[SPARK-13616][SQL] Following PR for SPARK-13616

[viirya]

What changes were proposed in this pull request?

As all operators except Join should support converting back to SQL when it is top node. We should only take care of Join for this purpose.

How was this patch tested?

All existing tests should be passed.

[viirya]

Fix the problem mentioned by @gatorsmile in #11466.

cc @rxin @liancheng

[liancheng]

Hey @viirya, I just commented in #11466 and I tend to revert #11466 first and revisit it after we have full SQL generation support for all major language structures.

Another thing confusing me is that, although @gatorsmile mentioned that #11466 brought regressions, I didn't find them (Jenkins PR builder looks fine).

[dilipbiswal]

@liancheng FYI - I did hit a problem while working on SQL generation for Generate. I found that Limit is not in the list of cases of top nodes and my test cases with Limit clause started failing after i rebased with master.

[SparkQA]

Test build #52515 has finished for PR 11537 at commit 1ae4f8b.

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

[liancheng]

@dilipbiswal Thanks, I think I'd better revert #11466 first for safe.

[viirya]

Let me close this now.

[viirya]

@liancheng I have no idea about these regressions. At least #11466 passes all tests when testing in Jenkins.

[gatorsmile]

We have to locally run "hive/test-only org.apache.spark.sql.hive.execution.HiveCompatibilitySuite" and check the log file, as mentioned in the JIRA: https://issues.apache.org/jira/browse/SPARK-11012

Yeah, the regressions does not stop the test cases. If a logical plan is inconvertible, HiveComparisonTest falls back to its original behavior. When we can complete SQL generation for all the all the test cases in HiveCompatibilitySuite, we need to disable the auto fallback.

Definitely, we need to add more test cases to LogicalPlanToSQLSuite.