It is easier to introduce a new lowering so the codegen will emit code for the old
tail-call optimizer to understand. Also, this is more flexible and would allow to
optimize cases, which are now feasible with the old optimizer.
Note, that because of bytecode inlining, we cannot replace the old one, but we cannot
emit code, that is simpler for it to optimize.
Since the markers replace ALOAD 0 as continuations, passed to suspend calls, in
JVM_IR we do not need this, since in JVM_IR all inline lambdas are static
functions.
When called by reflection the bit mask will be generated
discounting dispatch/extension receivers. Make sure that the
interpretation of the bit mask is consistent for direct and
reflective calls.
In addition, this also fixes the modifiers on java 8 parameter
metadata for the dispatch and extension receivers for these
inline class methods.
An uninferred parameter stub may leak through calculation of CST(Inv<Uninferred Stub>, Nothing) into a result type.
A stub type in the result type means a type error. So we can afford recalculating
CST with stub-containing types filtered out, since its an error anyway.
This prevents stub types leakages and helps with reporting type error diagnostics.
KT-35914 Fixed
KT-35943 Fixed
DiagnosticsTestWithStdLib now require stdlib only.
Move test data accordingly.
The only test that required kotlin-reflect is transformed into
codegen box test.
The problem in the added test was that a suspend lambda was represented
by a function reference with a bound argument for the ObjectRef value,
and the corresponding parameter was not the first parameter of the
referenced local function. This happens because
LocalDeclarationsLowering lifts the local function up and adds a
new parameter for the captured ObjectRef (which is bound at the call
site), but the original receiver parameter remains the first unbound
parameter. So, it's no longer correct to rely on the fact that all bound
parameters of a function reference are located in the beginning of the
parameter list, which was kind of assumed in the `withIndex` call in
`AddContinuationLowering.addCreate`.
Old version handled only private companions correctly.
Some situations require multiple companion object accessors
(including accessors for protected companion objects from supertypes)
to be generated in the corresponding class.
This is important for calls using reflection as the reflect
library assumes this ordering of arguments.
It would be nice if this could be handled in the lowerings.
Currently AddContinuationLowering is after
DefaultArgumentStubGenerator. If we could add the continuation
first and then do default stub generation maybe we could avoid
the reshuffling introduced in coroutine codegen in this change.
The first change here is swapping the isCompatibility and hasJvmDefault
checks. Otherwise behavior could be different depending on -Xjvm-default
mode even for non-JvmDefault declarations, which makes little sense.
Another change is avoiding generating $default stubs for fake overrides
in interfaces, which replicates the behavior of the current backend.
(Note that this change also fixes the first problem on the newly added
tests, but the first change seems useful anyway.)
This only reproduced when compiling (technically incorrect) code in the
standard library, where private functions monitorEnter/monitorExit are
accessed from another file, and their names with suffixes are not
recognized as intrinsics which should be replaced by
monitorenter/monitorexit JVM bytecode instructions.
SyntheticAccessorLowering was initially implemented under the assumption
that any access to an invisible declaration will cause an accessor to be
generated _in the same file_. Moreover, it's declared in the group of
phases that are performed by file.
But this assumption is incorrect for constructors which need to be
hidden (those which take parameters of inline class types), since such
constructor is public and can be called from anywhere. In this case,
SyntheticAccessorLowering actually generated a new accessor for the
hidden constructor for each (!) source file where that constructor is
called, which led to ClassFormatError because of the class file having
multiple methods with the same signature. The internal `functionMap`
cache didn't help because it's not shared among phase instances for
different files (well, it helped to generate not more than one accessor
per usage-file).
In this change, we use the global cache, stored in JvmBackendContext,
for accessors to hidden constructors. Note that after this change, calls
to hidden constructors are always transformed to the corresponding
accessor in SyntheticAccessorLowering right away, but that accessor
might be orphaned for a while (not declared in any parent's
declarations). Only when SyntheticAccessorLowering encounters the
original constructor which needs to be hidden, it adds the accessor
beside it.
The test is sensitive to the file order, so both variants are added.
Alexander Udalov