@Niro can you please add context to this issue? Trying to understand if this is the tension between APIs that provide separate "open" and "create" routines and APIs that provide a single "open" routine, or something else. (Personally I prefer the former option whenever there is no read+write nice, as then the two return tours are completely disjoint and it is typesafe!)
11/23/2023, 11:32 AM
This was open as part of a discussion I had with @Oz Katz regarding the expected functionality (and compatibility with fsspec) and support of a writer object.
As I understand the initial PRD suggested adding write modes to
, which was replaced by a
method which is provided with the data.
This design constricts us in scenarios where we want to provide a writer to another process which performs the data fabircation