In the future version of JavaIncompatibilityRulesOverridabilityCondition
for IR, we need to compute the original JVM signature of a declaration.
In descriptors, it's possible simply by using `original`.
For some reason, codegen box test was not added in 082c337faa. The main
point of it now is to check that everything works correctly in the
IrFakeOverrideBuilder mode.
It affects the `is IrGetField` check in
TypeOperatorLowering.computeNotNullAssertionText, which leads to missing
NPE messages when accessing backing fields of public properties.
#KT-64615
Replace every property with its getter and setter. This is needed
because later on, JVM backend assumes that all properties have been
lowered (by JvmPropertiesLowering) to this state.
#KT-64116 Fixed
We don't have true flexible types in the IR, but we approximate it with
internal type annotations, such as FlexibleNullability,
FlexibleMutability, RawType. These annotations are then handled
specially in JvmIrTypeSystemContext, which can construct a fake flexible
type so that type checker on IR types would behave exactly as on
frontend types.
As shown in KT-63441, one instance of flexible types where flexibility
was lost during conversion to IR is Java array/vararg types. It's
necessary to support it so that IR fake overrides could be constructed
correctly, because IR fake override checker requires parameter types to
be equal. So this change introduces another internal type annotation,
FlexibleArrayElementVariance, which is only applicable to types with
classifier kotlin/Array, and which signifies that the annotated type
`Array<X>` should rather be seen as `Array<X>..Array<out X>`.
#KT-63441 Fixed
#KT-63446 Fixed
In this commit, .ir.txt expectation test data files for the JVM IR
backend are renamed to .txt. This is done as two separate commits so
that Git would recognize that these files were logically moved, not that
the contents of the original .txt files were changed.
#KT-64680
It fixes the exception when REPL is run on an ARM Mac, as well as not
quitting after interrupting on the first line.
After this change, REPL runs fine on JDK 8, 17, 21. Note that on JDK 11,
it prints a warning during startup (but then also runs fine):
WARNING: An illegal reflective access operation has occurred
WARNING: Illegal reflective access by org.jline.terminal.impl.exec.ExecTerminalProvider$ReflectionRedirectPipeCreator to constructor java.lang.ProcessBuilder$RedirectPipeImpl()
WARNING: Please consider reporting this to the maintainers of org.jline.terminal.impl.exec.ExecTerminalProvider$ReflectionRedirectPipeCreator
WARNING: Use --illegal-access=warn to enable warnings of further illegal reflective access operations
WARNING: All illegal access operations will be denied in a future release
#KT-18355 Fixed
#KT-64013 Fixed
This call to isAbstractMethodOfAny was in Fir2IrLazyClass since its
inception in 0622be14a5, and there's no explanation why this was
necessary. Removing it does not seem to break anything, but fixes the
case when IrFakeOverrideRebuilder is enabled and when Java base class
declares an abstract equals/hashCode/toString which is not overridden in
the Kotlin subclass.
#KT-63443 Fixed
To make it possible to change nullability of IrCapturedType and change
its annotations, which is happening when determining overridability of
functions via IR.
#KT-63437 Fixed
IrCapturedType (which is a IrSimpleType) does not have variance, so it
is confusing to have it in IrSimpleTypeBuilder, and it's clearer to
construct type projection separately anyway.
- use identity in equals/hashCode. It was already effectively working
like this because of `constructor === other.constructor`
- implement variance and declare it in IrSimpleType
- remove obsolete TODO
- simplify code in captureFromArguments
- use data objects
In case there are Java sources in the module, K2 creates two instances
of `FirModuleDescriptor` with the same name: one for Kotlin sources, one
for Java sources. When building fake overrides in IR, it's important to
understand that those are the same module. Currently it's done by
comparing the names, but in the future it may be improved in KT-62534.
The changed test was in fact unrelated to KT-61805, and the problem
there was rather that no fake override was created for internal property
`schemeState` in `InspectionProfileModifiableModel`.
Types based on type parameters which have both hasEnhancedNullability
and hasFlexibleNullability were incorrectly converted to IR: only
`@EnhancedNullability` annotation was added to the IR type. This means
that the nullability of the type started to matter, which is incorrect
because the type is supposed to be flexible.
This backfired in the IR fake override builder mode, where nullability
of methods inherited from both Kotlin maps and Java maps clashed, which
resulted in the exception from FakeOverrideRebuilder.
If the type is both flexible and enhanced, we're now adding
`@FlexibleNullability` to the IR type. Note that `@EnhancedNullability`
is not needed because it only affects primitive types which need to be
handled as wrapper types on JVM, but `@FlexibleNullability` already has
that effect.
IR still slightly differs from K1 for collection subclasses, which can
be seen for example in the changed IR text tests. This is to be
investigated later.
