from_json should just work out of the box for it. Scan for JSON will need to be modified because we don't want to pass in a buffer. As we read the data a row at a time we want to turn it into a string column and send it to the GPU. Depending on what we want to do we can either filter out empty rows on the CPU before sending to the GPU or after. Either way once the input data is filtered, then we call into the parsing just like we would for from_json.
Is your feature request related to a problem? Please describe. Once https://github.com/NVIDIA/spark-rapids-jni/issues/2035 is done then we can start to move over to use it for our parsing.
from_json should just work out of the box for it. Scan for JSON will need to be modified because we don't want to pass in a buffer. As we read the data a row at a time we want to turn it into a string column and send it to the GPU. Depending on what we want to do we can either filter out empty rows on the CPU before sending to the GPU or after. Either way once the input data is filtered, then we call into the parsing just like we would for from_json.