The declarations that are printed by the kotlinp tool may be optionally
sorted. That's controlled by `sortDeclarations` flag in
`org.jetbrains.kotlin.kotlinp.Settings`. However, there was no sorting
implemented for `KmClass`es.
This commit fixes that.
^KT-62340
There are two types of type lists that can be met in Kotlin metadata
and that should be treated in a different way:
1. The order-sensitive type list. Examples: Context receivers.
2. The order-insensitive type list. Examples: Class supertypes,
type parameter upper bounds.
With (1) swapping elements in list causes source and/or ABI
incompatibility, while with (2) it does not. If it happens that we
compare two type lists in Kotlin metadata of (2) kind, we should group
them using the key computed with the help of
`dumpToString(dumpExtras = false)` instead of just the index as we do
in case of (1).
^KT-62753
There are few reasons why fake overrides should not be printed:
1. Fake overrides are not serialized in metadata. So, if one will run
"dump-metadata" they won't see there any fake overrides. Neither
their signatures. It would be inconsistent if
"dump-metadata-signatures" command would show them.
2. In order to properly build fake overrides the KLIB tool needs
the dependency library with the super class or interface, which
is not available unless this is stdlib or one of the built-in
Kotlin/Native platform libraries. KLIB tool does not support any
reasonable means for passing dependencies through CLI arguments.
If such possibility is added in the future, it would be the
right time to support printing fake overrides (probably under
a separate CLI option).
^KT-62340
Don't print signatures computed for private declarations. This
effectively means that only signatures for `public`, `protected`,
and `internal` declarations will be printed.
Note: This new behavior is intact with `dump-ir-signatures` command.
^KT-62340
With `KlibToolOutput` it's possible to forward the output of each
KLIB tool's command to either system stdout/stderr streams or to
some buffers which then can be read by tests.
^KT-62340
This is an option that allows running the commands that support it
in a special "test mode". The "test mode" means (but not limited to)
that a command may, for example, sort the output which is unsorted
by default, and this way guarantee stable output. This is essentially
helpful for tests, which rely on the command output.
^KT-62340
What's done:
- Reduce the visibility of all entities inside the module to `internal`
or `private`. Only leave `fun main()` as `public` because it's the
single legal entry point.
- Add comments to few classes related to "contents" command that they
should be removed together with the command after 2.0.
- Drop useless `DeclarationHeaderRenderer` interface.
- Rename: KlibToolLinker -> KlibToolIrLinker
^KT-62340
This commit includes:
1. New subproject :tools:kotlinp-klib that contains the facade for
rendering the metadata in "kotlinp" format given just
KlibModuleMetadata
2. A tool for computing (external) IR signatures that are not directly
available in metadata: KlibKotlinpExternalSignatureComputer
^KT-62340
This helps to have a primitive array instead of an array of primitives
as the type of vararg value parameter. Also, this prevents
ABI-incompatibility with the libraries (KLIBs) compiled with K1.
^KT-65588
Change the package name for all *.kt files inside
the :tools:kotlinp-jvm subproject from "org.jetbrains.kotlin.kotlinp"
to "org.jetbrains.kotlin.kotlinp.jvm"
^KT-62340
The "common" subproject keeps only backend-neutral logic and depends
only on :kotlinx-metadata library. It takes the name of the former
project - :tools:kotlinp
The "jvm" subproject depends on the "common" one and also depends
on :kotlinx-metadata-jvm. It gets the new name - :tools:kotlinp-jvm
There is a lot of touched files in this commit. The majority of them
is just moved files (tests, test data, etc).
Only the following files were actually modified:
.space/CODEOWNERS
build.gradle.kts
libraries/tools/abi-comparator/build.gradle.kts
libraries/tools/kotlinp/build.gradle.kts
libraries/tools/kotlinp/jvm/build.gradle.kts
plugins/kapt3/kapt3-compiler/build.gradle.kts
settings.gradle
^KT-62340
This refactoring includes the following:
1. Separate backend-neutral part of the metadata printer `Kotlinp`
from the part that accesses JVM-specifics: `JvmKotlinp`.
