INTPYTHON-520 Ensure all parquet data types are handled

[sleepyStick]

I think this is what we want???
edit: docs will be completed via a docs ticket but in general the only parquet types that we don't inherently already support are on the left and what they can be converted to (that we do support) are on the right.

• uint8 -> pa.int32
• uint16 -> pa.int32
• uint32 -> pa.int64
• uint64 -> pa.int64
• int8 -> pa.int32
• int16 -> pa.int32
• float16 -> pa.float64
• float32 -> pa.float64
• float64 -> pa.float64
• timedelta -> pa.int64

[sleepyStick]

3.13t tests are failing because pandas (which I import for the test) isn't installed..is that something i need to fix here?

[blink1073]

3.13t tests are failing because pandas (which I import for the test) isn't installed..is that something i need to fix here?

The tests are run with and without optional deps (polars and pandas), so they need to have a skip condition when pandas isn't installed.

[blink1073]

Hmm, do we want to offer a "best effort" conversion mode where we automatically make these conversions in write()?

[sleepyStick]

Hmm, do we want to offer a "best effort" conversion mode where we automatically make these conversions in write()?

hmm would that look like adding an optional parameter to write? where if we do best effort conversion unless specified (via the parameter)?
(to be clear, all for the idea, just wanna make sure i'm envisioning it the same way as you? and if not, to see what the differences are)

[blink1073]

I see it as similar to allow_invalid in aggregate_polars_all and friends. It would be a single flag that would convert unsupported types to supported ones.

[blink1073]

Very nice, thank you!