See https://issuetracker.google.com/issues/162446295 for more context, but in short: this will fail if the project:
* targets java 8
* consumes external libraries targeting a higher version (java 9+). This includes Android SDK 30's android.jar, which targets java 9
* has `mapDiagnosticLocations` enabled for kapt
targetElement is a nullable type, so this seems like a pretty cut-and-dry NPE fix
Compiler plugins can add new source roots to the next compilation round
by using `AnalysisResult.RetryWithAdditionalRoots`. Some plugins added
already existing source roots to this list in some cases. For example,
this is reproducible with `square/anvil` with incremental compilation
(KT-45100, KT-44925). Psi2ir didn't deduplicate the source files before,
which resulted in several classes with the same name linked to the same
symbol. This led to a "symbol is already bound" exception, and in case
of KT-44925, to an additional NPE when we were rendering the class to
display it in the message.
The solution is to deduplicate classes before psi2ir. Note that this
commit has two changes, in CLI and in psi2ir. Any one of these is
sufficient for fixing the problem, however both are made just to make it
more future-proof against new components and/or changes in existing
subsystems (e.g. fir2ir).
In the old JVM backend, similar deduplication was happening in
`ClassFileFactory.registerSourceFiles`, which is why the problem is not
reproducible there.
#KT-45100 Fixed
This function was always experimental, as explained in its kdoc, but it
was introduced before opt-in requirement markers were supported. Thus,
breaking changes (such as in KT-42746) were always expected, and the
`@ExperimentalReflectionOnLambdas` annotation just makes it clearer.
#KT-45486 Fixed
Also:
* Revamp kotlin-native/README.md a bit: make it more up-to-date,
add useful links
* Remove "composite build" section from ReadMe.md:
it was used for Kotlin + Kotlin/Native composite;
this is no longer required.
The .konanLibrary is cached for each declaration for performance purposes, but it is only ok to do it
after IR has been lowered to its final state (or at least when declarations aren't being moved around),
so the fix respects that by only taking .konanLibrary of a IrFile (it is assumed that files stay on their place
during entire backend lowering procedure).
(cherry picked from commit e021138368b48e306ba99a96f47d93ecbe039f4d)
* test_support::TypeInfoHolder to create TypeInfo given payload description
* test_support::Object<Payload> to create objects with Payload and to get them from ObjHeader* after checking that their type_info are layout compatible.
* test_support::*Array<Count> to create various arrays with given length and similarly get them from ArrayHeader*.
- skip metadata target from importing. That lead to metadata
compilations being imported is well -> some source-sets were
participating in metadata compilations as well -> logic for determining
platforms used to consider metadata compilations as well, adding COMMON
platform to set of platforms
Seems like metadata was never needed in import/IDE, and got there purely
by accident
- Use only targets, actually present in the project, as the default
platform. This is needed mostly for corner-cases/miconfigurations, like
orphan source-sets (source-sets which are created but not included into
any configuraion). Still, for those source-sets the tooling is required
to behave properly; presence of non-existing target can lead to various
issues like showing gutters for test runs, which would fail on launch
(because tests for that target actually do not exist)
^KT-37127 Fixed
Note that previously addSourceSetsToCompilation had logic regarding
platforms. The reason for that is that it was the last place where
default soruce sets of compilations could be distinguished from all
participated source sets.
So, to move that logic, we have to preserve default source sets.
Therefore, this commit introduces 'KotlinCompilation.defaultSourceSets',
and 'ImportingContext.isDefaultSourceSet' method.
The rest is trivial code moving/methods introduction
The resulting code isn't really pretty and likely has non-orthogonal
predicates, this will be cleaned-up in the next commits
Yan Zhulanow