I’m working on removing both these storage formats, leaving us with just the 
SplitRowStore as the only rowstore format. This new format will also be the 
default format for temp blocks.

Are there any objections? Note that:
* It will be a feature regression in that we’ll no longer have a compressed row 
storage format at all. I don’t believe compression is useful currently anyway, 
and I’m planning to redo it from scratch in a few weeks.
* It will be a breaking change, not backwards compatible with existing stored 
catalogs. I’m NOT planning to write a new deserializer that allows backward 
compatibility with existing PackedRowStore on-disk relations.
* There will probably be no performance regression. I’ll try to make sure that 
the bulkInsert functions for SplitRowStore are not slowed down by 
variable-length attribute checks in the inner loop.

Reply via email to