Change the treatment of default implementations on interfaces in JVM
compatibility mode. Previously, the IR backend moved the actual
default implementation to the DefaultImpls class, and then bridged to
it from the interface default. The old backend did the reverse, at the
cost of an additional accessor, in order to gain better binary
compatibility properties. See #2612 for discussion.
The accessor needs to call a specific implementation, so must be
performed through an `invokespecial`. We trick the
SyntheticAccessorLowering into doing this for us, by marking the
bridging call as a super call. We do this in want of an explicit
`invokespecial` Ir Node.
InterfaceDefaultCallsPhase previously assumed the old behaviour of the
IR backend (that calls to default implementations, e.g. `foo$default`
should target `DefaultImpls.foo$default`). But now the bridge to
foo$default resides on `DefaultImpls` already, causing that pass to
create a recursive loop. We cut that loop with a simple check.
In old BE we generate `create` for this kind of suspend lambdas, which,
like in simple suspend lambdas is responsible for putting arguments
(including extension receiver) to fields.
But, if number of parameters of suspend lambda (including receiver) >= 1,
there is no need to generate `create`, since `create` is called only by
`createCoroutine` and friends from stdlib, and there is no version of
`createCoroutine` for lambdas with multiple parameters.
Thus, in old BE we generate a redundant method, which affects method
count.
In JVM_BE we decided to 'inline' create into `invoke` for suspend lambdas
with multiple parameters.
Updated test checks, that large unsigned numbers are converted to corresponding
negative signed numbers properly. Using unsinged constants instead of signed
in test allows to remove supressed OI error as well as use test with NI.
Vararg parameter in reflection type is interpreted as covariant
array type against array in expected functional type and as
vararg element type otherwise. For instance having function
fun foo(vararg args: Int): Unit { /*...*/ }
reference ::foo can be passed against expected
(Int) -> Unit,
(Int, Int) -> Unit, etc.
In none of such cases type for parameter in foo's reflection type
should be changed to array.
However, against expected type (IntArray) -> Unit args' type
must become IntArray.
^KT-25514 Fixed
* Extract replacement of IrGetField/IrSetField into a separate
file-level lowering (should reduce the amount of work to linear in
the number of classes rather than potentially quadratic)
* Extract static backing field construction into JvmDeclarationFactory
and move that lowering after PropertiesToFields lowering to reduce
code duplication
Fun fact: this is not actually validated when loading the class; and
even when compiling a java file against a class file that does this,
only javac 9 has a check (1.8 and before simply crashes with
NullPointerException somewhere deep inside the compiler).
- don't box/unbox when value is known to be an inline class
- add unbox state when coroutine resumed
- correctly handle suspension in case of inline class
- add tests
Code in inline lambdas can call multifile part members. These calls are
replaced in GenerateMultifileFacades with the call to the facade member.
Previously this didn't happen though because the lambda body was removed
before the GenerateMultifileFacades phase, which led to
IllegalAccessError in the -Xmultifile-parts-inherit mode (because the
part class is package private in another package).
The problem was that we tried to generate an `$annotations` method for a
property declared in an annotation class. That method is final and has a
body, which is not allowed in annotation classes. Now we're generating
this method in the separate `$DefaultImpls` class as for properties in
interfaces.
Note that the added test still doesn't find any annotations because the
proper support is needed in reflection (KT-22463). Currently it only
checks that no VerifyError happens.
In addition to fixing getContainingDeclaration, change origin of
multifile facades to FILE_CLASS since the corresponding class descriptor
should also be skipped when computing containing declaration. This fixes
the problem with internal function calls in -Xmultifile-parts-inherit
mode (previously we incorrectly mangled the function name in
MethodSignatureMapper), and also fixes coroutine intrinsic calls when
compiling kotlin-stdlib with JVM IR. In the latter case, all intrinsics
(such as isBuiltInSuspendCoroutineUninterceptedOrReturn) are present in
sources, and were previously not detected as intrinsics by the code in
`generateInlineIntrinsic` because the FQ name didn't match: it had an
additional component for the file class name.
Technically a backwards compatibility problem, as the new backend
*consistently* renamed `f$default` on `f` with `@JvmName("g")` to
`g` instead of `g$default`, so it all worked out. However, this
breaks when encountering libraries compiled with the non-IR backend.
Otherwise a local class in a field initializer or anonymous init block
is copied into each constructor of the containing class (because
InitializersLowering calls deepCopy).
Since the code structure no longer resembles the original source code
here, record a custom EnclosingMethod mapping before moving such
classes, and use it in codegen.
Create a separate _ir.txt bytecode listing file for JVM IR, to avoid
duplicate tests and to fix "Codegent tests on different JDKs"
configuration where this test is muted but passes because only execution
is checked there, not bytecode listing.
This directory is skipped in JVM IR test generator, so they won't show
up as failed anymore.
Note that only failing tests are moved to oldLanguageVersions/. Tests
which already pass are still being run. It may be useful not to break
them in case we _do_ need to support some pre-1.3 language feature
switches in JVM IR.
- Added tests to demonstrate broken behaviour: the interaction of inline
functions and callable references with varargs and defaults in various
combinations.
- Refactored InlineCallableReferencesToLambdaPhase to look like and use
some of the infrastructure from CallableReferenceLowering.
- Lifted some of this infrastructure out to be broadly reusable.
Take branching and method calls into account when finding the line
number of the continuation. If there is no line number before
branching instructions or method calls, the following code is
still on the line of the suspend call itself.
This fixes a couple of issues with incorrect line numbers for
multiple throws on the same line or multipe suspend calls on
the same line.
In addition, it avoids the need to spam the method node with
repeated line number instructions in the IR backend.
When creating delegation redirectors for default implementations of
interface functions, `overriddenSymbols` should be repaired across the
whole module, not just a particular file.
With the mangling added in 488418d960, there's no longer any risk in
writing "special" function name ("<anonymous>" in this case) to the
local variable table.
#KT-34356 Fixed
- Extract all backend codegen tests that specifically target behaviour
in to-be-deprecated functionality from language versions < 1.3"
- Remove those tests from the JVM IR test suite.
If "null" was the first entry in an optimizable "when" over enum,
mapRuntimeEnumEntry was called before mapConstEnumEntry, and the
$WhenMappings field was not created. Now both mapConstEnumEntry and
mapRuntimeEnumEntry create this field on the first access
Kristoffer Andersen