[SPARK-6865][SQL] DataFrame column names should be treated as string literals

[rxin]

For example, "a.b" should match a column named a.b.

[rxin]

note @marmbrus our new resolver semantics breaks this test. Not sure how important it is.

[marmbrus]

The more I think about this, the more I am worried that we can't make a change this large. There is no way to express self join queries if we don't handle . in column names. We are also going to break lots of existing user code...

[SparkQA]

Test build #30221 has finished for PR 5505 at commit 36f63a4.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.
• This patch does not change any dependencies.

[rxin]

and @liancheng I had to disable this test as well since it used "tablename.columnname".

[liancheng]

I guess it should be OK to disable or even remove this test now, since now we check for invalid field names explicitly and suggest users to add aliases. See #5263.

[SparkQA]

Test build #30227 has finished for PR 5505 at commit a5f6983.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.
• This patch does not change any dependencies.

[rxin]

I discussed with michael offline -- given this would break self-join, we've decided to treat dot as a special case.

[cloud-fan]

As #5638 handled self join correctly, should we reopen this PR?

[rxin]

That one actually doesn't handle most self join cases, since very often in self joins you join on different keys.