KotlinJvmBinaryClass.AnnotationArrayArgumentVisitor didn't cover the
case when the element type is an Annotation. Therefore, when the
compiler read an array of annotations from JVM binary classes built from
Kotlin sources, it got an empty array regardless of what was written in
the bytecode.
For example, Foo.value below is read as an empty array when SomeClass
resides in another Kotlin module.
@Foo(
value = [Bar(1), Bar(2)]
)
class SomeClass
This makes sense because this mode is the default in the production
compiler. Forgetting to enable it where necessary led to different
bizarre test failures, see for example changes around 3fee84b966 and
KT-34826
Only invariant array projections and non-null element types will be
supported soon (see KT-26568), so it makes no sense to store the
complete type in KClassValue. What we need is only the ClassId of the
class, and the number of times it's wrapped into kotlin/Array, which is
exactly what ClassLiteralValue represents.
This change helps in decoupling annotation values from
descriptors/types. The only constant value that depends on descriptors
is now AnnotationValue.
#KT-26582 Fixed
Reflection expects to see a callable method for a hidden constructor,
thus, it should be a synthetic accessor.
JVM method signature in metadata should point to the synthetic accessor.
Annotations for hidden constructor should be written on the synthetic
accessor.
In MemberDeserializer.loadProperty, we incorrectly passed 0 to
getAnnotations when loading annotations on property accessors in case
the protobuf field getter_flags/setter_flags was not present. The
correct behavior, as described in metadata.proto, was to pass a special
"default accessor flags" value, constructed from the main property
flags. Otherwise in case there were annotations both on the property and
on the accessor (as in PropertyAndAccessor.kt) and the accessor was
otherwise default, we would assume that it had no annotations and would
not load them in compiler and reflection
#KT-25499 In Progress
Note that this change brings an incompatibility: `Array<Foo>::class`
will be seen as `Foo::class` by the old deserializer. We consider this
OK because the compiler never had any logic that relied on reading class
literal arguments correctly (otherwise it wouldn't have worked because
it could only see `Array<*>::class` before this commit), and the support
of annotations on types in JVM reflection is only available in the
upcoming 1.3 release (KT-16795)
#KT-22069 Fixed
Also, fix the value of "hasAnnotations" flag to reflect if there are any
_non-source_ annotations on a declaration.
Unfortunately, after this change
IncrementalJsCompilerRunnerTestGenerated$PureKotlin.testAnnotations
starts to fail because of the following problem. The problem is that
annotations on property accessors are not serialized yet on JS (see
KT-14529), yet property proto message has setterFlags field which has
the hasAnnotations flag. Upon the full rebuild of the code in that test,
we correctly write hasAnnotations = true, but annotations themselves are
not serialized. After an incremental build, we deserialize property
setter descriptor, observe its Annotations object which happens to be an
instance of NonEmptyDeserializedAnnotationsWithPossibleTargets. Now,
because annotations itself are not serialized, that Annotations object
has no annotations, yet its isEmpty always returns false (see the code).
Everything worked correctly before the change because in
DescriptorSerializer.hasAnnotations, we used Annotations.isEmpty and the
result was the same in the full rebuild and in the incremental scenario.
But now we're actually loading annotations, to determine their
retention, and that's why the setterFlags are becoming different here
and the test fails
#KT-23360 Fixed
- Use FULL_JDK instead of mock JDK in some tests because mock JDK is
created from JDK 6 and full JDK is now JDK 8, so there are differences
in the behavior in the compiler and at runtime
- Remove some '*.runtime.txt' files which were workarounds to JDK 6
reflection issues regarding generic inner classes; code in these tests
is now loaded exactly the same in the compiler and at runtime
- Change supertype in SupertypesAndBounds.kt: the class in the supertype
is not relevant to that test, it checks that annotations can be loaded
on types
'SuspendFunction$n' class descriptors are created on demand by KotlinBuiltIns (and cached).
On serialization, types constructed with 'SuspendFunction$n' are written as 'Function$n' with extra flag (SUSPEND_TYPE).
On deserialization, corresponding 'SuspendFunction$n' classes are used.
Alexander Udalov