Kapt task incremental annotation processing cache directory
was using @LocalState annotation that was added in Gradle 4.3.
However, minimum supported version is 4.2, so this commit registers
this directory as output of a task for pre 4.3 versions. This is safe
as caching is enabled only for Gradle 4.3+.
Test: verified manually
Because Kapt Gradle task uses PathSensitivity.NONE for the input representing
the classpath structure, there are cases when paths of the files may change,
but because content is the same, there will be no incremental changes.
Classpath snapshot comparison did not handle this case correctly, but this
commit fixes that. In more details:
- classpath entries with the same path and that are not reported as changed
will have their information loaded from the previous snapshot
- any other entry (changed path or changed content) will be reloaded
Test: ClasspathSnapshotTest
The test is not designed to fail if the definitions are not discovered -
the script files have completely custom extensions and therefore ignored.
So checking for the warning in the build log is enough.
#KT-32697 fixed
also adding a short sleep in tests, otherwise in some cases the
classloader was not able to find class in the jar.
Fixes script to jar saving tests on windows
This commit adds a method to get the value of org.jetbrains.org.objectweb.asm.Oopcodes.API_VERSION.
In general, this should not be a problem, but some Gradle plugins package
Jetbrains ASM version. At runtime, their version may be loaded which may be
different than the one this code was compiled against.
Use @LocalState to mark incremental annotation processing cache directory.
This is to make sure this is not stored in the build cache as it contains
absolute paths and cannot be shared between machines.
Artifact transform that uses ASM to analyze KAPT classpath stored absolute
paths in the output artifact. This resulted in remote build cache misses.
This commit changes how analysis is stored. Actual analysis file is the
output of the transform, and there is not need to use a marker file any more.
Output does not store the classpath entry absolute path. Instead, it uses
task action incremental information to find analysis outputs that changed.
Additionally, class that handles analysis snapshot comparison has been
simplified, and lazy loading of the structural information is handled in
a more straightforward way.
The tests were broken some time ago by applying script isolation, so
it was necessary to add base libraries into the classpath.
Also refactored some classpath building utilities.
The kotlin-scripting-compiler-impl jar is used in the idea plugin and
therefore should not depend on the cli parts of the compiler to avoid
dependency from the plugin to the kotlin-compiler.jar.
Therefore the cli-dependent parts were moved to the scripting plugin
jar, which is used only in cli compiler based environments.
Also implement required abstractions to allow this movement and
drop some redundant dependencies to the cli parts in other projects.
- implement error result
- refactor other result classes
- implement handling in the script evaluation extension - also restores
previous script error reporting functionality
- add possibility to customize result fileds in script and REPL
- refactor result calculation in the backend: cleanup, rename (since
it is not only about REPL now)
Do not register an extension, use an extra property instead. This way,
it is less visible for a build author.
Hotfix for commit c292380aff
Issue #KT-29693
* Interface and abstract properties are now checked, too.
For them, copy the @Input/@Internal annotations from the
implementations (luckily, they were all consistent).
* Annotating a private property without a public accessor is not not
allowed. Remove one such annotation.
Fix dependencies that are added in non-published source sets, which were
omitted from the requested dependencies since their configurations were
not added to the extendsFrom set of the merged configurations.
Also, don't resolve both merged configurations (compile and runtime) in
GranularMetadataTransformation, as the IDE can anyway import only the
compile-scoped dependencies from api & implementation.
Issue #KT-32225 Fixed
Fix incorrect source sets visibility when a dependency is added to a
non-root source set and its visible source sets are mistakenly treated
as already visible by the dependsOn parents and are thus not extracted
from the metadata package.
To simplify the logic, reuse the GranularMetadataTransformation
instances from the dependsOn parent source sets: they can provide the
requested dependencies sets for the parents and their transformation
results, so that excluding the dependency source sets which are seen
the parents gets simpler.
The issue is then fixed by naturally excluding only the source sets
which are mentioned in the parents' transformation results.
Issue #KT-32204 Fixed
Ivan Gavrilovic