See the relevant test at t/
estData/codegen/box/sam/contravariantIntersectionType.kt
After the previous commit, when generating a call for `doOnSuccess {}`,
we have a SAM type with removed projection Consumer<Any?>, while
real function type of the lambda is a bit different
(Function1<ConcreteType & ConcreteType2>).
While for me, that looks questionable, but in K1 it was effectively
resolved via the same implicit cast that on JVM is generated to an
INVOKEDYNAMIC that wraps initial function-typed value.
^KT-53552 Fixed
It would be more consistently to prohibit the behavior from the unmuted
test (see KT-52428), but it was decided to postpone the breaking change.
Unfortunately, it didn't work to make a test where for computing
star projections we would need to substitute other type parameters
because effectively, it's not allowed to have SAM conversion when
star projections/wildcard is based on a type parameter which bounds
use other type parameters.
^KT-53552 In progress
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
Beside some corner cases, it's already prohibited in K1 because
adaptation have a bit strange nature
(they don't represent any existing real function exactly)
^KT-55137 Fixed
This fixes "IllegalStateException: Fir2IrSimpleFunctionSymbol for
<paramName> is already bound" when creating a callable reference to a
synthetic property
^KT-56072 Fixed
- The fix for KT-55570 caused some backend tests to fail, because errors
are now correctly reported for simple classes and actual/expect in
the same module is not supported in FIR. See KT-55177.
- The commit also adds separate tests for K2. Unfortunately, these have
to be disabled for K1 because K1 then reports "expect without actual"
errors.
In details, this commit changes the following:
- it converts FIR when without branches to empty IR block without when
- it doesn't drop empty else branches in when anymore
* `return` should only be added to the last statement if the return
type is not Unit
* If there is a `return` without an argument, then the expected return
type is Unit and the last expression is not a return argument (unless
it's an incomplete call, in which case it is inferred to return Unit;
this behavior is questionable, but inherited from K1)
* There should be a constraint on return arguments even if the expected
type is Unit, otherwise errors will be missed
* When the expected type is known, using the call completion results
writer is pointless (and probably subtly wrong).
^KT-54742 Fixed
This feature is not needed because it is unconditionally disabled for K1
(because of not fully correct implementation) and unconditionally enabled
in K2 (K2 does not support old behavior)
^KT-38895
Previously, FIR used `_context_receiver_n` while FE10 used `<this>` for
all context receiver parameters. This commit changes the code in FE10
to follow the convention from FIR.
Refactor the renderer, make BoundSymbolReferenceRenderer a static class
to prevent calling RenderIrElementVisitor's methods from it to avoid
infinite recursion in the future.
^KT-52677 Fixed
In FIR we desugar when with multiple conditions leading to same block
as tree of OR expressions
Given
```
when(some) {
"a", "b", "c" -> {}
else -> {}
}
```
actually desugared into
```
when(val <subj> = some) {
<subj> == "a" || <subj> == "b" || <subj> == "c" -> {}
else -> {}
}
```
There is a multiple ways of how we can organize such expressions in FIR
Previously it was just nesting-chain of OR expressions
While the most efficient way in terms of required stack depth is
a balanced tree
KT-53255
After we added "careful approximation of contravariant projections" in
584b70719e, some SAM conversions started to require an additional
implicit cast of the functional value before it is converted to the SAM
interface. The target type of this implicit cast was computed
incorrectly because it didn't contain nullability of the SAM type. This
could lead to a situation where a nullable value was incorrectly cast to
a non-null type, which caused a missing null check and NPE at runtime.
For example, let's consider the test `kt54600.kt`. SAM conversion
happens in the constructor call `J(filter)`. Before 584b70719e, the IR
for that argument was (irrelevant things are omitted for simplicity):
TYPE_OP SAM_CONVERSION type=Condition<String!>!
GET_VAR filter type=Function1<String, Boolean>?
After 584b70719e, the IR became:
TYPE_OP SAM_CONVERSION type=Condition<Any?>!
TYPE_OP IMPLICIT_CAST type=Function1<Any?, Boolean>
GET_VAR filter type=Function1<String, Boolean>?
Note the two changes:
* The resulting SAM type changed from `Condition<String!>` to
`Condition<Any?>`. This is exactly the point of the "careful
approximation" change, because just erasing the "in" projection from
the parameter type is incorrect, see the explanation for that change.
* The value is now implicitly cast to the _non-null_ function type
before it is SAM-converted. The presence of the cast is fine, but the
fact that it's to a non-null type is an oversight.
The target type for this cast is computed at
`KotlinType.getSubstitutedFunctionTypeForSamType` in psi2ir. Now it
extracts the nullability from the SAM type and retains it in the
resulting function type.
After this change, the IR for the argument becomes:
TYPE_OP SAM_CONVERSION type=Condition<Any?>!
TYPE_OP IMPLICIT_CAST type=Function1<Any?, Boolean>!
GET_VAR filter type=Function1<String, Boolean>?
Note that the target type is now flexible, as the resulting SAM type.
Another option would be to make it nullable, as the type of the
functional value, but there doesn't seem to be any difference.
#KT-54600 Fixed
In 1.7.20 we used the nearest Java-based receiver for such field
references in backend. Now we are using use-site receiver again,
it can lead to accidental usage of derived class property backing field.
This is effectively a revert of KT-49507 fix, see commits:
- fa914f20
- b0a6508d
#KT-54393 Fixed
#KT-49507 Planned
#KT-52338 Planned
Denis.Zharkov