fix reading properties without changing old configuration

[wuciawe]

it fixes #111, and unlike #112, it doesn't require to change old configuration, it's implemented inspired by the org.apache.spark.sql.execution.datasources.CaseInsensitiveMap in SPARK-SQL.

[pfcoperez]

@wuciawe We'll follow your approach since, as you well pointed out, would keep backward compatibility.
Thanks again for another good contribution.

[pfcoperez]

The logic here is wrong: You are filtering those config properties which does not contain the whole option list.

I'd try something like:

lowerCaseOptions.flatMap(key => config.properties.get(key).map(key -> _)) toMap

Or, cleaner:

private val clientOptions = {
    val lowerCaseOptions = MongodbConfig.ListMongoClientOptions.map(_.toLowerCase).toSet
    config.properties.filterKeys(lowerCaseOptions contains _)
  }

[pfcoperez]

Looking at your commits I see you had already noticed that filterKeys isn't compatible with a map to be distributed to Spark workers as its return instance is not serializable. In any case, you should apply the logic I proposed:

 private val clientOptions = {
   val lowerCaseOptions = MongodbConfig.ListMongoClientOptions.map(_.toLowerCase).toSet
   config.properties.filter { case (k, _) => lowerCaseOptions contains k }
 }

[pfcoperez]

@wuciawe Change as suggested, before committing make sure you pass the project test set, we'll pass them for a second time and if everything is OK we'll merge

[wuciawe]

@pfcoperez I merged your suggestion and project tests all pass on local machine. Sorry for wrong logic produced in previous pull request, I'm surprised that the wrong code compiles. Is it necessary to squash all those commits into one this time?

[pfcoperez]

@wuciawe No worries. I was surprised too, probably a consequence WrappedString#CanBuildFrom.
You don't need to squash I'll do it using new GitHub merge tool.

Thanks again.