In the old backend, BindingContextSuppressCache is used (which is now
created explicitly in GenerationState), which looks up `@Suppress`
annotations on elements right before reporting the diagnostic. In JVM
IR, we clear the binding context after psi2ir, so this approach doesn't
work. This change provides another implementation of KotlinSuppressCache
which eagerly precomputes all suppressions on all annotated elements in
all source files at the point of creation of GenerationState (when the
binding context is still full).
#KT-43047 Fixed
- Do not wrap exceptions resulting from generating error expressions
multiple times, as that could lead to stack traces where the identical
code is wrapped many times and is printed in the exception message on
each step, which was difficult to read
- Add element location (file name, line number & position) to the
message, similarly to exceptions from codegen, when catching and
rethrowing exceptions at the top level
The inline class is boxed when we pass it as lambda argument, now we
unbox it. If the underlying type is not Any or Any?, bridge method does
the unboxing.
#KT-32450 Fixed
#KT-39923 Fixed
#KT-32228 Fixed
#KT-40282 Fixed
This helps to reduce peak memory in lowerings/codegen by a lot.
A more robust approach would be to have a separate BindingContext for
each file, and clear each of them after running psi2ir on it. This would
also lower peak memory usage in psi2ir.
Provide a fallback workaround compiler argument
-Xir-do-not-clear-binding-context just in case BindingContext is in fact
used somewhere and it's not caught by tests.
instead of statically registering it.
Static registering can cause subtle errors when plugin implementation
(e.g. SerializationDescriptorPluginForKotlinxSerialization) is registered
from multiple classloaders: in multi-module with daemon compilation scenario
#KT-41857 Fixed
All inline classes should be boxed coming in and out of lambdas,
however, if the inline class was returned non-locally, it was not boxed.
This change fixes the issue in Old JVM BE.
#KT-41194 In progress
The JVM BE inlines calls to the underlying function in a $default stub
verbatim, e.g., without renaming LVT entries or regenerating anonymous
objects. This commit introduces the same behavior in the JVM IR BE.
Fixes KT-36769.
When a suspend lambda does not capture crossinline lambda, it is
generated with as state-machine, since it does not inline anything.
However, when regenerating, the inliner used to remove all DebugMetadata
annotations to avoid duplication. This lead to missing annotation if
the lambda is regenerated, but state-machine is not regenerated.
This change fixes the blind spot by readding the annotation after
regeneration.
#KT-41789 Fixed
Steven Schäfer