Before this change, we've tried to open the database in read/write, but not create mode. In situations where the database didn't exist yet, this logged an error to the console, and we proceeded to opening it again in read/write/create mode, which actually created the file. The reason we did this is so that we could detect really old caching databases from January 2016 in case a developer upgraded from an older SDK (iOS v3.1 and earlier, Android v3.2 and earlier) that didn't have https://github.com/mapbox/mapbox-gl-native/pull/3715 yet.
However, these error messages, while innocent, look scary in the console and some users suspect that it's a bug. This change opens the file directly in read/write/create mode, omitting the first failed attempt. To handle old cache databases, we're now deleting the `http_cache` table, which was the only table in those old databases, and create the new schema, rather than deleting the entire file and recreating the Database object. In most scenarios, this will lead to one fewer opening attempt, while the database migration will continue to work for the few users who upgrade all the way from a January 2016 SDK.
Additionally, this fixes a mismatch between the Qt and non-Qt implementation: Qt doesn't support opening a file in read/write mode without the create flag. This means that we've seen a different control flow on Qt compared to the non-Qt implementation when opening a database.
Unfortuntely, it's difficult to avoid all exceptions, because sqlite3_open_v2 does not reliably return SQLITE_NOTADB if the file is not a database. However, this should avoid cases where developers misinterpret the SQLITE_CANTOPEN exception as a crash, which is the common case.
- puts function definitions in a namespace ... {} rather than using namespace ...;
- remove trailing whitespace
- add trailing newline
- protect SQL statements from being formatted by clang-format
The migration was upgrading the schema, but not the schema version. As a result, the (expensive) migration was running every time an OfflineDatabase was constructed with a v2 database.
Fixes#4501
SQLite REPLACE is *not* UPSERT. If a conflict occurs, it first deletes the existing row, then inserts a new row. This means that AUTOINCREMENT primary keys change. This will break foreign keys to that value, which we use.
Instead we must try an UPDATE, and fall back to an INSERT if the UPDATE changes zero rows.
This results in a ~1% increase in database size, which is worth it for reducing complexity and making the tiles and resources tables more similar in structure.
* Under the hood, SQLite creates surrogate keys (ROWID) anyway. We may as well take advantage of this and use the surrogates for foreign keys as well, since they are simpler and more efficient than compound foreign keys.
* Create indexes for efficient eviction queries
When inserting an cached resource, or removing a region, remove least-recently used resources and tiles, not used by offline regions, until the used database size, as calculated by multiplying the number of in-use pages by the page size, is less than the maximum cache size minus 5 times the page size.
In addition, OfflineDatabase may be configured to ignore cache puts of individual resources larger than a certain size.
This policy is similar but not identical to the former SQLiteCache policy:
* It accounts for offline, by exempting resources required by offline regions from eviction.
* It must delete from two tables (resources and tiles), rather than one. Currently the strategy is naive: evict 50 rows at a time from each table.
* It makes maximumCacheSize and maximumCacheEntrySize completely independent. The SQLiteCache implementation evicted when `usedSize > maximumCacheSize - 2 * maximumCacheEntrySize`. This evicts when `usedSize > maximumCacheSize - 5 * pageSize`.
* It uses a non-unlimited default value for maximumCacheSize: 50 MB. We should have always had a limit in place; "a cache without an eviction policy is a resource leak".
