This commit removes usage of the iterator and replaces map key and positional parameter mappings with an index based token lookup.
Closes#3921
Original pull request: #3930.
Introduce EmptyDocument and utility methods in BsonUtils. Avoid entrySet and iterator creation for document iterations/inspections.
Original Pull Request: #3809
SessionSynchronization.NEVER bypasses all transactional integration in cases where applications do not want to make use of transactions so that transaction inspection overhead is avoided.
Original Pull Request: #3809
We now no longer attempt to convert query Documents into primitive types to avoid e.g. Document to String conversion.
Closes: #3783
Original Pull Request: #3797
Along the lines make sure to convert map like structures correctly if they do not come as a Document, eg. cause they got converted to a plain Map in a post load, pre convert event.
Closes#3702
Original pull request: #3704.
This commit fixes a regression that prevented custom converters from being applied to types considered store native ones.
Original pull request: #3703.
Fixes#3670
While maps that have numeric keys work if there is only one map with an integer key, when there are multiple maps with numeric keys in a given query, it fails.
Take the following example for a map called outer with numeric keys holding reference to another object with a map called inner with numeric keys: Updates that are meant to generate {"$set": {"outerMap.1234.inner.5678": "hello"}} are instead generating {"$set": {"outerMap.1234.inner.inner": "hello"}}, repeating the later map property name instead of using the integer key value.
This commit adds unit tests both for the UpdateMapper and QueryMapper, which check multiple consecutive maps with numeric keys, and adds a fix in the KeyMapper. Because we cannot easily change the path parsing to somehow parse path parts corresponding to map keys differently, we address the issue in the KeyMapper. We keep track of the partial path corresponding to the current property and use it to skip adding the duplicated property name for the map to the query, and instead add the key.
This is a bit redundant in that we now have both an iterator and an index-based way of accessing the path parts, but it gets the tests passing and fixes the issue without making a large change to the current approach.
Fixes: #3688
Original Pull Request: #3689
