Added - Support for `through` on node result set expressions.
Fixed - `unit` on properties in data modeling. This was typed as a string, but it is in fact a direct relation.
6.34.2
Fixed - Loading a `ContainerApply` from source failed with `KeyError` if `nullable`, `autoIncrement`, or `cursorable` were not set in the `ContainerProperty` and `BTreeIndex` classes even though they are optional. This is now fixed.
6.34.1
Added - Support for setting `data_set_id` and `metadata` in `ThreeDModelsAPI.create`. - Support for updating `data_set_id` in `ThreeDModelsAPI.update`.
6.34.0
Fixed - `PropertyType`s no longer fail on instantiation, but warn on missing SDK support for the new property(-ies).
Added - `PropertyType`s `Float32`, `Float64`, `Int32`, `Int64` now support `unit`.
6.33.3
Added - `functions.create()` now accepts a `data_set_id` parameter. Note: This is not for the Cognite function, but for the zipfile containing the source code files that is uploaded on the user's behalf (from which the function is then created). Specifying a data set may help resolve the error 'Resource not found' (403) that happens when a user is not allowed to create files outside a data set.
6.33.2
Fixed - When fetching datapoints from "a few time series" (implementation detail), all missing, non-ignorable time series are now raised together in a `CogniteNotFoundError` rather than only the first encountered.
Improved - Datapoints fetching has a lower peak memory consumption when fetching from multiple time series simultaneously.