Support output column on a different level
- Dominant language
- Scala
- Stars
- 33
- Forks
- 16
- PR merge metrics
- No merged PRs in 30d
Description
## Background
Currently, conformance rules require that output columns be at the same level as input columns (e.g. casting rule). This seems like a too strict limitation.
## Feature
Output columns can be at different struct levels as long as the array boundary is not crossed.
## Example [Optional]
Here is an example schema:
```
root
|-- id: long (nullable = true)
|-- key1: long (nullable = true)
|-- key2: long (nullable = true)
|-- struct1: struct (nullable = true)
| |-- key3: integer (nullable = true)
| |-- key4: integer (nullable = true)
|-- array1: array (nullable = true)
| |-- element: struct
| | |-- key7: long (nullable = true)
| | |-- key8: long (nullable = true)
| | |-- skey2: string (nullable = true)
```
- Currently you *can't* use the casting tole to convert `struct.key3` from to `string` and save it to `key3_String`. The output column should be at the same level, e.g. `struct1.key3_String`. This should be *allowed*.
- Converting `array1.key7` to `string` and saving it to `key7_String` should be *disallowed* since it crosses the array boundary. The conversion can only be allowed if the output column is also inside the array, e.g. `array1.key7_String`.
## Proposed Solution [Optional]
`spark-hats` allows a unified way of accessing columns from arbitrary levels of nesting as long as array boundaries are not crossed. We can utilize that in conformance rules. This is similar to how it is done for the broadcasting mapping rule strategy.
Contributor guide
Assessment
This issue has not been assessed yet.