Skip to content

Nullability Sentinel Very Hard to Use #352

Open
@WillAyd

Description

@WillAyd

This is arguably more feedback than an issue / question, but while I was trying to use the DataFrame Interchange API I found the requirement of supporting custom sentinels as a consumer rather burdensome. As far as I understand, if the array describes itself as using a sentinel, then it is up the consumer to query describe_null() and convert that PyObject to whatever type works with the buffer.

For types that span multiple words (ex: 256 bit decimal), such a sentinel seems really hard if not impossible to support. Even for more primitive types, you have to jump through some hoops with the CPython API to do that translation.

Trying to support all of those cases in a lower level extension makes it a ton of effort to fully adhere to this specification

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions