Hello, I am writing a compiler plugin to modify errors and warnings found by the compiler. I have a code like this, that can read the diagnostics

class WarnningsExtension(
) : AnalysisHandlerExtension {

    override fun analysisCompleted(project: Project, module: ModuleDescriptor, bindingTrace:
    BindingTrace, files: Collection<KtFile>): AnalysisResult? {

        val diagnostics = bindingTrace.bindingContext.diagnostics
        return AnalysisResult.Companion.success(bindingContext, module)
    } 
}

But, I also want to modify this list like

val newDiagnostics = diagnostics.filter { //myFilter }

And return this result back instead. Is something like this possible? Or is there any extension that allow me to modify diagnostics?

Hey I've created https://github.com/hadilq/trackable-compiler-plugin and all the tests in the sample project and android sample tests are passing. You can simply check it by running

./gradlew clean check

, but when I tried it on a simple new android project created by AS, the same tests throw

NoSuchMethodException: com.github.hadilq.emptycomposeandroid.TrackableClass.track()

I have to add that it seems the compiler plugin is installed because it compiles without

Unresolved ...

, but at the runtime there is no generated method! Any idea would be appreciated. Thank you for your time 🙂

Hi, How can I run the tests for the android extensions compiler plugin (in

plugins/android-extensions-compiler

)? I have a patch ready to fix a major bug but can't find a relevant gradle task or IDEA configuration to run the tests. If I can update these tests I hope to raise a pull request within the week

Are there ways to modify existing code? Specifically, I’m trying to remove an annotation from a class or to modify it:

// Original:
@MyAnnotation(values = ["a", "b"]
class MyClass

// Expected:
@MyAnnotation(values = ["a", "b", "c"]
class MyClass

I couldn’t find a hook to do this. I only found ways add a new annotation. Any pointers?

I've recently enabled the new inference algorithm with the Kotlin 1.4 nightly (1.4.0-dev-6071) compiler. I have an inline function that calls some Java code with signature:

public <T> T deserialize(..., final Type type)

like this:

inline fun <reified T: Set<Any?>> deser(...): T? = deserialize(..., object : TypeToken<T>() {}.type)

and this used to work fine with 1.4-M1 / old inference. However, now I get a runtime error when the java code attempts to return its value to the Kotlin code:

Caused by: java.lang.ClassCastException: class java.util.LinkedHashSet cannot be cast to class java.lang.Void (java.util.LinkedHashSet and java.lang.Void are in module java.base of loader 'bootstrap')

I can "solve" this by explicitly adding the

T

type to the deserialize call i.e.

deserialize<T>(...)

, but I'm wondering if this is a bug with the new type inference? The compiler does not report any errors if the

<T>

is left off.

Poll: what should (note: not, what does) this code produce? (Hint: it produces 1️⃣ in Kotlin 1.4-dev with new inference enabled)

fun <T> foo(): T? = "abc" as T

fun main() {
  val s: String? = if(false) null else foo()
}

1️⃣ A runtime exception

ClassCastException: class java.lang.String cannot be cast to java.lang.Void

2️⃣ A compile-time error due to the type

T

of

foo

being ambiguous (could be

String?

or could be

Nothing

) 3️⃣ A compile-time warning that the type of

T

was inferred to

Nothing

and that is probably not the desired result 4️⃣ An inference of

T

to be

String?

, so no compile-time or runtime errors 5️⃣ Other

1. I notice that when I declare a

private companion object

, the

Companion

class is compiled as package-private in the bytecode, but the static

Companion

field is compiled as public + deprecated in the bytecode. Why is this? 2. Because of this, in the IDE, from a Java call-site, I can can compile and run

Object object = DeferredBoolean.Attribute.Companion;

without issue. The

Companion

field doesn’t even get a strikethrough, which seems like an IDE bug. Assuming this hasn’t already been fixed in a new IDE version, and depending on the answer to #1, should I file an IDE bug?

