Add query support for $collation (Mongo 3.4+)

[ericyhwang]

Background

Mongo 3.4 added support for collation in string comparisons, for finding and sorting. This enables, among other things, sorting while ignoring case and accent differences based on a provided locale.

With a plain Mongo JS client, a case-insensitive sort would use cursor.collation:

db.my_collection.find(query)
  .sort({myStringField: 1})
  .limit(1)
  .collation({locale: 'en_US', strength: 1 /*or 2*/});

With a larger result set, a matching case-insensitive index would also be desirable for performance.

sharedb-mongo doesn't support collations yet in queries. Yet! That's what this is about.

In the meantime, there are some workarounds for doing case-insensitive sorting.

Implementation

Thankfully, it's straightforward to add support in sharedb-mongo for collation.

The usage that's equivalent to the Mongo JS driver example above would look like this:

connection.createSubscribeQuery('my_collection', {
  $sort: { myStringField: 1 },
  $collation: {locale: 'en_US', strength: 1 /*or 2*/},
});

sharedb-mongo has a map from dollar-prefixed query properties to cursor methods, and supporting collation is as easy as adding a $collation entry.

There are a couple additional pieces:

• Tests: The one slightly tricky part with the tests is that Share runs its tests against Mongo server 2.6, 3.6, and 4.0, and collation won't work when running against 2.6. I believe it should be possible to have the test code check the MONGODB_VERSION env variable set in Travis and conditionally run specific test cases that way.
• Error handling: Since collation was added in Mongo 3.4, any server versions before that will not handle collation.
  • Whoever implements this should check behavior against older servers and make sure something sensible happens error-wise.
  • Things are mostly fine client-wise. sharedb-mongo depends on mongodb ^2.1.2. That will result in installing [email protected], which does support cursor.collation. Users can pass in a pre-created Mongo client, though, which may be an issue.