Maybe better CRDTs for object_table, version_table and block_ref_table? #19

New Issue

Closed

opened 2020-11-21 15:01:01 +00:00 by lx · 1 comment

lx commented

2020-11-21 15:01:01 +00:00

Owner

(As started in #18)

But maybe not.

object_table: complex reconciliation logic, better suited for a custom CRDT.

version_table: we would need a crdt for a simple map, and not an LWWMap. Also we need to make it clear that Map CRDTs are ordered (they are, because the data is in fact just stored in a vec, it isn't even a btree or what), because the block list relies on the fact that blocks are ordered...

block_ref_table: the only crdt is a boolean, we could use crdt::Bool for that but then we would have to add a migration, and the logic is so simple it can't be messed up so there is no advantage of using crdt::Bool instead of just a bool

(As started in #18) But maybe not. object_table: complex reconciliation logic, better suited for a custom CRDT. version_table: we would need a crdt for a simple map, and not an LWWMap. Also we need to make it clear that Map CRDTs are ordered (they are, because the data is in fact just stored in a vec, it isn't even a btree or what), because the block list relies on the fact that blocks are ordered... block_ref_table: the only crdt is a boolean, we could use crdt::Bool for that but then we would have to add a migration, and the logic is so simple it can't be messed up so there is no advantage of using crdt::Bool instead of just a bool