0.13 is coming soon

[konsoletyper]

I finished all the features that I wanted to complete for release 0.13. I'd like to receive feedback before I perform actual release. Please, try to switch to 0.13.0-dev-10 in your project and let me know if you have some problems compared to stable 0.12 releases.

[Ihromant]

It's impossible to test prerelease build in IDEA. Now it fails with following error:

I remember that you told that you want to disable dev builds of the plugin. So, I assume that you know what you're doing.

[konsoletyper]

After 0.13.0 using TeaVM with Maven

What do you mean?

this comparison claims that 55% Java devs use Maven and 25% use Gradle

Maven is terrible. Let them switch to Gradle or it's their problem. Also, 99% Java devs don't program in Java for the browser, so the statistics is not correct in our context.

Build time increased from 10 secs in IDEA to 3 mins from console/IDEA

That's strange that build time has raised so much. This should not be the case. If you are experiencing this in your project, then perhaps something is misconfigured on your side.

If I have some questions regarding Maven -> Gradle migration, can I ask them on Discord

Sure

[konsoletyper]

For instance, 99% Android devs use Gradle, but among entire Java community it's a minority.

[Ihromant]

Maven is terrible.

I agree, personally I have no experience with gradle, but I liked ant and switching to Maven was big downgrade for me and until they fixed most annoying parts, it was a pain. But in my working career I spotted gradle only once in neighbor team project.

What do you mean?
That's strange that build time has raised so much.

I mean that combination of IDEA plugin and Maven was completely fine. When IDEA plugin is gone - then you need rebuild all dependencies and then the module with TeaVM entry point. This is what raises build time from 10 sec to 3 mins. Maybe I'm doing something wrong, but consider Maven multi-module user and try to guess how should he use TeaVM for development.

For instance, 99% Android devs use Gradle, but among entire Java community it's a minority.

Google is notorious of forcing developers to use specific tools, but they are also notorious of dropping support of that tools. I don't think that it is good example to follow. When I considered alternative for GWT, I looked at J2CL, but at that time it came with only Bazel build tool and I was not able even run "hello world" with it, so I continue searching and eventually found TeaVM. And GWT is officially dropped by Google after they included lots of Java devs into it. If you want to make TeaVM more famous, you need to invest some time in newcomer-friendly learning curve.

[konsoletyper]

When IDEA plugin is gone - then you need rebuild all dependencies and then the module with TeaVM entry point

Stop, maven does not rebuild dependencies if they did not change. Then you are definitely doing something wrong.

Also, to reduce TeaVM build time, you can turn off any optimizations and set fastGlobalAnalysis. This setting is not documented, but who cares about documentation when there's code?

This is what raises build time from 10 sec to 3 mins

So you are doing general conclusions based on your personal experience?

For instance, recently I helped to develop TeaVM backend for WebFX project, which is based on Maven. And in my case it was about 8 seconds to rebuild. It's far worse than, say, 0.5 seconds, but it's still good enough. If I bothered much, I would have experimented with mvnw and with parallel builds.

[konsoletyper]

Also, don't run Maven fron console, run it as a run configuration. IDEA wraps and accelerates Maven when you using it this way.

[zlataovce]

getting this on dev-10, compiling to wasm gc:

Class org.teavm.platform.Platform is not supported on current target
    at java.lang.ref.ReferenceQueue$RemoveCallback.run(TReferenceQueue.java:93)
    at org.teavm.runtime.EventQueue.run(EventQueue.java:66)

[zlataovce]

tried reassembling wasm-dis output of the module and wasm-as segfaults 😂

I think I'm going to just stay with the JS backend as it's fast enough for my purposes

[konsoletyper]

@zlataovce I asked on WebAssembly discord, they gave an idea and this allowed to figure out what's wrong. I pushed the fix and with it your project passes validation. However, I even did not try to actually run it.

[zlataovce]

@zlataovce I asked on WebAssembly discord, they gave an idea and this allowed to figure out what's wrong. I pushed the fix and with it your project passes validation. However, I even did not try to actually run it.

everything works perfectly now, thank you so much for the time and effort you've invested into debugging the problem!

