K2-only issue. In an incremental build, sourceSet boundary isn't
preserved in certain conditions:
SourceSet A depends on SourceSet B
A and B overload a function, for example:
A -> fun foo(p: Parent) // (1)
B -> fun foo(c: Child) // (2)
If some source file in A is calling foo(..), only (1) is supposed to be visible
for that call site.
However, if
a) it is an incremental build,
b) the declaration of (1) is not a part of the compilation set, and
c) call to foo(c: Child) is applicable,
then (2) would be called from the generated code. So, the build result is not
consistent between the full build and an incremental build.
As a workaround, we fallback to a non-incremental build, if any source from A
needs to be compiled.
To enable "risky" incremental builds, use Gradle property
kotlin.internal.incremental.enableUnsafeOptimizationsForMultiplatform=true
^KT-62686
^KT-63837 Fixed
Merge-request: KT-MR-13695
Merged-by: Evgenii Mazhukin <evgenii.mazhukin@jetbrains.com>
This is already the case for Kotlin/JS.
This is required for fixing KT-63670: `linkViaSignatures` implies that
`SymbolTable` is used in FIR2IR, and in case if two declarations
have the same `IdSignature`, we have a crash in FIR2IR:
exception: java.lang.IllegalStateException: IR declaration with
signature "/foo|foo(){}[0]" found in SymbolTable and not found in
declaration storage
If we're not using SymbolTable, FIR2IR finishes successfully, and we can
safely proceed to the KLIB serialization stage, at which point we
can show the user a diagnostic about clashing signatures.
Since we've disabled SymbolTable, we also needed to fix the logic that
relied on its presence. Fortunately, such logic only includes
filtering out unused libraries. Instead of iterating over symbols in
SymbolTable, we iterate over cached symbols in Fir2IrDeclarationStorage
and Fir2IrClassifierStorage, which seems to do the trick just as well.
Apparently, using such an approach is a popular case especially for
configuring 'dependsOn' parameter. And we should simplify user migration
to more restricted DSL.
^KT-64722 Verification Pending
Invocation of Logger.fatal() may cause severe side effects such as
throwing an exception or even terminating the current JVM process
(check various implementations of this function for details).
The code that uses Logger.fatal() sometimes expects a particular kind
of side effect. This is totally a design flaw. And it's definitely not
a responsibility of Logger to influence the execution flow of
the program.
We should provide valid getReferenceName as it is used
in many places (e.g. from com.intellij.psi.impl.PsiImplUtil.findAnnotation)
^KT-61605 Fixed
^KT-64605 Fixed
While not beeing final solution, this is closer to what
we want to have in the end. Enabling on non-JVM targets
would help better testing.
Enabling in JVM is now not possible yet, as some of the bugs are
not fixed yet (check KT-61360 for details)
^KT-62476
Because of KT-64743 there is unexpanded type-alias in a place we don't
expect it to occur. To workaround this, we just expand typialiases in
one more place.
^KT-64743
Ilya Kirillov