[SPARK-12581][SQL] Support case-sensitive table names in postgresql

[maropu]

Table names in postgresql is originally case-insensitive.
To support case-sensitive table names, we need to wrap names in double quotes.

[maropu]

retest this please

[SparkQA]

Test build #48482 has finished for PR 10523 at commit 65c048c.

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

[SparkQA]

Test build #48703 has finished for PR 10523 at commit d261683.

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

[maropu]

retest this please

[SparkQA]

Test build #48725 has finished for PR 10523 at commit ba0e434.

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

[SparkQA]

Test build #48723 has finished for PR 10523 at commit ba0e434.

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

[SparkQA]

Test build #48730 has finished for PR 10523 at commit f5537e9.

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

[SparkQA]

Test build #48734 has finished for PR 10523 at commit a6e3a73.

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

[maropu]

@liancheng @yhuai Could you review this?

[yhuai]

@maropu Thank you for the PR. So, when you quote a table name in postgres, the name is case-sensitive, right? Looks like we are always quoting the table name? If so, seems we are changing the semantic because if I use a table name fOO and I want it case-insensitive, we still treat it as a case sensitive table name?

[maropu]

@yhuai Yes, quoted tables in postgres are always case-sensitive. We need to support case-insensitive table names? Table names in sparksql (DataFrame#registerTempTable) and typical databases such as oracle and mysql are also case-sensitive, so IMO we need to comply with the rule.

[yhuai]

@maropu Let me explain. The case sensitivity in spark sql is actually configurable. For HiveContext, we use case insensitive resolution.

[maropu]

@yhuai Ah..., I misunderstood. Okay and how about turning on/off case sensitivity in configurations, e.g., spark.sql.jdbc.caseSensitivity? For example, in case-insensitive mode, we always uncapitalize table names input in jdbc, and in case-sensitive one, we apply JdbcDialects#quoiteIdentifier.

[yhuai]

Is it a postgresql specific thing? Maybe it is better to allow users to set a conf in the data source optionsS (a conf that can be set by doing sqlContext.read.option(key, value)) instead of introducing a new sql conf?

[maropu]

@yhuai yes and it's a pg-specific thing. I agree that we add it as a data source option instead of sql conf. If the fix is acceptable, I'll fix this pr to add a data source option like sqlContext.read.option("caseSensitiveTableName" -> "true").

[SparkQA]

Test build #48927 has finished for PR 10523 at commit 98c7af9.

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

[maropu]

Close this for now and, if needed, I'll reopen this.