Unexpected error when testing project with jdk11

[Tomahna]

Hello,

First thank for your efforts, bloop is definitely an awesome tools.

I'm trying to get bloop working with java 11, compilation works fine but on some projects tests fail to run.

System:
nixos on unstable channel
java openjdk 11.0.3-internal 2019-04-16
bloop 1.3.2+2-ac2e07ff

Reproducing it:

> sbt new scala/scala-seed.g8
> cd [project-name]
> sbt bloopInstall
> bloop test 
Compiling root (1 Scala source)
Compiled root (174ms)
Compiling root-test (1 Scala source)
Compiled root-test (378ms)
[E] Failed to initialize communication: jdk/internal/reflect/ConstructorAccessorImpl
[E] Internal error when running tests: sbt.ForkMain$Run$RunAborted: java.net.SocketException: Broken pipe (Write failed)
===============================================
Total duration: 0ms
No test suites were run.
===============================================

I experience this only on java 11, running it on java 8 works fine. Bloop is running as a systemd service. Note that this does not seem to occur on all projects. For example, on cats, bloop test coreJVM runs fine.

[jvican]

Hey @Tomahna, thanks for reporting! Yeah, unfortunately this bug fell through the crack because we're not testing Java 11 on our Nailgun infrastructure. I'll have a look into it.

[rtar]

I had the same issue on JDK11, but I noticed that it always works fine if I launch bloop server from terminal, and does not work and fails with the error above if I use Xfce launcher.

I added env command to see what is the difference, and the only thing which was really different is PATH environment variable. I copy-pasted the one I had in terminal into bloop launcher script and it now works perfectly.

I do not know which entry exactly from PATH was to blame though.

[jvican]

That's great news! I'm honestly not sure why this is happening, whenever we add JDK11 to our testing infrastructure we'll see if this is an issue or not. If this isn't, I'll be closing it and instead tell people to look better into their Java installations.

Recommendations on how to improve the workflow of inspecting and setting JAVA_HOME while debugging this is welcome. Note that bloop about prints out the Java home used by the bloop server. If someone wants to brainstorm ideas, please do so either here or in Gitter.

[Tomahna]

Note that I tested both on my workstation (nixos) and the one of a colleague (manjaro) and we both had the same issue even when launched from the command line with bloop server.

[rtar]

It seems my path included JDK8, that why it worked. When JDK8 removed, I get the same error consistently.

[jvican]

Fixing this is a priority for the next release 1.3.3. Will keep you all posted, thanks for the comments and detailed explanations.

[jvican]

Upstream nailgun ticket about supporting JDK 11, looks like we will need to make changes in our fork to support JDK 11. facebookarchive/nailgun#172

[jvican]

After a little bit more investigation, I can confirm this doesn't look like an issue caused by nailgun but by sbt's test interface. My guess is that if we've got this far with JDK 11 is that nailgun is indeed compatible with JDK 11. I will confirm this test by running tests on JDK 11 in our CI.

[Tomahna]

I can confirm that it works well with bloop v1.3.2+125-2d6bf327. Thanks @jvican for your hard work.

[mushtaq]

@Tomahna curious how can I try v1.3.2+125-2d6bf327. I can use resolvers += Resolver.bintrayRepo("scalacenter", "releases") in plugins.sbt, but from where to download and install the bloop binary for CLI?

[Tomahna]

@mushtaq it depends on how you install bloop. If you install bloop on linux using the curl method. Instead of piping the install script to python, you can download it and run it with a version.
python install.py -v 1.3.2+125-2d6bf327 or something along these line.

[mushtaq]

@Tomahna thanks!