partest' jvm/natives.scala missbehave on non-intel arches

[judovana]

On ppc64le or aarch64:
...
!! 31 - jvm/natives.scala
...
Other tests are ok.

reproduction steps

// ensure jdk8 is your jdk, with jdk11 the tesr may pass
wget https://piccolo.link/sbt-1.3.4.tgz
wget https://github.com/scala/scala/archive/v2.12.12.tar.gz
tar -xf sbt-1.3.4.tgz
tar -xf v2.12.12.tar.gz
pushd scala-2.12.12
bash ../sbt/bin/sbt partest
popd

problem

When partest is run by jdk8 on non intels, it download intel binaries.
I run scala on ppc64le and aarch64.
I'm not sure why the issue is not observable when run by jdk11, nor was lookng on what sbt is doing here.

Thanx for any hint!

[eed3si9n]

I'm not sure if this is much of a hint, but scala/test/files/jvm/natives.scala tests Java Native - https://github.com/scala/scala/blob/v2.12.12/test/files/jvm/natives.scala

  val libName =
    if (sysType == "Mac OS X")
      "natives"
    else
      "natives-" + sysWordSize

  System.loadLibrary(libName)

That means your build needs to build libnatives-64.so - https://github.com/scala/scala/blob/v2.12.12/test/files/jvm/mkLibNatives.sh. Make sure that part works.

[judovana]

Ok, that part clearly do not work.
There is clearly downlaoded file:
.../scala-partest/scala-2.12.12/test/files/jvm/libnatives-64.so: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, not stripped

Thanx for pointer. Will lookm ore to that mkLibNatives.sh

[judovana]

Seeing the cygwin/darwin resolution and seeing the CC_OPTIONS="-c -arch ppc -arch i386 -arch x86_64" makes me a bit worried...
I will try to re-dust my rotting.. no rusting:) ... C knowledge.

[eed3si9n]

Yes. There's libnatives-64.so pre-built and checked in to scala/scala, but it's x86-64. So I guess the test currently assumes you're on Intel. I have no idea why it would work on JDK 11 for you.

[judovana]

I see. Well using intel prebuild binary on non intell really sounds like a bug.
I also cinsider passing on jdk11 as bug. Very likely the jdk11 is failing to open the library, but the error is somehow surpassed.

[lrytz]

I really think this test is not worth the trouble it causes. On macOS you get

So I think we should remove it, and possibly replace it by a bytecode test in https://github.com/scala/scala/blob/2.13.x/test/junit/scala/tools/nsc/backend/jvm/BytecodeTest.scala

[judovana]

Oh gosh:)

[judovana]

I belive such a test should be around, although it is more Java test then scala test.
However this test should be skipped if there is no gcc/msvc.
And if there is, the native hsould be compiled. I dont know, but hope, that self compiled library would be ok on Mac.

[judovana]

Can I somehow run only the jvm/natives.scala or at least only jvm group?

[lrytz]

$ sbt
> partest test/files/jvm/natives.scala

But again, I really think asserting the bytecode structure would be the better way forward for testing this feature. Running native code is going to cause maintenance burden again and again, and I wouldn't like the team to spend time on this.

[judovana]

Eh.. that do not work... Not even files/jvm/natives.scala or jvm/natives.scala or natives.scala. Those I tried first. I addition the partest --help, as documented on readme.md is not working (or I'm misinterpreting it)

Any other dealing with this issue is up to scala team. I can fix the partest natives so it passes on nonintels (where running it alone would help) but I'm unable to write test asserting bytecode for you, You have gave me most of what I needed, and provided great support. Thank you. Crossing fingers to move on whatever you feel best for sacala. The issue can be closed now as wont fix, or wait until the test is removed/replaced/fixed.

[lrytz]

partest --help, as documented on readme.md is not working (or I'm misinterpreting it)

That's weird.. did you make sure to run it within the sbt console? If you pass it to sbt you'd need to add quotes

$ sbt "partest test/files/jvm/natives.scala"

otherwise sbt takes it as two commands.

I'll try to do the bytecode test soon. Having natives.scala fail is confusing contributors too many times.

[judovana]

Oh, the quotes did the job. Sorry for confusion. Readme.md is really not clear about it and sbt is not exactly wide-known tool.
Also the "partest --help" works, only it is longest wait for help ever:)
Tyvm, good luck with bytecode check.