This makes Stash's database handler a bit more robust. The changes
started as me trying to add an entry expiry, but I've realized that the
database system is a little fragile and it assumed the database does not
change, ever. Well that's not true, it does change and when it does
there's a chance that everything implodes.
We now wrap migrations in transaction for atomicity and track version
via PRAGMA user_version (0 -> 3). We also check column existence before
ALTER TABLE and use `last_insert_rowid()` instead of `next_sequence()`.
Last but not least, a bunch of regression tests have been added to the
database system because I'd rather not discover regressions in
production.
Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: Ifeab42b0816a5161d736767cb82065346a6a6964
Adds a `content_hash` column and index for deduplication, and a
`last_accessed` column & index for time tracking. We now de-duplicate on
copy by not copying if present, but instead bubbling up matching entry.
Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: Icbcdbd6ac28bbb21324785cae30911f96a6a6964
This includes breaking changes to the database entries, where we have
started deduplicating based on hashes instead of full entries. Entry
collisions are possible, but highly unlikely.
Additionally we use `Box<str>` for error variants to reduce allocations.
This is *yet* to give me a non-marginal performance benefit but doesn't
hurt to be more correct.
Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: I6a6a6964d0a33392da61372214ca3088551564ac
