classes, instead of MemberScope.
The primary motivation was to fix issues around type-mapping for inline
classes in FIR, which uses wrapped descriptors that have empty
MemberScopes.
- Allow participating subtypes of functional types in conversions
- Fix several subtle inconsistencies
- Place logic about conversions at one place
Now conversions operations have two stages: before usual subtyping
check and after one. This is needed to support conversions of
subtypes (of functional types, for example). First, the compiler
checks if it possible to resolve an argument without conversion and
only then it tries to perform conversion.
Note that it'd be incorrect to perform conversion eagerly as it can
change resolve (Runnable & () -> Unit <: KRunnable), plus we can't
guess whether conversion is needed at all as it's important not to
look into supertypes if resolution doesn't actually needed it
#KT-36448 Fixed
#KT-37574 Fixed
#KT-38604 Fixed
Before this commit we considered !isOverride as a sign that
function / field / accessor has no overridden symbols.
However, it's false for deserialized, because isOverride
is always false there.
This commit fixes 68 BB tests but breaks 25 BB tests (not yet muted)
Attributes are used to name continuation classes and are generated
before inline classes processing. During the processing, for override
functions in inlined classes, the compiler generates
STATIC_INLINE_CLASS_REPLACEMENT function with body of the override.
The override's body is replaced with delegating call to
STATIC_INLINE_CLASS_REPLACEMENT. However, since we need to keep the name
of the continuation class, we copy attributes from the override to
STATIC_INLINE_CLASS_REPLACEMENT. This leads to attribute clash during
AddContinuationLowering.
So, to fix the issue, do not use the attribute of
STATIC_INLINE_CLASS_REPLACEMENT in original->suspend map.
As an optimization, do not generate continuation for the override
function.
This fixes the weird cases when a class gets overwritten by an imperfect
copy, reduces the number of classes in the output if an inline function
contains an inline call that causes it to have regenerated anonymous
objects, and makes inlining of same module functions a bit faster in
general. On the other hand, this may increase memory footprint a bit
because classes cannot be flushed to the output jar, as the inliner
would not be able to locate classes for anonymous objects if they have
already been unloaded from memory.
For a class literal Type::class we are resolving Type as a constructor,
getting all diagnostics (about missing arguments, for example) and then
just not committing this trace with errors
#KT-37626 Fixed
If an inline class is mapped to a reference type (or an array), it's Ok
to treat JVM view on a suspend function as returning a value of
corresponding inline class (although in reality it returns 'Any?'
because of COROUTINE_SUSPENDED).
even when arguments are to be rearranged.
Lambdas have no side effects, and storing them in temporary variables
prevents processing in the backend
(such as inserting continuation parameter in AddContinuationLowering).
The fields containing crossinline lambdas should be package-private to
avoid generating synthetic accessors, which break object regeneration.
Note that the inline methods cannot actually be called, as call sites
will attempt to read the captured lambda from a field through a *copy*
of the local containing the object, so these reads will not be inlined,
causing an exception at runtime:
inline fun f(crossinline g: () -> Unit) = object : I {
inline fun h() = g()
// effectively `val tmp = this; return tmp.$g()`:
override fun run() = h()
}
f {}.run() // NoSuchFieldError: $g
This particular example can be fixed by reusing locals for receiver
parameters in IrInlineCodegen, but explicitly assigning `this` to
another variable and calling an inline method on it will break it again.
(This is only applicable to the JVM_IR backend, as the non-IR one fails
to generate `f` at all for some other reason.)
Now FE IR -> BE IR transformation is performed in multiple stages
controller by Fir2IrConverter. Stages are
* files & classes registration
* supertypes & type parameters handling
* functions & properties signature generation
* body generation
After each step we have guarantee (with exception of local classes &
type inference combination, and external symbols) that required symbols
(class/function/property/variable/type parameter)
are already bound to real declarations and have correct parents.
This commit also fixes incorrect parents for local classes
Even if a function is known to be tail call because it's a compiler
generated bridge, the tail return might still need to be added in case
of Unit return type.
Jinseong Jeon