* Record total_rows for each column
* Record (value, count) if there is just a single distinct value
* Do not calculate most/least common if all values are distinct
* Calculate table count once per table, not once per column
Takes my test down from ten minutes to four seconds!
* Removed unnecessary update() optimization
* Added column_order= to .transform() and .transform_sql()
* Tests for reusing lookup table in extract()
Closes#172
* Pass a tracer= function to Database constructor
* New db.tracer() contextmanager
* Neater SQL indentation, because tracer means it could be visible now
* New db.execute() and db.executescript() methods
Closes#150
Deletes all rows in the table (if it exists) before inserting new rows.
SQLite doesn't implement a TRUNCATE TABLE statement but does optimize an
unqualified DELETE FROM.
This can be handy if you want to refresh the entire contents of a table
but a) don't have a PK (so can't use --replace), b) don't want the table
to disappear (even briefly) for other connections, and c) have to handle
records that used to exist being deleted.
Ideally the replacement of rows would appear instantaneous to other
connections by putting the DELETE + INSERT in a transaction, but this is
very difficult without breaking other code as the current transaction
handling is inconsistent and non-systematic. There exists the
possibility for the DELETE to succeed but the INSERT to fail, leaving an
empty table. This is not much worse, however, than the current
possibility of one chunked INSERT succeeding and being committed while
the next chunked INSERT fails, leaving a partially complete operation.
* Failing test showing that DML in `sqlite-utils query` doesn't work
* Run `sqlite-utils query` in a transaction so that DML is committed
Thanks, @tsibley!