2. Introduce `org.jetbrains.kotlin.kotlinp.Printer`: The component
that renders both individual `Km*` nodes and the whole metadata
tree, and does so with the proper indentation.
^KT-62340
This is done to make the output of "dump-metadata" command consistent
with the output of "dump-metadata-signatures" and "dump-ir-signatures",
which dump signatures for all non-private declarations.
^KT-62340
Fir-based lazy type alias is required in IR text tests to check that
lazily loaded IR is equivalent to deserialized IR (module bodies)
and does not cause ABI compatibility issues in KLIB-based backends.
^KT-65033
Support new metadata attributes that were added to Kotlin metadata
but are not yet supported in `MetadataDeclarationsComparator`:
- DNN type flag
- context receivers
- underlying type & property name for inline classes
^KT-62753
Invocation of Logger.fatal() may cause severe side effects such as
throwing an exception or even terminating the current JVM process
(check various implementations of this function for details).
The code that uses Logger.fatal() sometimes expects a particular kind
of side effect. This is totally a design flaw. And it's definitely not
a responsibility of Logger to influence the execution flow of
the program.
IrMangledNameAndSignatureDumpHandler can dump either signature mangled
names by Fir or by descriptors, but not both at the same time. And
mangled names should be equal to preserve ABI compatibility.
^KT-57788
- Make the messages that are reported by KLIB resolver prettier
- For those messages that affect the resolve process add
prefix "KLIB resolver: "
- Don't log warning on duplicated libraries on the classpath. This
does not make any sense.
^KT-63573
This function is only used for probing if a library at the given
path exists. Any errors or warnings logged by this function could
only bring a user to a confusion. So, it's better to avoid any
logging here.
^KT-63573
The reason of this change is to make messages (especially warnings)
that are reported by the KLIB resolver become visible to the end user.
This can be achieved to forwarding such messages to the appropriate
compiler's components such as
`org.jetbrains.kotlin.cli.common.messages.MessageCollector` and
`org.jetbrains.kotlin.ir.util.IrMessageLogger`.
Also: The default `DummyLogger` should be used as minimal as possible.
Because it just forwards messages to the standard output (console)
where they can remain unattended. When the compiler is executed
from the Gradle plugin such messages appear only in DEBUG Gradle's log.
^KT-63573
Rework rendering of kt-like dump and signatures dump in order to avoid
unstable blank line between declarations of the same level:
1. No blank line for the first declaration inside the member scope of
the class.
2. Always a single blank line between each two subsequent declarations
inside the member scope of the class.
Making enum class-related tests unmuted requires implementing
a special "compatibility" mode for IR tree dumper to filter out
fake override declarations leaking from java.enum.Enum and
kotlin.Enum (JVM-only) classes.
1. Local declarations don't participate in IR-linkage, because they
can be referenced only inside the same body -> can be dropped
from IR text tests.
2. Mangled names for private declarations computed by descriptors/fir
are actually not used anywhere (they are recomputed by IR
immediately before serialization of IR). But sometimes such
mangled names diverge between K1 and K2 -> don't check them, but
always check mangled names computed by IR even for private
declarations.
3. Also: Drop DUMP_LOCAL_DECLARATION_SIGNATURES test directive.
^KT-57428 Obsolete
^KT-57430 Obsolete
^KT-57434 Obsolete
^KT-57778 Obsolete
^KT-57775 Obsolete
There are a lot of cases when a DEF file for a K/N platform library
has incomplete set of dependencies in 'depends=' property in
the manifest file. This leads to use of "KLIB resolver" - the
component inside the compiler that is responsible for finding the
missing libraries and implicitly (i.e. invisibly for the end user)
adding them to the set of dependencies of the compiler invocation.
The goal of this commit is to fix 'depends=' in manifest files so that
they should reflect the REAL set of dependencies, and as a result don't
trigger use of "KLIB resolver". This way we could be more prepared to
removed in the "KLIB resolver" in the future.
^KT-61098
This test can be used to make small performance measurements of
the core commonizer on occasional basis. This test is not intended
for regular execution, e.g. at the CI server.
^KT-63159
Dmitriy Dolovov