Parquet regression: exceptions.ArrowErrorException: NotYetImplemented("Can't read Dictionary(UInt32, LargeUtf8, false) from parquet")

[ritchie46]

CI round trips fail on writing/reading dictionary arrays with keys UInt32 and values LargeUtf8.

[jorgecarleitao]

The issue is that we do not have statistics structs for dictionary arrays - the dictionary is read, but the statistics are not.

[ritchie46]

So I trigger that error when reading statistics? I need to check the error type and ignore the statistic on this error type?

Or should I check the schema first and if any dict in there don't try to parse statistics?

[ritchie46]

What did we do before this? Just not include the dictionary column in the statistics?

[jorgecarleitao]

I am working on this - the statistics API is clumsy (we should just have stats as arrays, not individual values), I will just fix this together with the API

[ritchie46]

All right. I'll wait a bit. :)

[ritchie46]

Fixed. Confirmed by the CI run in pola-rs/polars#3181