Class exit node should only unify data flow if it is an anonymous
object, in which case it only has one control flow input. Coversely,
any other class exit node has normal merge semantics for its control
flow inputs, but it won't have any data flow inputs.
manifested in the IC tests by a seemingly redundant lookup,
although theoretically may lead to more severe issues.
Also fixes lookup recording ignoring logic, since the old
one may break with the new us-site file calculation.
#KT-55195 fixed
Fix merge conflict related to renaming of one of the tests
Regenerate tests affected by previous commit
Fix merge conflict related to renaming of one of the tests
Improve wording of the `-Xuse-k2 -language-version 1.*` CLI warning
Improve wording of LV2.0-specific CLI warnings some more
Fix output of one of the tests for -Xuse-k2 deprecation
Tweak wording of LV2.0-specific CLI warnings
* make it clear in every warning that -Xuse-k2 is deprecated
* add a reminder to remove -Xuse-k2 for `-Xuse-k2 -language-version 2.0`
* add both possibly wanted fixes for `-Xuse-k2 -language-version 1.*`
Remove deprecated -Xuse-k2 flag from one of the tests
Merge-request: KT-MR-8449
Merged-by: Stanislav Ruban <Stanislav.Ruban@jetbrains.com>
K1 allows writing access to a public setter of a protected synthetic
property only if the call is inside a subclass. K2 previously allowed
that unconditionally. This changes brings the behavior in line with K1.
^KT-56050 Fixed
In details, this commit changes the following:
- it converts FIR when without branches to empty IR block without when
- it doesn't drop empty else branches in when anymore
^ KTIJ-24385
Temp property to store receiver is generated for `a.b++` expression.
If this property's psi corresponds to receiver expr, then FirProperty
would be found by mapper if receiver is requested.
It works unexpectedly, because FirProperty is normally not expected by expression.
This change set fake sources for generated FirProperty, so it won't be found
by source psi
Currently, designation path requires all intermediate classes to be
regular named classes. This is not certainly possible, especially for
invalid cases such as:
---
class Foo {
object {
class Bar {
fun foo() {} // What's the designation path for 'foo()'?
}
}
}
---
If there was a way to get a parent for a given FIR element, we could
just follow parents until the root FIR node (which would be a FirFile).
'LLFirProvider'/'LLFirProviderHelper' do not normally search for classes
in the 'kotlin.' package. While it has a positive impact on performance,
as 'kotlin.' package is normally forbidden in arbitrary modules,
during designation path calculation we need to handle also invalid
cases.
As 'initialiseVirtualFileFinderServices()' that is run during
test initialization collects transitive dependencies, all of them
must be ready. However, 'KtNotUnderContentRootModuleForTest' has a
built-in dependency provided by 'LLFirBuiltinsSessionFactory'.
Before, 'FirSession's for not-under content root modules were stored in
a project-global service, which essentially means that even if one
creates a non-cached 'LLFirResolveSession', 'FirSession's for such
modules will be cached and reused later. Such exact behavior happened in
sealed class inheritor pre-analysis handler.
In some cases, 'LLFirProvider' doesn't return the existing class, e.g.
when the class is inside the 'kotlin.' package but when the
'-Xallow-kotlin-package' flag is not passed.
This fixes the test 'DiagnosisCompilerTestFE10TestdataTestGenerated.
TestsWithStdLib.testInaccessibleInternalClass()'.'
The old 'FirProviderInterceptor' is too invasive, ad-hoc, and only a
few places in the compiler are aware of it.
This fixes the test 'FirOnAirResolveTestGenerated#testOnAirTypesResolve'
which started failing after the designation path collection refactoring.
The newly appeared class existence check started to fail:
```
val declaration = firProvider.getFirClassifierByFqName(classId)
check(declaration != null)
```
pyos