[SPARK-17685][SQL] Make SortMergeJoinExec's currentVars is null when calling createJoinKey

[wangyum]

What changes were proposed in this pull request?

The following SQL query cause IndexOutOfBoundsException issue when LIMIT > 1310720:

CREATE TABLE tab1(int int, int2 int, str string);
CREATE TABLE tab2(int int, int2 int, str string);
INSERT INTO tab1 values(1,1,'str');
INSERT INTO tab1 values(2,2,'str');
INSERT INTO tab2 values(1,1,'str');
INSERT INTO tab2 values(2,3,'str');

SELECT
  count(*)
FROM
  (
    SELECT t1.int, t2.int2 
    FROM (SELECT * FROM tab1 LIMIT 1310721) t1
    INNER JOIN (SELECT * FROM tab2 LIMIT 1310721) t2 
    ON (t1.int = t2.int AND t1.int2 = t2.int2)
  ) t;

This pull request fix this issue.

How was this patch tested?

New test case added in DataFrameJoinSuite.

[SparkQA]

Test build #65963 has finished for PR 15259 at commit 2bb54b5.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[hvanhovell]

@wangyum could you explain what causes this problem?

[wangyum]

@hvanhovell

The following SQL query thrown IndexOutOfBoundsException:

SELECT
  count(int)
FROM
  (
    SELECT t1.int, t2.int2 
    FROM (SELECT * FROM tab1 LIMIT 1310721) t1
    INNER JOIN (SELECT * FROM tab2 LIMIT 1310721) t2 
    ON (t1.int = t2.int AND t1.int2 = t2.int2)
  ) t;

SortMergeJoinExec.createJoinKey()‘sctx.currentVars is the expression of count(int), keys is the join condition int and int2, as shown in the following screenshot:

---

ordinal is the key's index of keys when constructingBoundReference, as shown in the following screenshot:

---

So ctx.currentVars(ordinal)(ordinal=1) will throw IndexOutOfBoundsException when doGenCode, as shown in the following screenshot:

---

I think ctx.currentVars have nothing to do with keys at this time, so I want to set ctx.currentVars=null to avoid IndexOutOfBoundsException.

[wangyum]

@hvanhovell I updated PR description.

[wangyum]

Jenkins, retest this please

[SparkQA]

Test build #76610 has finished for PR 15259 at commit 2bb54b5.

• This patch fails SparkR unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #76645 has started for PR 15259 at commit 2bb54b5.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/76645/
Test FAILed.