- Move the following from 'deserialization' to 'descriptors':
NotFoundClasses.kt
AdditionalClassPartsProvider.kt
ClassDescriptorFactory.kt
PlatformDependentDeclarationFilter.kt
findClassInModule.kt
- Move the following form 'descriptors' to 'deserialization':
BuiltInSerializerProtocol.kt
builtInsPackageFragmentProvider.kt
- Extract a marker interface from BuiltInsPackageFragment and move its
implementation to 'deserialization'
- Change the type of parameters in PlatformDependentDeclarationFilter
and AdditionalClassPartsProvider to ClassDescriptor
This will help in getting rid of the circular dependency of
'descriptors' <-> 'deserialization'
The problem was that an overload of CodedOutputStream.newInstance
without size uses 4000 as a default value, thus allocating
a 4k-sized byte array.
At the same time the array remained effectively unused
in the most cases since it only used for storing data
for unknown fields.
Because such arrays are being created for each read of
protobuf message, during compilation of IntelliJ project
it was producing 25% of redundant memory traffic.
Of course these arrays were all in the young gen, but still
they have some effect on GC
This fixes KT-17001 because now 'header' modifier is loaded correctly
for deserialized members and the standard disambiguation in
OverloadingConflictResolver.compareCallsByUsedArguments takes place,
where header members are discriminated against the corresponding impl
members
#KT-17001 Fixed
Some IDE features relates onto that same descriptors will remain same
between resolve calls
Fix it to be true for TypeAliasConstructorDescriptor's
#KT-16265 fixed
Notes: if arity == 0, it means that type was `suspend () -> String` => arguments.size == 1, and it means that SuspendFunction0 has only one type argument and it is return type.
Since now `suspend (Int) -> String` will be serialized as `(Int, Continuation<String>) -> Any?` + suspend flag.
Before this change such type serialized like this: Function2<Int, String> + suspend flag. And yes, type `Function2<Int, String>` isn't correct, because Function2 expect 3 type arguments.
We have special logic for this case and we deserialize such error-written types correctly.
(cherry picked from commit 3518cbe)
Instead of multiple .kjsm files for different classes and .kotlin_string_table,
.kotlin_file_table, .kotlin_classes files for each package, serialize the
contents of each package to a single foo/bar/baz/baz.kjsm file. The short name
of the file is the last segment in the FQ name of the package, or
"root-package" if the package is root.
There are two main reasons for this change:
1) Such structure takes less space, is more IO-friendly and will not cause
multiple exceptions as the old one, where we sometimes tried to read
non-existing files
2) This is exactly the same format that is used to serialize built-in
declarations (.kotlin_builtins) at the moment, which will allow us to reuse
some code
Also write a separate Header protobuf message to the beginning of the .kjsm
file. This will be used as arguments of the kotlin.Metadata annotation are used
in the JVM-specific parts of the compiler: to be able to provide some general
information about the binary file without parsing the whole protobuf data.
This commit breaks decompiled text & stub builder consistency tests. This is OK
because they're removed in a future commit.
Fixes EA-79605, EA-81947, EA-84277 and maybe EA-86787
#KT-10894 Fixed
#KT-14124 Fixed
#KT-15755 Fixed
Note that now DeserializedClassDescriptor.getSealedSubclasses works a lot
faster than before, for all newly compiled sealed classes except empty ones. It
may be optimized further if we look at the metadata version of the file the
class was loaded from, however it's not easy currently because
DeserializedClassDescriptor is declared in common (non-JVM) code and has no
direct access to the binary version information.
This will also allow to add a reflection API to get subclasses of a sealed
class
#KT-12795 Fixed
'SuspendFunction$n' class descriptors are created on demand by KotlinBuiltIns (and cached).
On serialization, types constructed with 'SuspendFunction$n' are written as 'Function$n' with extra flag (SUSPEND_TYPE).
On deserialization, corresponding 'SuspendFunction$n' classes are used.
This is a way for future compilers to cause previous compilers to report
diagnostics on usages of some declarations. Diagnostic can have a message
(and/or error code), level (error, warning, or completely hide the declaration
from the resolution), and Kotlin version, since which the diagnostic should no
longer be reported
Similarly to pre-release classes, load metadata for the class anyway and allow
the resolution to select it as the result and prohibit its usage in the end
with the special diagnostic reported in MissingDependencyClassChecker
Before creating a MetadataPackageFragment, check that the corresponding
directory (across the classpath) contains at least one .kotlin_metadata file.
Otherwise we're creating packages for every simple name queried during the
resolution and sometimes prefer a (empty) package to the existing class, for
example when the latter class is star-imported
Extract AbstractDeserializedPackageFragmentProvider out of
JvmBuiltInsPackageFragmentProvider and implement it a little bit differently in
MetadataPackageFragmentProvider. The main difference is in how the package
fragment scope is constructed: for built-ins, it's just a single scope that
loads everything from one protobuf message. For metadata, package scope can
consist of many files, some of which store information about classes and others
are similar to package parts on JVM, so a ChainedMemberScope instance is
created.
Introduce a bunch of interfaces/methods to deliver the needed behavior to the
'deserialization' module which is not JVM-specific and does not depend on the
compiler code: MetadataFinderFactory,
PackagePartProvider#findMetadataPackageParts, KotlinMetadataFinder#findMetadata.
Note that these declarations are currently only implemented in the compiler; no
metadata package parts/fragments will be found in IDE or reflection
Alexander Udalov