Is there a way to make the Kotlin compiler list its phases? In Scala, for example, I can run

scalac -Xshow-phases

to get something like this:

phase name  id  description
    ----------  --  -----------
        parser   1  parse source into ASTs, perform simple desugaring
         namer   2  resolve names, attach symbols to named trees
packageobjects   3  load package objects
         typer   4  the meat and potatoes: type the trees

I may have hit a compiler bug. If I use the HTML DSL in my project with tbody { tr{ td{ }}} , the more tr I add in my code , the slower it compiles. Having one it compiles in ~30s, adding another one it takes 1min30, and I stopped it after 4 min when I have three… (edited)

Am I missing something obvious? I feel like compiler plugins can easily break with incremental compilation. My plugin generates a new Kotlin file when another file contains an annotation. I add the plugin with

dependencies {
  kotlinCompilerPluginClasspath project(':myplugin')
}

If I change something in my compiler plugin, then I see that the Kotlin compilation tasks aren’t up to date, because my compiler plugin changed. Good! The problem now is that the incremental compilation kicks in and doesn’t recompile the code in the module, because nothing has changed for the source files individually, only the Kotlin compiler plugin dependency has changed. Is there a way to recompile a module, if one of its compiler plugins has changed? (I tested disabling incremental compilation and this fixes the problem. But I’m afraid of the slow compilation long term.)

Where can I find kotlin idea plugin dependency like

org.jetbrains.kotlin.idea.highlighter.KotlinHighlightingColors

? I could find it in the kotlin source code, but I don't know how to add the dependency.

Is there a way to remove/replace existing methods with a compiler plugin?

Is there a way to analyse Kotlin files with the new compiler frontend using the

kotlin-compiler

artifact ?

I'm writing a compiler plugin with Arrow Meta. I manage to print some debugging output with

messageCollector?.report(CompilerMessageSeverity.WARNING, message, HERE)

but using any lower severity level does not produce any output on the console, even if I pass

-verbose

as compiler argument. Why? How can I use different severity levels or otherwise print messages to the console from my compiler plugin?

Hello! How can I get IrClass of other module from symbol? when I call symbol.owner I got error: java.lang.IllegalStateException: Symbol for ... is unbound

Is it possible to report errors from

IrGenerationExtension

?

Hello! In ir transformation phase i need to add private default setter to property if it is not exists (is val). For that case i can assign IrProperty.setter = to needed implementation. But after transformation java still generate final field and throw 'Update to non-static final field ... attempted from a different method (...) than the initializer method <init>' How can i make this work? I check IrProperty class and find out that it has isVar flag that is not correlate with availability of setter method. It is just final field initialized in constructor.

I'm using kotlin 1.4-M1 with IR enabled and found out bug in generated code. I want to know is it known bug or I should repot it?

abstract class TestBase<T> {
        suspend fun test(value: T) {
            println(doTest(value))
        }

        protected abstract suspend fun doTest(value: T): T
    }

    class TestInherit : TestBase<String>() {
        override suspend fun doTest(value: String): String {
            return value + "/" + value
        }
    }

    @Test
    fun test() {
        runBlocking {
            TestInherit().test("aaa") /* throws exception here java.lang.AbstractMethodError: Receiver class TestInherit does not define or inherit an implementation of the resolved method 'abstract java.lang.Object doTest(java.lang.Object, kotlin.coroutines.Continuation)' of abstract class TestBase. */
        }
    }

If define TestBase without generic it works ok.

abstract class TestBase {
        suspend fun test(value: String) {
            println(doTest(value))
        }

        protected abstract suspend fun doTest(value: String): String
    }

    class TestInherit : TestBase() {
        override suspend fun doTest(value: String): String {
            return value + "/" + value
        }
    }

    @Test
    fun test() {
        runBlocking {
            TestInherit().test("aaa") // OK
        }
    }

Is it possible to restart the compiler from inside a compiler plugin or just restart the plugin ?

When IrField.isStatic == true initializer block is not generated when compiling to JS. Only var fieldName; is generated

Is it possible to add synthetic class to package level in SyntheticResolveExtension?

