javakit: improve error messages when class missing from classpath

[ktoso]

Today it seems we just:

Error: org/gradle/tooling/GradleConnector

which you can guess what it means but it'll be better if we spell out "Class not found" etc

[lhoward]

Also seeing the same thing. Can I assume this means class not found? Do I need to explicitly add the current directory to classPath when mapping a class that's defined within a Swift target? Can a single Swift module wrap both classes from an external JAR and those that are located within it?

Edit: it seems I need to include the build directory in the class path:

  "classPath": "/opt/flutter/bin/cache/artifacts/engine/android-arm64/flutter.jar:.build/plugins/outputs/flutterswift/AndroidFlutter/destination/JavaCompilerPlugin/Java",

This doesn't seem right. Particularly as I'm not sure I can guarantee the order of plugin invocation.

[ktoso]

You'll have to build a "proper" classpath that explicitly lists all jars and paths we need to include etc (plain old java classpath), that's an adventure in of itself but I'm sure we can figure it out.

Most of the time such message is a class not found, but can't guarantee it's just that, that's what this issue is about, making those errors better

[lhoward]

OK, so maybe I need to fix this (GH) issue to fix my actual issue :)

Should Swift packages wrapping external JARs be in distinct packages from those that have Java classes (with native methods) inside them?

[lhoward]

Can you give me a hint as to where this error message is emitted from? git grep offers no hints.

[ktoso]

Should Swift packages wrapping external JARs be in distinct packages from those that have Java classes (with native methods) inside them?

I don't think we have a complete opinion formed here yet. Can you explain in more detail the question. There's various "wrappings" that make sense, for example "a jar that contains java code including native dylib that it calls into (either using JNI or FMM (jextract-swift)" is definitely a thing, mixed language projects eventually too... but this isn't figured out yet. What specifically are you asking about here?

[ktoso]

Can you give me a hint as to where this error message is emitted from? git grep offers no hints.

That's part of this issue I'm afraid 😅 It might be during a FindClass call perhaps? Idk if the Java runtime does the print here or who does, we don't seem to have such Error: ... print at least.

You could also launch your jvm with lldb -- java -cp ... <main> and this way you could maybe find where this happens by stepping through a sample app if nothing else is helping.

[lhoward]

Should Swift packages wrapping external JARs be in distinct packages from those that have Java classes (with native methods) inside them?

I don't think we have a complete opinion formed here yet. Can you explain in more detail the question. There's various "wrappings" that make sense, for example "a jar that contains java code including native dylib that it calls into (either using JNI or FMM (jextract-swift)" is definitely a thing, mixed language projects eventually too... but this isn't figured out yet. What specifically are you asking about here?

Ultimately I'm building a shared object to be embedded in an Android APK that will implement some native methods. I'm not asking about the distribution method as much as whether the Swift projections for imported Java API should be in a separate Swift target (module) to Java classes whose source is included in the target. The reason I ask is that it appears that if I don't do this, Java2Swift.config's classPath appears to need to explicitly refer to the JavaCompilerPlugin's output directory, which seems wrong.

[lhoward]

Also, does the classPath in the configuration file augment, or replace the CLASSPATH environment variable?

[ktoso]

Also, does the classPath in the configuration file augment, or replace the CLASSPATH environment variable?

AFAICS we don't take into account env variable at all; it's a good question if it should. Probably not, we just need to do a better job at forming classpaths -- right now it's incredibly basic. I have ideas here to form classpaths that include dependencies: #137

(I really need to rename it to classpath, the uppercase in the Swift word is very weird 😅)

[ktoso]

The reason I ask is that it appears that if I don't do this, Java2Swift.config's classPath appears to need to explicitly refer to the JavaCompilerPlugin's output directory, which seems wrong.

I wholly expect all of this to not be the end state we want to arrive at. Let's start with "what should work" and then work backwards to fix how we form these configurations.

[lhoward]

We should also validate a malformed class path if we can't get a more useful error from the JVM than this:

Error occurred during initialization of VM
java.lang.InternalError: Bad level: 2
	at jdk.internal.misc.VM.initLevel(java.base@21.0.3/Unknown Source)
	at java.lang.System.initPhase2(java.base@21.0.3/Unknown Source)

In this case it had a trailing white space.