleSupertypesOneByOne`, etc)
- Unification with functions that are prefixed with "are" and return
Boolean
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
For StrongIncompatible `actual` declaration is considered as overload
and error reports on expected declaration. For WeakIncompatible the
error is reported straight away
Before the refactoring `areCompatibleClassScopes` returned just
`Incompatible`. It is bad because StrongIncompatible isn't possible for
classes (classes can't be overloaded). Now all class incompatibilities
are weak.
The commit has a minor impact on observable behavior (cases where we
reported the compilation problems are still reported but on another
elements):
- We no longer report type parameter class incompatibilities on expect
declaration, we report them only on actuals (it happened because all
WeakIncompatible are reported only on actuals)
- In a sense, Java implicit actualization was the only way to "overload"
classes (it would be a redeclaration compilation problem, so it
doesn't count as a valid "overload"). And since type parameters
incompatibility was StrongIncompatible for classes, we counted them as
"overloads" and didn't report incompatibility problems on Kotlin
class. Now we do report. (see
implicitJavaActualization_multipleActuals)
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
Extract main logic of `areCompatibleCallables` into two functions:
`areStrongIncompatibleCallables` and `areWeakIncompatibleCallables`.
The main point is that `areStrongIncompatibleCallables` &
`areWeakIncompatibleCallables` have very specific return types.
This commit doesn't change any logic. The commit makes the API more
type-safe ensuring that bugs like in previous commit (KT-60902) won't
happen again
^KT-60902 Fixed
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
We should prioritize to return STRONG incompatibilities over WEAK
incompatibilities. But this invariant broke in `areCompatibleCallables`,
because `areCompatibleTypeParameters` returns incompatibilities of both
types, and `areCompatibleTypeParameters` is called in WEAK
incompatibilities section.
The fix is to split `areCompatibleTypeParameters` into two functions:
`areStrongIncompatibleTypeParameters` and
`areWeakIncompatibleTypeParameters`. And call each of this function in
appropriate `areCompatibleCallables` sections.
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
This is a preparation refactoring for the following KT-60902 fix and
type-safety refactoring.
This commit doesn't have observable side effects (and can be called pure refactoring) because the following conditions:
1. `expectDeclaration is ConstructorSymbolMarker && actualDeclaration is ConstructorSymbolMarker`
2. `expectDeclaration is FunctionSymbolMarker != actualDeclaration is FunctionSymbolMarker`
can't be both `true` at the same time
^KT-59665 Fixed
Review: https://jetbrains.team/p/kt/reviews/11039/timeline
It's better to have this logic in common place
(AbstractExpectActualCompatibilityChecker) to avoid missing compilation
errors in the future
This commit fixes:
1. Missing compilation error for actual function with default arguments
for 'actual typealias' KT-59665
2. Missing compilation error for actual function with default arguments
for actual fake-override KT-59665
Alternative solution for KT-59665 is to create a special checker.
"incompatibility" vs "special checker":
Arguments for common incompatibility:
- What if we had a rule that expect and actual default params must
match? If so then it certainly would be an incompatibility.
- Technically, we do the matching of expect and actual params (because
we allow default params in common ancestors of expect and actual
declarations).
- It's hard to check that the actual definition doesn't use default
params because `ExpectedActualResolver.findActualForExpected` filters
out fake-overrides and doesn't return them. It's not clear logic for
me, that I'm afraid to touch.
implicitActualFakeOverride_AbstractMap.kt test breaks if you drop this
weird logic
- WEAK incompatibilities can be considered as "checkers". So it doesn't
matter how it's implemented, as a "incompatibility" or a "checker"
Arguments against common incompatibility:
- Although we match expect and actual declarations to allow default
params in common ancestors of expect and actual declarations, it's
still can be considered that we check that the actual declaration
doesn't have default params. And it doesn't feel right that we check
correctness of the actual declaration in expect-actual matcher.