I observe an interesting type inference behavior regarding

it

parameter in lambdas contained in pairs, lists or other data structures. This code sample

var pair: Pair<Int, (String) -> String> = 2 to { "param $it" }
var pair2: Pair<Int, (String) -> String> = 2 to { p -> "param $p" }

var list: List<(String) -> String> = listOf { "param $it" }
var list2: List<(String) -> String> = listOf { p -> "param $p" }

looks fine to IDE, but fails to compile with the following errors for the lines using

it

lambda parameter:

Type inference failed. Expected type mismatch: inferred type is Pair<Int, () -> String> but Pair<Int, (String) -> String> was expected
Unresolved reference: it
Type inference failed. Expected type mismatch: inferred type is List<() -> String> but List<(String) -> String> was expected
Unresolved reference: it

But the compiler is happy with the root-level lambda declarations like this

var func: (String) -> String = { "param $it" }
var func2: (String) -> String = { p -> "param $p" }

Is this a bug or an intended behavior? Are there any workarounds that would make the compiler happy and at the same time keep the code neat (that is, using

it

but without full lambda type annotations at usage site)? I'm using Kotlin 1.3.71 in multiplatform project and can see this both in JVM and JS targets.

Hello! Is there any way to itarate through all class descriptors in module descriptor?

I'm using an IR rewrite pass to replace a property with a new one with the same name, but a new backing field, and accessors that get/set a property of that backing field. This works fine for the most part, but in

init

blocks and class methods, references to the property's accessors seem to be replaced with the backing field. Ex:

class Test(@Watch var x: Int){
    fun print(){
        println("Prop: $x")
    }
}
fun main() {
    val t = Test(3)
    
    println(t.x)
    t.print()
}

The

println(t.x)

works fine and prints 3, but the

println

in

print

prints the backing field. I'm using 1.4-M1. Ideas? I've dumped the IR and there doesn't seem to be any differences

ModuleDescriptor.getSubPackagesOf doesnt return subpackages if there is no .kt files in searched package. Is it a bug or i miss smth. Please advice me.

I'm trying to add a synthetic property to classes w/ my annotation. It works as expected, except it turns any property access statements on that class into get_fields on the added property's backing field. It does this before it even gets to the IR phase, so I'm not sure what's causing it. The SyntheticResolveExtension is just:

class KlairvoyantResolveExtension() : SyntheticResolveExtension {
    override fun generateSyntheticProperties(thisDescriptor: ClassDescriptor, name: Name, bindingContext: BindingContext, fromSupertypes: ArrayList<PropertyDescriptor>, result: MutableSet<PropertyDescriptor>) {
        if (!thisDescriptor.hasWatchAnnotation())
            return

        if(result.any { it.name == Name.identifier(Names.ClassWatcherProp) })
            return

        with(FrontendReferencer(thisDescriptor.module)) {
            val classWatcherDescriptor = SimpleSyntheticPropertyDescriptor(thisDescriptor, Names.ClassWatcherProp, KotlinTypeFactory.simpleNotNullType(Annotations.EMPTY, CompositeWatcher, listOf()), visibility = Visibilities.PUBLIC)
            result.add(classWatcherDescriptor)
        }
    }
}

I'm writing an IDE plugin for my synthetic property, and I'm including my compiler plugin as a dependency and reusing the SyntheticResolveExtension class. However when I run the IDE, it complains that my extension doesn't implement SyntheticResolveExtension. I can use reflection to get the class, and while it has a superinterface of SyntheticResolveExtension, it is not equal to the SyntheticResolveExtension it is checking for. I suspect this comes from mine coming from "org.jetbrains.kotlin:kotlin-compiler-embeddable:1.4-M1", and the IDEs isn't (I can't seem to get the

runIde

task to use the 1.4-M1 Kotlin plugin, which may be part of the reason why). Is there a way to make this work without duplicating a bunch of code in the ide plugin?

With my IDE plugin, my field resolves fine once it is typed, but doesn't appear in autocomplete. Is there an extension point I'm missing? I'm using

generateSyntheticProperties

in my

SyntheticResolveExtension

I manage to get the compiler to spew this.

e: org.jetbrains.kotlin.codegen.CompilationException: Back-end (JVM) Internal error: Couldn't inline method call 'usingConnection' into

I can reproduce it in my project but I'm having trouble making a self-contained reproducer. I would publish the project but the contents are personal/sensitive (My access tokens are hard coded). What to do? I fix the error when I stop taking advantage of contracts in suspend methods.

// Breaks compiler
val value: String
withContext(...) {
    value = loadFromDb()
}
println(value)

// Doesn't break
val value: String = withContext(...) {
    loadFromDb()
}
println(value)