Le 11/03/2026 à 12:56, Rok Mihevc a écrit :
Option 1 - current FIXED_SIZE_LIST proposal
Option 2 - introduce VECTOR repetition type
Option 3 - move nullability into a new column, move towards gradually
removing definition/repetition levels
---
Hi all,
great to see new ideas and appetite to support vector-like data.
How much extra work do we expect Option 2 would be compared to Option 1? I
suppose at the minimum readers would have to be aware of the new repetition
type so they can safely ignore it, which IIUC would be most of the extra
work compared to Option 1. (All implementations would obviously need more
changes to read/write new types. But changes would be of
similar magnitude.)
There are no new types in Option 2. Option 1 does have a new type.
If we come to the conclusion Option 2 is not feasible now we are then
picking between two long term efforts - Option 2 vs Option 3.
I don't think Option 2 is "long term".
And I'm not sure why we're supposed to talk about Option 3, which is
almost like replacing Parquet with another file format. Please let's focus.
Regards
Antoine.
