Merging array types with different element types, for example
`[Lj/l/String;` and `[Lj/l/Object;`, now produces `[Lj/l/Object;`
(instead of `Lj/l/Object;`), which allows for more precise tracking of
null values because we assume that AALOAD on a non-array typed value is
possible only if that value is null.
#KT-54802 Fixed
JS scripting uses the old IR to JS transformer.
The new IR to JS transformer can not be used for
JS scripting out of the box. Patching the new transformer for
JS scripting is potentially dangerous and requires a lot of effort.
Dropping JS scripting and the old IR to JS transformer allows to
refactor and simplify JS BE codebase.
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
These should only have ever been necessary for field type inference in
coroutines, which should have been fixed by 0fc676a20c.
Unlike 903a5d69a4, this time the change is in an optimization pass.
Advantage: optimization passes have a closer to the JVM view of the
bytecode, which makes this change significantly more likely to be correct.
Disadvantage: technically we don't really guarantee optimization passes
other than FixStack will run at all. For KT-54581 this is 100% fine
since the problem itself is caused by redundant checkcast elimination in
the first place (otherwise there would've been a redundant cast to
String), but for KT-53146 this means the fix is somewhat incidental and
not necessarily guaranteed.
^KT-53146 Fixed
^KT-54581 Fixed
Since SimpleDiagnosticsCollectorWithSuppress have been replaced by
PendingDiagnosticsCollectorWithSuppress after commit 9add6f3d55,
diagnostics need to be committed now.
#KT-54366 Fixed
'equals' from any made available for overriding in inline classes
'typed' equals made available for definition in inline classes
'typed' equals definition made compulsory if 'untyped' is overridden
'operator' keyword is allowed in 'typed' equals definition
^KT-24874: Fixed
E.g. when substituting T -> Array<T>, write the bytecode for the
Array<...> part for typeOf.
This fixes various issues where either Array nesting levels, nullability
information (for typeOf), or entire reification markers were missing,
causing incorrect outputs ranging from missing `?`s to missing `[]`s to
just reified types not really being reified.
^KT-53761 Fixed
If the super class is in a file that has already been lowered, the base
method has an extra continuation parameter which breaks things.
Also, SAM wrappers around functional objects are tail-call and do not
need continuations ever, so don't even try.
^KT-50950 Fixed
We already check for {POP, Unit} sequence before ARETURN, but if the
there are multiple sequence before ARETURN, the compiler assumes, that
TCO misses.
The fix is to check, that the instruction after the sequence is either
ARETURN or another {POP, Unit} sequence.
#KT-50835
#KT-54152 Fixed
For non-suspend lambdas annotations are carried over to the
invoke method so that tooling can find the annotation there.
It seems reasonable that annotations are carried over to
the invokeSuspend method on suspend lambdas as well so that
similar tooling can be built and work for suspend lambdas.
When suspend function type is serialized, there is special logic that
adds Continuation parameter, before doing so, type-aliases has to be
expanded, attributes for resulting type should also derive from
expanded type
KT-53193, KT-54062
The previous attempt stopped this code from throwing an exception during
serialization, but the arity is still wrong so an exception is now throw
when reading the serialized type.
^KT-50997 Fixed
After an accidental change in 846537b367, we started generating metadata
into anonymous classes for callable references. This metadata contained
the Kotlin representation of the referenced function. In KT-53794, this
led to a problem where Java's protected visibility could not be
represented in Kotlin terms, which crashed the backend.
But also, this metadata is useless because there's no real use case for
interpreting it (since you already have a function reference object at
runtime with all the needed information), so it would take some extra
space in the bytecode.
#KT-53794 Fixed
This test uses a hacky mode of the compiler which is not worth it to
support further (especially in K2), `USE_SINGLE_MODULE`, where
everything is compiled in one module. The purpose of the test is just to
check that metadata for local/anonymous classes is written correctly.
So we can replace it with the tests on kotlinp, which uses
kotlinx-metadata-jvm and dumps all loaded metadata to text.
This replacement is not perfect, in particular because it won't check
that the reflection machinery is able to load this metadata, and because
it won't check that annotations are loaded correctly from the bytecode.
But IMHO it's good enough, there are box tests on reflection on local
classes (e.g. `reflection/annotations/localClassLiteral.kt`), so this
way is better than having to support the weird compiler mode for just
one test.
Ilya Chernikov