Yes, it's a beauty in some scenarios, but not in all.
We (with Hubert š ) are actually trying to build a generic service that can load parquet files from defined locations. The parquet files are created by upstream services.
We want to be able to detect schema changes and classify them as:
⢠non-breaking changes - new column added - allowed
⢠breaking changes - column removed - not allowed
For example if a completely different file is dropped into a location by mistake, we don't want autoloader to create "n" additional columns and fill the rest with NULLs. š