If there's a frontend error in a test, or bytecode generation crashed
with an exception, it makes no sense to run JVM test handlers such as
lambda inlining checker or D8 checker.
This is needed in order to have a single convenient place where to
register frontend services implemented _outside_ of the 'frontend'
module, such as the control flow analysis, extracted to a separate
module in a subsequent commit.
Since nullability-flexible types in IR are represented by an annotation
on an IrSimpleType, not a special instance, the common implementation of
`KotlinTypeMarker.isMarkedNullable` in `TypeSystemContext` doesn't work.
This method is used for example to generate `typeOf` in JVM IR, in
`TypeSystemCommonBackendContext.generateTypeOf`, and this issue led to a
difference in behavior of `typeOf` for nullability-flexible types.
#KT-44726 Fixed
Avoid generating synthetic overrides in subclass
It has been already working for PSI2IR because fake overrides there don't
inherit default values for parameters, while they do it in FIR
The difference is how we deal with intermediate fake overrides
E.g., in case
interface A { /* $1 */ fun foo() }
interface B : A {
/* $2 */ fake_override fun foo()
}
interface C : B {
/* $3 */ override fun foo()
}
We've got FIR declarations only for $1 and $3, but we've got
a fake override for $2 in IR.
Previously, override $3 had $1 as its overridden IR symbol, just because
FIR declaration of $3 doesn't know anything about $2.
Now, when generating IR for $2, we save the necessary information
and using it for $3, so it has $2 as overridden.
So, it's consistent with the overridden structure of FE 1.0 and this
structure is necessary prerequisite for proper building of bridges
for special built-ins.
if it is unbound and the underlying type is reference type.
If the underlying type is primitive, it is boxed and unboxed
correctly, otherwise, it is simply casted and not unboxed.
Additionally, generate functions for inliner with inline classes
in signature, so unboxing works.
The unboxing is removed after inlining.
#KT-44722 Fixed
Use the same logic as for type constructors of classes, based on the
fully-qualified name of the classifier, with special cases for error
types and local declarations, with an additional check that the type
constructors' declaration descriptors are structurally equal via
`DescriptorEquivalenceForOverrides`. The latter is required because type
parameters of overloaded functions must be different, even though their
full FQ name is the same.
This (hopefully) has no effect for the compiler, but is useful for
kotlin-reflect where `KType.equals` runs the type checker on the
underlying `KotlinType` instances, which eventually ends up comparing
type constructors. Descriptors and types in kotlin-reflect are cached on
soft references, so they may be suddenly garbage-collected and
recomputed, and we want copies of the same type parameter to be equal to
each other.
This fixes flaky codegen tests which started to fail after migration to
the new test infrastructure, where tests are now run in parallel in the
same process, thus with higher memory pressure and more soft references
being GC'd:
* `codegen/box/reflection/types/createType/typeParameter.kt`
* `codegen/box/reflection/supertypes/genericSubstitution.kt`
Also, add a new test to check that we do the instanceof check in
overrides of `AbstractTypeConstructor.isSameClassifier`.
#KT-44850 Fixed
Taking just the `jvmClasspathRoots` is not correct because it also
contains stuff needed for resolve to work correctly, such as JDK (full
or mock), stdlib (full or mock), reflect. JDK is obviously not needed in
the classpath, and stdlib/reflect are available via the parent class
loader, which is specifically reused across all tests to make them run
faster.
Also, don't try to create class loader for Java-only modules in
`JvmBoxRunner.processModule`. This happens, for example, for all tests
which were moved from `boxAgainstJava`.
In the old backend, this was unnecessary because nested objects would
reference their lambdas' captures through the original this$0. On
JVM_IR, using loose capture fields means a name/descriptor clash can
occur on any level of nesting, not just the top.
Dmitriy Novozhilov