In short, some of the bytecode analyzers assume that there could be
no stores instructions into parameter vars with value of different
types (even when the value type is a subtype)
See the issue for details
#KT-19713 Fixed
So (for the most often reproduction case) #KT-19433 Fixed
Before this commit, internal names for nested classes were written as test/Foo/Bar (comparing to test/Foo$Bar in the normal mode), as getting qualified names from such internal names was trivial. But, because of IC, we needed to write class files to the disk, so our decompiler could find such "broken" classes and read it in a wrong way.
FixStack transformation divides on phases:
- Fixing stack before break/continue
- Fixing stack for inline markers/try-catch blocks
After the first stage all ALWAYS_TRUE markers are replaced
with simple GOTO's and if we're skipping break/continue edges
we won't reach the code after while (true) statement.
At the same time it's fine to not to skip them in the second phase
as the stack for them is already corrected in the first phase
#KT-19475 Fixed
Note that this isn't fully correct, consider the following situation:
S : T, T : Any?
=> CS(S, T) = T, but for now it will be T?, which is reliable but not so specific as just T
Problem manifests when a class property name matches a companion object
property name, and class property is referenced in closure context.
#KT-19367 Fixed Target versions 1.1.5
SAM interface wrapper for an argument is required,
if in the function descriptor for SAM adapter
type for the corresponding value parameter
doesn't match type of the corresponding value parameter
in the original (Java) descriptor.
#KT-19251 Fixed Target versions 1.1.5
Do the same thing as for secondary constructor (looks like it was a
workaround for R&I bug that was used only for secondary constructors
for some reason).
#KT-17464 Fixed Target versions 1.1.5
If 'this' (implicit or explicit) was used as an extension receiver,
and the corresponding call required a smart-cast,
this information was effectively lost in "old" resolution & inference,
but is required by "old" JVM BE to generate proper CHECKCASTs.
In "short" version, LHS is always evaluated before RHS,
and RHS may be not evaluated if LHS is null.
So, it makes sense to use "short" version in cases when RHS can't have
side effects (because of more compact bytecode and more opportunities
for code elimination).
Stanislav Erokhin