This allows the compiler to limit imports from "magic" packages like
cnames.structs to forward declarations actually present in dependent
cinterop klibs.
The manifest already had `exportForwardDeclarations` field, but
- it didn't include Objective-C forward declarations (@class, @protocol)
- it has a slightly different meaning (the export list enables importing
the listed declarations through the interop package of the library,
while the include list shouldn't do that), so we better avoid mixing
them.
The test relies on the fact that any name can be imported from a "magic"
package like `objcnames.classes`.
This was the case for K1, but likely won't be for K2: the compiler
should allow importing only those names that are actually
forward-declared in a cinterop library.
This commit changes the test that way. In particular, moves it to a
different task that has a cinterop dependency.
It affected the changed test in the following way: "strict metadata
version semantics" is a flag which is written to the .kotlin_module file
between the version and the protobuf data, see `ModuleMapping:73`. But
flags are written only for Kotlin 1.4+, i.e. if `isKotlin1Dot4OrLater`
returns true. So in the test, the flag was not written, thus we did not
detect that the module actually has incompatible metadata, and thus
incorrectly did not report an error.
It also has an effect that version requirements for nested classes will
now be written correctly to protobuf with metadata version 2.0+.
We have several tests which tests how compiler with LV = LATEST_STABLE
can consume binaries compiled with LV = LATEST_STABLE + 1 with
different CLI flags. For LV = 2.0 we have special rules for those flags,
so testdata was changed
After switching LV in repository to 2.0 this testdata should be rolled back
Also consider delegated properties which are optimized since Kotlin
1.7.20 because of KT-23397, where backend doesn't generate a field
`foo$delegate`, but generates a method `getFoo$delegate` instead.
Using `ReflectionProperties.lazy` is incorrect because it allows several
threads to observe different resulting values if they're computing it
simultaneously (unlike `lazy(PUBLICATION)`, which always returns the
value that "won the race").
In the case of property delegates, for example, if we're invoking
`isAccessible = true` and then `getDelegate()` concurrently, it might
happen that when some thread invokes `getDelegate()`, it gets the
underlying Field object which was written by another thread and which
has not yet been made accessible, leading to
IllegalPropertyDelegateAccessException.
#KT-27585 Fixed
It's hard to achieve that without implementing part of `externRefToAny` using an intrinsic. It should be rewritten when all VMs and tools update to the latest spec.
Also, stop using (deprecated) instructions unsupported by SpiderMonkey.
#KT-56166 In-Progress
Previous implementation set Nth ModuleDescriptor dependencies to the
first N modules. This is not always correct: for example, when producing
a metadata klib (-Xmetadata-klib) for HMPP, metadata dependencies might
have inexact or missing depends= in their manifests.
This commit sets ModuleDescriptor dependencies to all other
deserialized modules. The same approach is used for K1. That way, the
implementation is more reliable.
^KT-56071
Old implementation expected that builtIns should be based on a library
with no dependencies, usually stdlib. But that doesn't work properly in
some cases, e.g. when producing a metadata klib (-Xmetadata-klib) for
HMPP. In that case, some of the metadata dependencies can have no
depends= in the manifest, leading to the compiler choosing such a module
instead of stdlib.
This commit makes it look for a library that is explicitly stdlib
instead.
^KT-56071
There were still some cases where the generated function reference class
would contain references to generic parameters that were not in scope.
One example of that was `Array<*>::get`. Before this fix the generated
function reference class would still contain a reference to the
type parameter declared in the `Array` class.
Also, the logic of generifying the generated function reference class
was incorrect. For example, for function `fun <A, B> foo()` if we had
a reference `::foo<T, T>` (where `T` comes from the outer scope),
the class would still contain two generic parameter, despite that only
one generic parameter from the outer scope was referenced.
This is also fixed here.
The existing threshold for determining between medium and large page
does not take the dummy block into account, meaning that an allocation
on the threshold will lead to a loop that only terminates when the
process goes out of memory.
Co-authored-by: Troels Lund <troels@google.com>
Merge-request: KOTLIN-MR-618
Merged-by: Alexander Shabalin <alexander.shabalin@jetbrains.com>
Sebastian Sellmair