Invert the logic of IR/FIR2IR-based generators: change the CLI argument
to -Xuse-fir-fake-override-builder, test directive to
USE_FIR_BASED_FAKE_OVERRIDE_GENERATOR, etc.
The changes in test data caused by using IR fake override builder by
default are in the subsequent commit.
#KT-61514
This commit explicitly splits raw fir builder in lazy mode on
AST and stubs.
By default, in tests we have an AST tree loaded, so all our
`disallowTreeLoading` is useless in unit tests mode
^KT-64646 Fixed
^KT-64898
^KT-64899
^KT-64900
^KT-64901
^KT-58765
Now that JDK 21 was released as stable, we can remove separate test
configurations for testing the compiler with JDK 21 and reintegrate
the corresponding test data into standard test configurations.
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).
After this commit there are three different versions of
IR source range tests: classic (K1), FIR/PSI (K2), FIR/LT (K2).
Since 5 tests behave differently for FIR/PSI and FIR/LT,
in this commit their test data was set to FIR/LT state,
so relevant 5 tests are failing for FIR/PSI right now.
They will be fixed in two subsequent commits
Related to KT-59864, KT-60111, KT-59584
There was only one real inheritor of AbstractFirBaseDiagnosticTest,
so there is no need in the whole hierarchy of abstract test runners
Also AbstractFirOldFrontendLightClassesTest is not intended to check any
diagnostics or CFG dumps, so this logic was removed (they still can be
checked using non-legacy AbstractFirDiagnosticTest from the main test
infrastructure)
In the vast majority of tests, diagnostics are reported at a slightly
different location in LT vs PSI. This is because in the light tree mode,
backend errors are reported basically on the start of the IR element,
which is for example the keyword "class" for a class and not its name as
in K1 or in K2+PSI. Similarly, the end of the diagnostic is at the
ending offset of the IR element.
Normally it would be a bit user-unfriendly to highlight the whole class
as red, starting from the keyword "class" and ending with the closing
brace "}". But remember that these are backend diagnostics in the K2+LT
mode, which is only possible in the compiler scenario, not in the IDE.
And in this case, the ending offset doesn't matter at all because it's
not presented to the user, and minor change in the starting offset is
not a problem as well.
There are some tests which legitimately fail in the LT mode because the
corresponding diagnostics haven't been supported. They will be dealt
with in subsequent commits:
testPropertyInlineCycle
testInlineCycle
MultifileClasses.testJvmSynthetic
testSuspendInlineCycle
testIndirectInlineCycle
#KT-59586
Add some more filters on private/synthetic stuff (which doesn't matter
in practice) to make full and light analysis mode dumps as similar as
possible, so that all existing tests will pass for JVM IR. Unmute some
tests which were failing with the old JVM backend.
Tests on repeatable annotations are muted because in full analysis,
annotations are wrapped into the container (e.g. `@A(1) @A(2)` ->
`@A$Container(A(1), A(2))`), but they are no in the light analysis mode.
So there's always going to be a difference for these tests between full
and light analysis, unless we're going to change behavior of kapt, which
would be a kind of a breaking change.
#KT-58497 Fixed
According to
`FirNativeCodegenBoxTestGenerated.testNestedClassesInAnnotations`,
the annotation
`kotlin.internal.PlatformDependent` is
unresolved reference.
^KT-58549 Fixed
After it's released, there would be no need in them, but right now
they are unavailable through the toolchain, so we can't require it.
See KT-58765 for tracking
But there should be a dedicated Build configuration with JDK_21_0 env
properly set.
^KT-58716 Fixed
We've muted some irText tests on JS in the previous commits because
test expectations in some tests are different when targeting JS.
AbstractKlibJsTextTestCase tests use a different logic:
they compare the dump of the deserialized IR with the frontend-generated
IR, not with the expectations in text files, thus muted tests weren't
actually failing.
Here we introduce a temporary fix, namely
a separate // IGNORE directive for klib tests.
When klib tests are moved to the new test infrastructure, there will be
no need to do this.
Iaroslav Postovalov