highline 1.6.9 hangs indefinitely with JRuby

[graaff]

With jruby 1.5.6 (ruby 1.8.7 patchlevel 249) (2012-01-08 6586) (OpenJDK 64-Bit Server VM 1.6.0_22) [amd64-java] tests hang indefinitely. I've tested both with and without your patch for my ruby 1.8 bug. It shows two dots when running its tests and then hangs. I realize that our jruby is old but we can't currently upgrade since 1.6 versions don't run properly on our system.

[JEG2]

In understand. However, this is an old JRuby and we're just talking about the tests, right?

Does the library still work or is it also broken?

[graaff]

Not sure since I only tried the tests, and I don't have a clear test case for the code. I'm doing this testing so that I can add the new version to Gentoo Linux. I can understand if you want to put this on hold until we upgrade to a new jruby version.

[JEG2]

Yeah, I actually think that may be older than what we now support. We use JLine under the hood and I can't seem to find when that was added to JRuby, but I believe it's recent-ish.

We couldn't really find another working option for JRuby support, so I think we need it unless someone offers up another solution.

[tomdz]

jline seems to have been in JRuby since about 1.6.0 or 1.6.1 (april 2011). Would it be more better if highline has its own jline bundled ? I.e. a highline-java version that includes jline (similar to jruby) ?

[JEG2]

That might be an option. Would JLine run on the older JRuby versions?

[tomdz]

I don't think jline integration needs a lot of jruby features, so it should work with, say, 1.5.6. Bigger problem is the bundling - the jline jar itself is somewhat big, so a highline-java or even highline-jruby-old variant (since with 1.6.x the separate jar is not needed anymore) might be a good idea ?

[JEG2]

Yeah, this is sounding like some hassle to support some pretty old stuff. I'm feeling like it's not worth the bother. Do you guys disagree?

[graaff]

It looks like our jruby 1.5.6 already has jline support. Also, with that version all tests pass for highline 1.6.8 pass. So unless something changed accordingly in highline 1.6.9 it may still be another problem. In any case I'm fine with postponing this until we can test with jruby 1.6.

[JEG2]

OK, closing this for now.

[Flameeyes]

Okay I think I found the problem. The test contains:

@input    = StringIO.new
@output   = StringIO.new
@terminal = HighLine.new(@input, @output)  

Unfortunately, in the case of JRUBY, the two parameters are ignored:

  @java_input = Channels.newInputStream($stdin.to_channel)
  @java_output = OutputStreamWriter.new(Channels.newOutputStream($stdout.to_channel))
  @java_terminal = Terminal.getTerminal
  @java_console = ConsoleReader.new(@java_input, @java_output)
  @java_console.setUseHistory(false)
  @java_console.setBellEnabled(true)

simply using @input.to_channel and @output.to_channel don't seem to work though.

[graaff]

This issue should probably be re-opened because this hang still happens with jruby 1.6.7.2 (ruby-1.8.7-p357) (2012-05-17 fffffff) (OpenJDK 64-Bit Server VM 1.6.0_24) [linux-amd64-java]

[graaff]

Forgot to add that this still happens with highline 1.6.13.

[JEG2]

I'll try to look into this when I have some time.

[mnzaki]

I have faced this problem while working on the jruby 1.7 support. I have made some preliminary changes that will alleviate the hang, but 18 tests fail.
The problem as of now (mnzaki/highline@2daa26e) is that the output contains the input that was passed as well, so the asserts fail. I think this happens because when using jline the ConsoleReader echos the input (unless you explicitly turn it off) while other methods depend on the underlying system to echo characters. One solution I see is to not even create a ConsoleReader when @input.tty? is false

[JEG2]

Yeah, that sounds reasonable. I would welcome such a fix.

[abinoam]

It seems io/console is available on recent vesions of JRuby and it works pretty similar to how it works on MRI.
So I'm trying to unify the approaches for MRI and JRuby on this PR #149

The test suit just 'skips' the readline cases when at jruby. Because I couldn't figured out a way to test them. (If you know how to do it, please tell me so).
So the tests will not hang indefinitely (waiting for console user input).

But you can be sure that it is working fine by using rake acceptance
This rake task will show you some use cases of HighLine and ask you if it's working or not.

By skipping this particular readline case on test suit we have all tests passing in travis https://travis-ci.org/abinoam/highline (jruby-19mode)

Could you people interested in (and familiar to) JRuby support give some feedback about the PR #149 ?

Hooking anyone that has commented in any HighLine JRuby issue: @rsutphin @mnzaki @graaff @Flameeyes @tomdz @andlaz @jeremyd @rsutphin @kgx @thinkgareth @presidentbeef @JEG2 @alexch @stomar @BanzaiMan @ai @celeduc @oreoshake @antond @jcoyne @keshavkuruva @cfbrobak @astounding

[abinoam]

... and I think PR #149 also closes issue #51

[abinoam]

After PR #149 I think most of JRuby problems are gone. Closing.