The change in tests is needed because the `kt48495*.kt` tests are now
passing on K2 in the IR fake override builder mode, while failing on the
vanilla K2.
This is related to KT-61370 but does not entirely fix it, although
curiously it fixes all remaining tests which were muted because of it.
Deduplicate Fir2IrExtensions creation and move the
convertToIrAndActualizeForJvm function to 'cli', which allows to remove
the dependency of 'fir.entrypoint' on 'backend.jvm'.
Note that behavior slightly changes in IncrementalFirJvmCompilerRunner:
previously the value of linkViaSignatures was always false, now it is
taken from the compiler configuration, which seems more correct.
These tests are mostly used to check ABI in case you're compiling a
library or library-like code, which is no longer possible (since old JVM
backend is only used in the IDE).
Mute them for the old JVM backend instead. JVM IR behavior is fine in
all cases except maybe `destructured.kt`, for which there's now an issue
KT-63075.
Also, remove setting the value of allowUnstableDependencies to true if
K2 is used because we want K2 to report errors (as K1 does) on
unstable-ABI dependencies.
#KT-61598 Fixed
Split it to 4 functions for clarity: resolveFakeOverride,
resolveFakeOverrideOrFail, resolveFakeOverrideMaybeAbstract,
resolveFakeOverrideMaybeAbstractOrFail. Remove/inline duplicated
utilities and remove unused parameters.
It was added as a placeholder in f3c58a1df7 and was not used anywhere.
The used error is declared in `JvmBackendErrors` and is reported by the
backend.
This flag will be used on JVM to determine whether or not to generate
external enum entries mappings ("$EntriesMappings") classes. Note that
from the frontend's point of view, every enum has `entries`, so for
backend purposes we have to reach out to the underlying deserialized
data to read the flag from the metadata.
The enumEntriesIntrinsicMultipleEnums.kt test was supposed to check that
JVM backend generates 3 `$EntriesIntrinsicMappings` classes: for X, for
Y, and for Z. Mappings classes are generated for enums without
`entries`, i.e. Kotlin enums compiled without `EnumEntries` language
feature, and Java enums. The test incorrectly _enabled_ the language
feature for X though, and `$EntriesIntrinsicMappings` for X was
generated anyway because of KT-61208.
To keep the original intention of the test, I'm disabling the language
feature for X, so that it will be considered as enum without `entries`.
KT-61208 will be fixed in a separate commit (with separate tests).
The boxInline and bytecodeText tests changed their meaning when language
feature EnumEntries started to be enabled by default, so those changes
are a continuation to ebd43fc8c0. The behavior did not change after
enabling the feature, once again because of KT-61208.
Also, remove obsolete error suppressions which are no longer needed
after 64c8ce18a0.
See the comment in the code for more info.
No proper test added because the compiler test infrastructure lacks the
ability to run old compiler versions, however I've verified manually
that the change fixes the issue.
#KT-62531 Fixed
In Kotlin subclasses of `MutableCollection<Int>`, the method
`remove(Int)` has its argument boxed, so that it wouldn't clash with the
method from `java.util.List`. So `JavaOverrideChecker` should understand
that a Java method `boolean remove(java.lang.Integer)` overrides it,
otherwise platform declaration clash was reported.
The code is adapted from `forceSingleValueParameterBoxing` in K1's
`methodSignatureMapping.kt`.
The test has been moved and adapted from diagnostic to codegen box
tests, to check correct backend execution + runtime.
#KT-62316 Fixed
These two tests are now failing with IR fake override builder because
e.g. the fake override `MyMutableList.foo` has two overridden symbols:
`MyList.foo` and `MyMutableCollection.foo`, as opposed to the normal K2
mode which only has the former.
This seems to only affect the printed IR dump. Note that in K1 there
were two overridden symbols as well.
Extract the common logic of remapping fake overrides into a function.
Also, slightly change behavior of the check to be stricter: now it's
using `as S?` instead of `as? S`, which means that if the remapped
symbol is present in `fakeOverridesMap`, then it must be of the same
type.
The change 9de6df6104 caused tests `requireKotlinCompilerVersion{,K2}`
to fail once the compiler version was changed to 2.0.0-dev-*. The reason
is that when patch version is 0, compiler version is equal to the
language version (which is itself always equal to JvmMetadataVersion).
So if we're replacing JvmMetadataVersion first, we end up with
`$ABI_VERSION$-dev-...` in the output, which is incorrect.
This is needed because in order to figure out which declarations are
visible from anonymous objects in terms of overridability (see
`FirVisibilityChecker.isVisibleForOverriding`), we need to get the
package name of that anonymous object, because there's package-private
visibility on JVM.
