William Ayd

If you drop a debugger here you will also be able to see the values at a lower level: https://github.com/innobi/pantab/blob/7cf2421047b90ca4b32d9e6326d681bc5ca26dfa/src/pantab/writer.cpp#L821 On the sixth loop iteration, any of the code samples...

That's a nice idea kyle - actually I think the easiest way to validate is to use nanoarrow: ```python import nanoarrow as na capsule = df_filtered.__arrow_c_stream__() stream = na.ArrayStream(capsule) with...

Ah OK great thanks! I can reproduce with that - will take a closer look

OK after looking at this some more, I think the actual issue is that the Tableau Hyper API may choose to reorder elements. The rules for that are not clear...

OK thanks - and you only see it when working with "date" types right? It seems like it may also be related to writing a pyarrow table in batches, whereas...

Thanks everyone for the continued discussion. Just so I am on the same page, do we think the issue is only with polars at this point in time or do...

Thanks @dvukolov for continued examples. Can you try to create an MRE out of that? See https://matthewrocklin.com/minimal-bug-reports for some guidance on that

Thanks @dvukolov that first example is _extremely_ helpful. I can confirm that the issue is in writer.cpp when the Date32 values are being written. For whatever reason, each iteration of...

I'll have to look at how cibuildwheel is set up. My (possibly flawed) understanding is that the generation of the sdist and repairing of the wheel happen in two different...

The Hyper database (and to be fair, most major databases) do not offer an unsigned integral storage type. I think what the PostgreSQL ADBC driver does is map: - uint8...