[konsoletyper]

@zlataovce happy to hear. Just in case you have some performance benchmarks, it would be great to learn if wasm provides any benefit

[zlataovce]

on the aforementioned project (a Java decompiler), I did a tiny benchmark with a JS script with 20 warm-up runs and 150 benchmarked runs of decompilation of a single class (a trivial heap sort implementation), compiled with TeaVM 0.13.0-dev-11, running with Node v24.11.0:

compiled with JS backend: average time per run: 57.58ms, total time for 150 runs: 8637ms
compiled with WASM GC backend: average time per run: 17.7ms, total time for 150 runs: 2655ms

in a less scientific, more real-world benchmark in Chrome, it managed to decompile around 700 classes roughly two times faster compared to the JS-only version
however in Firefox it was at best on par with the JS version and sometimes slower, while consuming three times the memory

also the WASM module is around three times the size of the JS output on disk (obfuscated, the default optimization level for both)

[lax1dude]

I am currently debugging an issue on a private project where replacing IFEQ/IFNE with POP followed by GOTO (to trick the dependency analyzer) causes InstructionCopyReader to fail to copy the jump:

Caused by: java.lang.IndexOutOfBoundsException: Index -1 out of bounds for length 264
	at java.base/jdk.internal.util.Preconditions.outOfBounds(Preconditions.java:64)
	at java.base/jdk.internal.util.Preconditions.outOfBoundsCheckIndex(Preconditions.java:70)
	at java.base/jdk.internal.util.Preconditions.checkIndex(Preconditions.java:266)
	at org.teavm.model.Program.basicBlockAt(Program.java:60)
	at org.teavm.model.util.InstructionCopyReader.copyBlock(InstructionCopyReader.java:110)
	at org.teavm.model.util.InstructionCopyReader.jump(InstructionCopyReader.java:267)
	at org.teavm.model.InstructionReadVisitor.visit(InstructionReadVisitor.java:112)
	at org.teavm.model.instructions.JumpInstruction.acceptVisitor(JumpInstruction.java:34)
	at org.teavm.model.BasicBlock$4.read(BasicBlock.java:299)
	at org.teavm.model.util.ProgramUtils.copyBasicBlock(ProgramUtils.java:131)
	at org.teavm.model.util.ProgramUtils.copy(ProgramUtils.java:116)
	at org.teavm.model.util.ModelUtils.copyMethod(ModelUtils.java:76)
	at org.teavm.model.util.ModelUtils.copyClass(ModelUtils.java:46)
	at org.teavm.model.util.ModelUtils.copyClass(ModelUtils.java:64)
	at org.teavm.model.util.ModelUtils.copyClass(ModelUtils.java:60)
	at org.teavm.vm.TeaVM.link(TeaVM.java:618)
	at org.teavm.vm.TeaVM.eagerPipeline(TeaVM.java:477)
	at org.teavm.vm.TeaVM.build(TeaVM.java:440)
	at org.teavm.tooling.TeaVMTool.generate(TeaVMTool.java:540)

I have not confirmed yet if this is a problem with TeaVM or just my bytecode transformers messing something up, I'll keep you updated if it turns out that there is some problem with TeaVM related to GOTO instructions, since I don't believe I've ever seen that instruction used before in a regular compiled java source file so its usage may not be covered in the regular test cases.

[lax1dude]

Unfortunately I can't get a standalone reproducer together, I've been trying different patterns like 2 hours and nothing is crashing it. The ATHROW instructions being inserted by ASM are total nonsense, there's nothing on the stack to for these instructions to even throw, and would probably cause class validation errors if the JVM wasn't aware of the fact that the instructions are unreachable. I think I might just work around this by registering something like the @PlatformMarker transformer into TeaVM itself.

[lax1dude]

still crashes when I hack ASM to not emit nonsense ATHROW instructions

[konsoletyper]

Then give me your project and instructions how to build. It's strange, because I believe I did not touch the code that mentioned in the stack trace.

[lax1dude]

I'm going to keep trying to find the root cause tomorrow, I think there is a real bug here if it still crashes even without any dangling ATHROW instructions in the unreachable code. There's no reason I'd expect TeaVM to crash just because it had to eliminate a basic block that had nothing in it but some NOP instructions.

[lax1dude]

Then give me your project and instructions how to build. It's strange, because I believe I did not touch the code that mentioned in the stack trace.

This is probably not a regression. The code is a private project with like 30,000 classes and is very much still a work in progress, I would have to add you to the repository if you want to reproduce it yourself. My discord is lax1dude if you could send me a friend request. Its getting very late so I'll have to get back to you tomorrow.

[pizzadox9999]

[konsoletyper]

Where did you take this information about dropping Maven support?

[pizzadox9999]

From the conversation at the start, if I misunderstood it I am sorry.

[SquidDev]

Have been using the 0.13 betas for a while now. Just updated to the latest commit, and it appears to be working perfectly! Thank you for all your work!

[kenchis]

Thank you for the update!
I am a bit late, but JSIDPlay2 in the gallery is now up-to-date (dev-10). No build problems or such.
I will test further all versions.
First impression is: it is stable.

[kenchis]

dev-10 seems to be working great. No problems found, yet. Its running non-stop for the last hours.

[kenchis]

I am currently working on the SharedArrayBuffer version and it start working for me (pixels and sounds already using it). Just need proper synchronization until i can publish it.
Thank you for providing dev-10 !