[SPARK-6444] [SQL] A quick fix for SPARK-6444

[liancheng]

As explained in the JIRA ticket comment, this PR introduces a simple quick fix for SPARK-6444. The root cause of this issue is that we don't check data types of input arguments for SQL functions. PR #4685 is heading the right direction. The reason why moving PromoteStrings before WidenTypes fixes this issue is that, PromoteStrings helps Sum to check and convert the data type of its argument.

[SparkQA]

Test build #28961 has started for PR 5127 at commit cd8f778.

• This patch merges cleanly.

[SparkQA]

Test build #28961 timed out for PR 5127 at commit cd8f778 after a configured wait of 120m.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28961/
Test FAILed.

[liancheng]

retest this please

[SparkQA]

Test build #28995 has started for PR 5127 at commit cd8f778.

• This patch merges cleanly.

[SparkQA]

Test build #28995 has finished for PR 5127 at commit cd8f778.

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

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28995/
Test PASSed.

[liancheng]

@yhuai Would you mind to take a look? This is the UNION ALL case we've discussed offline the other day. Thanks!

[liancheng]

Discussed with @yhuai, we don't want to backport this to branch-1.3 since it's not a regression. So we have plenty of time to deliver a proper fix for this. Closing it.