- ~~It may change the rules of expect actual matching~~ (It's not true,
because ActualFunctionWithDefaultParameters is declared as WEAK
incompatibility)
Currently, property `hasSourceAnnotationsErased` returns the opposite
to what is stated in name. Invert it both in implementation and on
call site.
^KT-58551
Many errors are reported in stdlib with these annotations
(SinceKotlin, Deprecated, so on).
But having them only on expect is a valid case. E.g. SinceKotlin added
if some old platform-specific API becomes commonized.
^KT-58551
This implementation only checks annotations set on expect/actual
declarations and requires further refinement (e.g. checking of other
annotation targets, class scopes within typealiases).
^KT-58551
This commit is a follow-up to 0d070f8ba9.
Here we remove accidentally returning "Not a sub-type" for a constraint
like TypeVariable <: DNN type
#KT-59241 Fixed
In terms of MPP there are no such thing as `expect constructor` for enums,
but they are physically exist in FIR and IR, so we need a switch which
skips matching for them
FE 1.0 implementation did not touch to avoid any hidden changes
Also use TypeSystemContext instead of TypeSystemInferenceExtensionContext
in AbstractExpectActualCompatibilityChecker
This is needed to have an ability to implement ExpectActualMatchingContext
for IR backend. IrTypeSystemContext may operate with type substitutors,
but there is no sense to implement all methods from TypeSystemInferenceExtensionContext
in it
Previously, it was led to plainly adding NullableType <: T constraint
which silently led to successful call completion.
What is suggested is just marking such initial constraint
as unsuccessful.
In K1, the error was reported just via additional type checking
mechanism being run after call completion.
^KT-58665 Fixed
There's a heuristic for approximation of a captured type that once
it has non-trivial lower bound (other than Nothing), it's worth
approximating it to sub-type even while the containing
top-level type is being approximated to super-type.
And that sounds reasonable in case the lower bound is indeed non-trivial,
but that's not the case because nullability here comes from
the nullability of captured type position.
So, the fix is basically not to treat such approximations as non-trivial.
And while that seems to be a bit of a change in the language semantics,
it still looks reasonable (see other changes in test data and KT-58087)
^KT-57958 Fixed
^KT-58087 Fixed
This fixes a FIR-specific issue where a type variable is inferred to
Any instead of dynamic. K1 wasn't affected because of a FIR-specific
condition in ResultTypeResolver.
#KT-57962 Fixed
See the comments in the code, but mostly the motivation is that once
it was decided to stick with such a legacy thing as raw types,
we are ok with some corner-cases hacks for them
(if there are not too many of them) and they don't break anything
when there are no raw types in the code.
^KT-56616 Fixed
Namely, do not choose `Nothing?` result type when fixing a variable
that has other constraints besides the ones that came from
the relevant type parameter's upper bounds.
See more details in KT-55691.
In K1, the case from specialCallWithMaterializeAndExpectedType.kt
was working (inferred to String?) just because the branches
were analyzed independently with `String?` expected type.
This change became necessary after the previous commit when we united
inference subsystems for if/when branches (see motivation there).
NB: For K1, the behavior is left the same, but the code
was refactored a bit.
^KT-55691 Fixed
^KT-56448 Fixed
When constraint system has forks in it usually we solve all of them before
starting full completion of corresponding call. But if some call with
forks was a last statement of postponed lambda, we will never call
completion for it with FULL mode. Instead of it we complete it in PARTIAL
mode and then just merge its constraint storage into storage of outer
call. So all forks from this inner call just remain unresolved inside
outer system without this fix
^KT-55966 Fixed
Callable symbols with SinceKotlin are filtered out by a
resolution stage, but in K1 classifiers and property accessors
report API_NOT_AVAILABLE. K2 filters out properties with
unavailable accessors, but does nothing for classifiers.
This change fixes it.
^KT-54410 Fixed
Roman Efremov