Jay Sprenkle wrote:
Every time you open a :memory: database you get a separate instance, identified only by its sqlite3* handle. I'm not sure if ATTACH would work with :memory:, but even if it does it would just create a new, empty in-memory database, not refer to the one (of possibly many) you already have open and populated.sqlite> create table x(y text); sqlite> insert into x(y) values('one'); sqlite> select * from x; one sqlite> attach ":memory:" as db1; sqlite> create table db1.x(y text); sqlite> insert into db1.x(y) values('two'); sqlite> select * from db1.x; two sqlite> select * from x; one sqlite> attach ":memory:" as db2; sqlite> create table db2.x(y text); sqlite> insert into db2.x(y) values('three'); sqlite> select * from db2.x; three sqlite> select * from db1.x; two sqlite> select * from x; one sqlite>
Would that work with pre-existing memory databases? Here, you create several new memory databases attached to an existing one. The whole contraption hangs off of a single sqlite* handle. For all intents and purposes, you have a single in-memory database, just with somewhat unusual table names.
This technique may be cute, but I don't quite see how it's useful. It defeats the whole point of having multiple independent memory databases - for example the ability to write to them concurrently from different threads.
Igor Tandetnik
