I just checked our Android code and we are reusing compiled statements in our benchmarks (Inserts only since compiled statements are not available for multi-column queries, afawk). Here’s the code we use for inserts:
public void testBatchInserts() throws PerformanceTestException {
SQLiteStatement stmt = db.compileStatement("INSERT INTO "
+ EmployeeDatabaseHelper.TABLE_EMPLOYEES + " VALUES(?1, ?2, ?3, ?4)");
db.beginTransaction();
for (int row = 0; row < getNumInserts(); row++) {
stmt.clearBindings();
stmt.bindString(2, getEmployeeName(row));
stmt.bindLong(3, getEmployeeAge(row));
stmt.bindLong(4, getEmployeeHiredStatus(row));
stmt.executeInsert();
}
db.setTransactionSuccessful();
db.endTransaction();
}
So there you go: no intention to mislead. Us overlooking prepared statements when we originally benched iOS was definitely a regrettable error but we do try to improve. Thanks for helping us along the way.Would you have the source code for the benchmarks available somewhere? On the iOS post there was a link, but I didn't see it on the android one.
I haven’t looked too deep personally but I don’t think the speedup comes from anything new we’re doing in the C++ layer. It’s more likely coming from the fact that we tried to find smart ways around Java & the JVM to avoid the kind of copying or hijacking other ORMs do. We did hit some of that stuff again yesterday/today as we just added migrations into the mix but we have some theories on how to get some of that speed back.
EDIT: typos.
EDIT 2: clearly referencing that this message was about iOS and pointing to the other comment for Android.
Thanks again for looking into it and updating the material - far too many groups just gloss over (glaring) problems with their micro benchmarks, and it just sours all their future claims. Glad you're not one of them :)
I'm still very much interested in Realm, and LOVE having a competitor to SQLite. But don't mislead your customers.
[1]: https://news.ycombinator.com/item?id=8043332 where seepel finds SQLite to be more than twice as fast as Realm after that one change to the benchmark.