@granturing Appreciate for the patch!
I'll manually test it in this evening with my windows XP and share the result.

It works well on my windows XP with oracle jdk 1.7.0_79. great work!

It works on Windows 7 as well.

I tried it on Windows 7, works well.

Except for these two test failure in cassandra, when I do
SPARK_VER="1.6.0" HADOOP_VER="2.3" PROFILE="-Pspark-1.6 -Phadoop-2.3 -Ppyspark -Pscalding" BUILD_FLAG="package -Pbuild-distr" TEST_FLAG="verify -Pusing-packaged-distr" TEST_PROJECTS=""
mvn $BUILD_FLAG $PROFILE -B

    - should parse multi-line comment *** FAILED ***
    [6.7] parsed: Comment(This is a comment

    line1

    line2

    line3

    ) did not match the given pattern (ParagraphParserTest.scala:113)

    - should parse a block queries with comments *** FAILED ***
    [24.7] parsed: List(Comment(

    This example show how to force a

    timestamp on the query

    ), Comment(Timestamp in the past), Timestamp(10), SimpleStm(CREATE TABLE IF NOT EXISTS spark_demo.ts(key int PRIMARY KEY, value text);), SimpleStm(TRUNCATE spark_demo.ts;), Comment(Force timestamp directly in the first INSERT), SimpleStm(INSERT INTO spark_demo.ts(key,value) VALUES(1,'val1') USING TIMESTAMP 100;), Comment(Selec   t some data to loose some time), SimpleStm(SELECT * FROM spark_demo.albums LIMIT 100;), Comment(Use @timestamp value set at the beginning(10)), SimpleStm(INSERT INTO spark_demo.ts(key,value) VALUES(1,'val2');), Comment(Check the result), SimpleStm(SELECT * FROM spark_demo.ts WHERE key=1;)) did not match the given pattern (ParagraphParse   rTest.scala:427)

@doanduyhai is cassandra supported on windows?

@felixcheung

Yes, Cassandra is supported on Windows.

The error is about parsing multi line comments. The first test failure looks interesting. Basically it says that the following test fails:

  "Parser" should "parse multi-line comment" in {
    val query:String =
      """/*This is a comment
        |line1
        |line2
        |line3
        |*/
      """.stripMargin

    val parsed = parser.parseAll(parser.multiLineComment, query)
    parsed should matchPattern {
      case parser.Success(Comment("This is a comment\nline1\nline2\nline3\n"), _) =>
    }
  }

because it does not match the regexp pattern: (?s)/\*(.*)\*/ e.g:

• (?s) = DOT ALL mode
• \* = start with \*
• (.*) = any character
• */ = end with */

I'm wondering if it is not related to the difference between Linux/Unix and Windows for the line terminator

Let me checkout the project on Windows and try (I need to find a Windows :D)

The last CI failure is because of spark release download timeout.
I think we are good to go - how about we track the Windows Cassandra test failure in a new JIRA @prabhjyotsingh @doanduyhai ?

Hi Felix, agree with the Cassandra test failure. I have a couple more things I want to verify and then we should be good to go. I'm writing up some docs related to Windows deployment that will be part of this PR.

+1 for track the Windows Cassandra test failure in a new JIRA

So, interestingly enough I only see an issue with Spark (using Hadoop 2.6). Running Flink seems fine, but it could be dependent on the Hadoop client versions being used by the two. But yes, the HDFS interpreter will need to have the necessary Windows dependencies. You want me to pull that line out?

Hmm, interesting. Let's keep this here for now then.

Couple questions:

• The Shell interpreter uses bash, which obviously won't work on Windows. I've confirmed the interpreter will work with a couple changes to support the Windows CMD shell. I can either set a value to decide which interpreter to use at runtime, based on the OS, or we make it configurable. Is it worth making it configurable?

  private static final boolean isWindows = System.getProperty("os.name")
          .startsWith("Windows");

  private final String shell = isWindows ? "cmd /c" : "bash -c";

• Units tests in the zeppelin-interpreter project have a reference to "../bin/interpreter.sh", which of course won't work on Windows. I've verified the unit tests using "../bin/interpreter.cmd". I just didn't like the fact that when I'm in Windows I can't run all the unit tests. I can either have a static variable which defines what script to use in each unit test or I add a helper class to centralize it and all unit tests reference that. What would you prefer?

  private static final String INTERPRETER_SCRIPT =
          System.getProperty("os.name").startsWith("Windows") ?
                  "../bin/interpreter.cmd" :
                  "../bin/interpreter.sh";

I've been running this extensively the past few weeks and have nothing left to add at this point, unless there's additional feedback.

@granturing Thanks for great work. LGTM. Can #749 and #769 also be applied?

The script seems to have a problem with CLASSPATH's that contain spaces (e.g. C:\Program Files\xyz\bin). I'm not knowledgeable enough with with Windows Batch scripting to determine if this a problem with the startup script or my particular environment.

The script fails in the following section of "zeppelin.cmd":

if "%CLASSPATH%"=="" (
    set CLASSPATH=%ZEPPELIN_CLASSPATH%
) else (
    set CLASSPATH=%CLASSPATH%;%ZEPPELIN_CLASSPATH%
)

The error thrown:
Error: Could not find or load main class Files\xyx\bin\

@dnldxn thanks for testing that. I changed a few things to better handle spaces. Unfortunately, using the bundled Spark with a ZEPPELIN_HOME with spaces will not work, due to Spark issues. External Spark works fine though.

Is it okay to be merged?

I have no more changes. @dnldxn were you able to validate the latest commit on your setup?

@granturing Yep, it works great on my Windows 10 system.

@granturing Selenium tests failed, could you close/re-open this PR to launch CI again?

One version failed because it couldn't download Spark, the other because of connection refused errors running API tests. My last commit was only against the Windows batch files which aren't being tested in the CI build though.

Most of the profiles are green, including those with tests, so LGTM

LGTM. Merge if there're no more discussions.