When replacing an enum entry reference with `this`, you need to take
`this` from the function's `dispatchReceiverParameter`, not the class's
`thisReceiver`. Otherwise the code generator fails to find the reference
among accessible variables.
This fixes the problem where compiling a class initializer that contains
a call to an `assert`ing function in a separate module causes the
assertion to always be enabled (i.e. the attached test used to fail in
CompileKotlinAgainstInlineKotlin mode).
where "fix" means "work around a bug that is much harder to fix". When
lambdas are extracted into the nearest class, captured type parameters
are not transformed into type parameters of the new method; because of
this, to check whether a reified type parameter is captured by a method
we need to check the outermost named function, not the current function.
Rely on the frontend weeding out cases that are not supported.
In psi2ir, introduce all the parameters before processing default
values.
Change the DefaultArgumentStubGenerator to generate code that
matches the behavior of the current backend.
and fake lambda types, too. (But those only matter for debugging.)
Also, share object name generators between methods with the same name to
avoid rewriting objects from one with objects from the other.
CheckLocalVariablesTableTests will now check the validity of
the locals table against types of locals computed based on the
bytecode.
These checks and the new destructuringInFor test act as a
regression test for the changes in
https://github.com/JetBrains/kotlin/pull/2613
These checks also caught a similar issue for destructuring
lambda parameters, where the local is introduced before the
value has been written to the local slot. This change also
fixes that.
Finally, this change fixes the asmLike tests to correctly
look up the name of parameters in the locals table.
If the delegated property operators involved are inline, and delegated
property metadata parameter is not used (which is often the case, e.g.,
'lazy'), we can skip those properties in metadata generation.
NOT implemented: special case when only 'kProperty.name' is used by the
corresponding delegated property operators.
Also a sneak fix for KT-34060.
Steven Schäfer