Fix type of nested inner-class field access #532
Conversation
| @@ -98,6 +102,54 @@ class KotlinTypeMapping( | |||
| return type(type, parent, signature) | |||
| } | |||
|
|
|||
| fun type(psi: PsiElement?, type: Any?, parent: Any?): JavaType? { | |||
There was a problem hiding this comment.
The new type method seems to blend the functionality of the PsiElementAssociations, KotlinSignatureBuilder, and KotlinTypeMapping.
What do you think about detecting the case in the PsiElementAssocation which is meant to associate the PSI to the FIR? And have a new method in the signature builder that is capable of returning the expected signature. Then the TypeMapping may use the corrected signature to return the correct type?
Note:
If I understand the changes correctly, it will fail for val x = foo.bar.A.B.A.C(). Please test that as well.
There was a problem hiding this comment.
I introduced this method to be able to break down a leaf FIR node (like A.B.A in this case) into pieces (like breaking down the atomic nucleus into quarks), this requires PSI's help, so that's why it has a psi as a parameter.
I didn't find another way to break down a leaf FIR node like A.B.A to A, B, and A.
There was a problem hiding this comment.
I mean to apply the same concept, but in a different way that encapsulates the functionality in the applicable classes. I'll take a look tomorrow
There was a problem hiding this comment.
I took a look at the changes, and there are still issues. The changes cause parameterization to be lost since the signature created by the signature build is based on flattening a dot-qualified expression rather than using the FIR type attribution.
I tested the original suggestion about using the ClassId and FirSession to associate the PSI to the FIR and confirmed it's possible. So, we can detect the mismatch in the PsiElementAssociation and find the right FIR element, then call typeMapping#type(...) on the associated FIR.
To recap ClassId and FirSession:
A FirResolvedQualifier will have a symbol if it is resolved to a type.
The symbol will have access to a ClassId, which enables the detection of outer classes.
Then, ClassId can be used to look up the associated FIR element, which we can use for type attribution.
I think we can close this PR since it still isn't quite right and it'll be safer to use the FirSession instead.
fixed by the 2nd commit, and added the test case as well. |
|
An alternate PR : https://github.com/openrewrite/rewrite-kotlin/pull/547/files |
fixes #517