#KT-62017 Fixed
This fixes the difference at least in the IR text test
testFakeOverridesForJavaNonStaticMembers. There's no fake override
created anymore for `packagePrivateMethod` in Test2.
#KT-61366 Fixed
`FakeOverrideBuilder.provideFakeOverrides` recursively changes overrides
for all superclasses in the hierarchy, including lazy IR, which is a lot
of extra work.
Also it leads to some tests failing in the IR fake override builder mode
because it changes correct fake overrides of Java classes to incorrect
ones. Those tests are unmuted but it doesn't mean they are fixed -- most
likely we'll generate fake overrides via IR for lazy IR too, at which
point they'll start to fail again.
Make the declaration external at the call site instead of creating two
instances of FakeOverrideCopier. It's possible because IR elements are
mutable, in particular isExternal is a var.
Maybe the same should be done to IrUnimplementedOverridesStrategy, but
that is postponed for another time.
Fake overrides created by K1, as well as fir2ir in K2, use DEFINED. So
this gets rid of a lot of differences in IR text dumps between the K2
normal mode and K2 with IR fake overrides, and will ultimately help in
implementing and testing KT-61360.
The change in AddContinuationLowering is needed to fix tests like
kt47549. When creating anonymous classes for SAM implementations, we're
reusing the same fake override building machinery. And since we can't
extract any meaningful information from the fake override parameter's
origin anymore, we have to get the parameter from the original symbol.
For example, `java.util.ArrayList.elementData` is a generic
package-private method. Not unwrapping intersection/substitution fake
overrides resulted in incorrect FQ names being compared in the package
visibility check, which led to the fake override always being created in
any subclass of ArrayList.
This change will help in testing IR fake overrides (KT-61360) because in
the current (temporary) mode, IrFakeOverrideRebuilder throws an
exception if initial fake overrides constructed by fir2ir differ from
the ones rebuilt by the IR -- and here they'll be different after fixing
KT-61366.
This is an addition to d16f33cf5b. Apparently there's another call site
of JvmIrCodegenFactory in intellij besides Evaluate Expression, namely
Android LiveEdit plugin, where it seems needed to collect all dependency
modules and resolve them via linker. LiveEdit also uses
shouldStubAndNotLinkUnboundSymbols = true, so we need to differentiate
between it and Evaluate Expression, hence the new flag. This new flag
will be set to true only for Evaluate Expression in intellij.
#IDEA-329915
Assertion thrown in `check` was wrapped into something and ignored at
the call site, for some reason. So the "Looks like this test can be
unmuted..." error was never happening for this test.
Collecting all modules takes a lot of time in Evaluate Expression in
IDE, and it's useless because we then invoke JvmIrLinker to load module
headers, but that linker is discarded in the IDE (see `val irProviders`
below).
#IDEA-329915
Note that 3 tests are still muted, but for another reason: for FIR
versions of the tests, we need to compile the "pre-release library" with
the next language version which is 2.1. But since currently
LanguageVersion.LATEST_STABLE is 1.9, the compiler refuses to read
metadata of version 2.1, regardless of its own language version. Which
is correct, but it leads to irrelevant errors in the test output -- the
ones about the incompatible metadata version, NOT about the
prereleaseness.
These 3 tests can be unmuted once the default language version is
switched to 2.0.
#KT-60780 Fixed
Before the change, the compiler threw exception in the unmuted tests,
because it tried to load metadata even though it had an unsupported
version. Use the same approach as in K1 (see
DeserializedDescriptorResolver).
Now the tests are unmuted, but note that test data differs from K1. K1
does not report errors related to the class `a.A` because it loads this
class as a _Java class_, so calling its constructor and methods somehow
works. This behavior is questionable since the compiler surely knows
that it is a Kotlin class, but with an unsupported metadata version.
Trying to interpret it as a Java class may lead to subtle problems. So
it's safer for now to avoid loading Kotlin classes with unsupported
metadata versions in K2.
#KT-60795 Fixed
"Container source" means "the source of the container". So, if you
called this extension on a class (`FirClassLikeDeclaration`), it was
reasonable to assume that you would get the source of its _containing
class_, whereas the function actually returned the source of the class
itself.
Instead of inventing another name for this function, I've decided to
inline and remove it because it has only one usage.
Optimization in 2439c22ff6 introduced an array `absentArguments` with
default values of all parameter types, which is copied in `callBy`
instead of being recomputed each time. Unfortunately, its size was
computed incorrectly: `maskSize` should only take into account
value parameters (see the `parameter.kind == KParameter.Kind.VALUE`
check in `callBy`).
This led to an issue where if the function had 32 * N value parameters
and 1 receiver parameter, `maskSize` was greater by 1 than expected,
which caused an exception due to mismatching argument array sizes.
#KT-61304 Fixed
Alexander Udalov