Smartcasts for public properties from different module are not
stable because module declaring a property in general can be
compiled separately from the module using it. However, if client
module has dependsOn relation with declaring module their simultaneous
compilation is guaranteed which makes this smart cast safe.
Cache all transitive 'expected by' modules in module dependencies.
Extend test to check smart casts are allowed for properties from transitive
'expected by' dependencies and prohibited otherwise.
^KT-42754 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.
Compute FQ name of the main class right after running the analysis and
before invoking codegen. This is needed because MainFunctionDetector
depends on BindingContext, and JVM IR needs to clear BindingContext as
soon as it's not necessary to reduce peak memory usage, thus breaking
any usages of data from it after the codegen.
Also refactor and use the extracted, but not properly reused previously,
copy of findMainClass in findMainClass.kt.
Note that this replaces NPE in KT-42868 with an UOE.
The current backend does that for some bridges. We do it consistently
for all bridges.
The line number used differs. For the JVM_IR backend, we use the
line number of the class to which the bridge is added. For the
JVM backend, that does not appear to be the case for bridges
in lambdas. I prefer the line number for a lambda invoke bridge
to be the line for the lambda instead os some surrounding class.
Reimplement the same hacky approach used in the old backend (see
cc2fe6b0c6).
Previously, the debugger incorrectly stepped into Collections.kt on
"step over" inline function calls from stdlib like 'any'.
Since `if` and `when` expressions are represented the same way in IR,
the behavior is fixed for both of them. It's not the case in the old JVM
backend, where stepping over `when` conditions still suffers from the
same problem, which the newly added test checks.
Before this change, a test with incorrect .ir.out could pass if the test
execution result was equivalent to the main .out file, which was a bit
confusing.
Loads of temporary variables that contain constants are replaced
with a copy of the constant. This avoids locals loads and stores.
However, the copy of the constant needs to have the offset of
the load and not of the original constant.
Fixes KT-41963.
This fixes smart step into for delegated member functions.
Additionally, we align on the string "memberFunctionName(...)"
for expression non-null checks for both JVM_IR and JVM
backends.
Bytecode generated by JVM IR ensures that on multiline expressions,
every call is stepped over, which seems better than in the old backend.
Also remove kt34905.kt since it duplicates this test case.
The only difference seems to be that step into IImpl.fooOverride is
actually supported in JVM_IR. Also fix slightly incorrect line numbers
in the comments.
The difference is that JVM_IR generates line numbers for the
constructor field initialization.
In this case:
```
29: class B {
30: // comment
31: val bPropVal: Int,
32: // comment
33: var bProbVar: Int
34: ) {
```
the line numbers for the field initializations are 31 and 33 where
in the current backend the whole constructor has line number 29.
Dmitriy Novozhilov