Deploying UDFs to Azure Databricks

[alastairs]

I've been following the README instructions to deploy my .NET Spark app (including a user-defined function) to Azure Databricks. I've found the application runs correctly, except for the UDF, which fails with the documented file not found error. I've run into the following unexpected behaviour related to that:

1. My UDF is defined in the same assembly as the Spark program. Why do I need to upload it to DBFS or similar and set DOTNET_ASSEMBLY_SEARCH_PATHS for it to be resolved?
2. Uploading a DLL to the same path as the zip file containing my application results in a JVM error from Spark trying to load it as a JAR, and causes the job to fail before it even starts.
3. Specifying DOTNET_ASSEMBLY_SEARCH_PATHS=. does not resolve the file not found error. The documentation is vague on what . means, and I'm surprised that this a) isn't a default search path anyway, and b) the required assembly isn't in that path (whatever it resolves to).

Thanks in advance!