[SPARK-20395][BUILD] Update Scala to 2.11.11 and zinc to 0.3.15

[srowen]

What changes were proposed in this pull request?

Update Scala to 2.11.11 and zinc to 0.3.15

How was this patch tested?

Existing tests.

[SparkQA]

Test build #76933 has finished for PR 17982 at commit 2028dbd.

• This patch fails to build.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

@ScrapCodes it's good you're here because this might be something you can help with. (CC @dragos as well)

I'm updating to scala 2.11.11, and it looks like the loadFiles method that SparkILoop hooks into was removed in 2.11.9. I think it's about the same to hook into process() instead and do Spark init after it's done. Do you know of any reason that's a bad idea? I'm pushing the change anyway to see.

[ScrapCodes]

Hi Sean,
Taking a quick look at the changes that removed loadFiles. changes

It seems startup is the correct alternative, i.e. just before the call to printWelcome.
See startup.

[srowen]

@ScrapCodes ah I should have pointed out the other constraint: we want this to still work with 2.11.8. Right now Spark would not run vs Scala 2.11.9+ at runtime, but if we tried to hook into startup it wouldn't work with 2.11.8 because that method didn't exist. That's why I was looking at process.

[ScrapCodes]

Actually, now I am wondering, how would it work without createInerpreter is executed. Your constraint, makes things a bit tricky. I will think more before I reply.

[ScrapCodes]

If I am not able to come up with the better alternative, then following not so ideal option exist.

1. we can do a scala version check and have two different code paths based on scala version.
   In scala 2.11.8+ we can hook our init just before printWelcome() [EDIT] This won't help either. :(
   In scala 2.11.11 we can do it just before loopPostInit().

[dragos]

If I'm not mistaken the current patch will fail on files passed via -i to spark-shell, since Spark is initialized after process is done (the SparkContext is not available during initialization, when -i files are loaded).

I'm not sure if there's a better way, especially one that works on all 2.11 series. Mea culpa for not adding a comment on loadFiles in the Scala REPL when I worked on the Spark Shell. Maybe @som-snytt has a suggestion for inserting initializeSpark in the initialization pipeline?

[SparkQA]

Test build #76936 has finished for PR 17982 at commit 794d18d.

• This patch fails from timeout after a configured wait of `250m`.
• This patch merges cleanly.
• This patch adds no public classes.

[som-snytt]

The contract was never well-defined.

sbt overrides createInterpreter and runs its initialCommands before returning. So that runs before REPL finishes init in loopPostInit: no $intp, power mode not entered, start code and -i not run yet.

I can't tell just by looking if that suits your needs here, but it's the most ancient hook enshrined by sbt.

https://github.com/sbt/sbt/blob/0.13/compile/interface/src/main/scala/xsbt/ConsoleInterface.scala#L27

@adriaanm is taking executive action to modularize/make API for REPL reuse. Possibly you're aware or he's only just started and is in stealth mode. That will address the issue in future.

[srowen]

Yeah, this was always a bit of a hack as this isn't an API. createInterpreter sounds like a promising place to hook into for now as it exists across these versions, and would mean Spark is initted a little earlier, before instead of after this middle block in here:

  def process(settings: Settings): Boolean = savingContextLoader {
    this.settings = settings
    createInterpreter()

    // sets in to some kind of reader depending on environmental cues
    in = in0.fold(chooseReader(settings))(r => SimpleReader(r, out, interactive = true))
    globalFuture = future {
      intp.initializeSynchronous()
      loopPostInit()
      !intp.reporter.hasErrors
    }
    loadFiles(settings)
    printWelcome()
...

[SparkQA]

Test build #76946 has finished for PR 17982 at commit e1482c3.

• This patch fails from timeout after a configured wait of `250m`.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #3728 has started for PR 17982 at commit e1482c3.

[SparkQA]

Test build #3741 has finished for PR 17982 at commit e1482c3.

• This patch fails from timeout after a configured wait of `250m`.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #77096 has finished for PR 17982 at commit b5a7035.

• This patch fails from timeout after a configured wait of `250m`.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

Darn. I don't know if this is going to work. I'm closing this for now.

[som-snytt]

Thanks for the effort. I'll take a hack soon. If it's hopeless, I'll at least try to track developments with the new REPL API.

[tovbinm]

@som-snytt @srowen what about 2.11.12?

[srowen]

Unless it changed the internals of the interpreter to afford a different place to hack in the init, don't think it is any different. However the build and most tests already work with 2.12.d

[tovbinm]

2.12 would be even better. I haven't seen a ticket related to 2.12 upgrade. I can give it a try.