Generate trees for Kotlin using the `kotlin-compiler-embeddable` parser

[tsantalis]

The tree-sitter parser for Kotlin is quite unreliable.

In many cases, it generates ERROR nodes (more details in the issue below)
bonede/tree-sitter-ng#108
and it also parses functional interfaces as functions.

I would strongly recommend to use kotlin-compiler-embeddable
https://mvnrepository.com/artifact/org.jetbrains.kotlin/kotlin-compiler-embeddable/versions?repo=central
for generating the AST.

It provides a org.jetbrains.kotlin.psi.KtVisitor

Parsing is easy as follows:

String filePath = ""; // the path of a kt file
String fileContent = "" // the string content of a kt file
KotlinCoreEnvironment environment = KotlinCoreEnvironment.createForProduction(
  	Disposer.newDisposable(),
  	new CompilerConfiguration(),
  	EnvironmentConfigFiles.JVM_CONFIG_FILES
);
PsiFileFactoryImpl factory = (PsiFileFactoryImpl) PsiFileFactory.getInstance(environment.getProject());
PsiFile psiFile = factory.createFileFromText(filePath, KotlinLanguage.INSTANCE, fileContent);
KtFile ktFile = (KtFile)psiFile;

[jrfaller]

Hi @tsantalis, good find! It seems that the only required code to write is a convertor from a PSI tree to a GumTree tree it should be too hard.

[tsantalis]

@jrfaller Thank you for considering this feature.
The doc for KTVisitor

https://fossies.org/dox/kotlin-2.3.20/classorg_1_1jetbrains_1_1kotlin_1_1psi_1_1KtVisitor.html

visitKtElement(@NotNull KtElement element, D data)
is perhaps the method to override, as KtElement is the super-interface for all Kotlin AST nodes.

[jrfaller]

Nice! I think you spotted the best method to override, thanks!

[tsantalis]

kotlin-treesitter does not model finally block in try-expression
@jrfaller
This is a very bad parser. Very bad AST diff. If you don't have time, I will ask a student to work on this issue.

[jrfaller]

Sorry @tsantalis, I do not have much time currently to work on the topic. I have investigated a little bit but it turns out that it's maybe not straightforward as I have the impression that IntelliJ work on an abstract PSI API that is then used by the parsers. Therefore I am not sure whether it would be better to plug a tree generator directly on PSI tree that would enable handling all languages for which intellij has a parser.

[tsantalis]

@jrfaller
Each language has its own PSI implementation built to handle its unique syntax and semantics. For Kotlin, code is represented using Kt classes (e.g., KtClass, KtFunction), while Java relies on standard PsiClass, PsiMethod, and PsiField implementations.

The JDT parser that you use for Java is really great. The problem is with Kotlin.
treesitter-kotlin is very unreliable.
I will ask some student to work on it.