pysqlite and APSW approached the problem of providing access to SQLite from Python from fundamentally different directions.
APSW only wraps version 3 of SQLite and provides access in whatever way is normal for SQLite. It makes no effort to hide how SQLite is different from other databases.
pysqlite tries to provide a DBAPI compliant wrapper for SQLite and in doing so needs to make it have the same behaviour as other databases. Consequently it does hide some of SQLite’s nuances.
I suggest using APSW when you want to directly use SQLite and its functionality or are using your own code to deal with database independence rather than DBAPI. Use pysqlite and DBAPI if your needs are simple, and you don’t want to use SQLite features.
APSW has the following enhancements/differences over pysqlite 2 (wrapping SQLite 3):
APSW stays up to date with SQLite. As features are added and functionality changed in SQLite, APSW tracks them.
You can use the same Connection across threads with APSW without needing any additional level of locking. pysqlite requires that the Connection and any cursors are used in the same thread. You can disable its checking, but unless you are very careful with your own mutexes you will have a crash or a deadlock.
APSW is a single file for the extension, apsw.pyd on Windows and apsw.so on Unix/Mac (Note PEP 3149). There are no other files needed and the build instructions show you how to include SQLite statically in this file. You can put this file anywhere your Python session can reach. pysqlite is one binary file and several .py files, all of which need to be available.
Nothing happens behind your back. By default pysqlite tries to manage transactions by parsing your SQL for you, but you can turn it off. This can result in very unexpected behaviour with pysqlite.
When using a Connection as a context manager APSW uses SQLite’s ability to have nested transactions. pysqlite only deals with one transaction at a time and cannot nest them. (Savepoints were introduced in SQLite 3.6.8 - another illustration of the benefits of keeping up to date with SQLite.)
APSW always handles Unicode correctly (this was one of the major reasons for writing it in the first place). pysqlite has since fixed many of its issues but you are still stuck with some.
You can use semi-colons at the end of commands and you can have multiple commands in the execute string in APSW. There are no restrictions on the type of commands used. For example this will work fine in APSW but is not allowed in pysqlite:
import apsw con=apsw.Connection(":memory:") cur=con.cursor() for row in cur.execute("create table foo(x,y,z);insert into foo values (?,?,?);" "insert into foo values(?,?,?);select * from foo;drop table foo;" "create table bar(x,y);insert into bar values(?,?);" "insert into bar values(?,?);select * from bar;", (1,2,3,4,5,6,7,8,9,10)): print row
And the output as you would expect:
(1, 2, 3) (4, 5, 6) (7, 8) (9, 10)
Cursor.executemany() also works with statements that return data such as selects, and you can have multiple statements. pysqlite’s executescript() method doesn’t allow any form of data being returned (it silently ignores any returned data).
pysqlite swallows exceptions in your callbacks making it far harder to debug problems. That also prevents you from raising exceptions in your callbacks to be handled in your code that called SQLite. pysqlite does let you turn on printing of tracebacks, but that is a poor substitute. apsw does the right thing as demonstrated by this example.
def badfunc(t): return 1/0 # pysqlite from pysqlite2 import dbapi2 as sqlite con = sqlite.connect(":memory:") con.create_function("badfunc", 1, badfunc) cur = con.cursor() cur.execute("select badfunc(3)") # apsw import apsw con = apsw.Connection(":memory:") con.createscalarfunction("badfunc", badfunc, 1) cur = con.cursor() cur.execute("select badfunc(3)")
# pysqlite Traceback (most recent call last): File "func.py", line 8, in ? cur.execute("select badfunc(3)") pysqlite2.dbapi2.OperationalError: user-defined function raised exception # apsw Traceback (most recent call last): File "t.py", line 8, in ? cur.execute("select badfunc(3)") File "apsw.c", line 3660, in resetcursor File "apsw.c", line 1871, in user-defined-scalar-badfunc File "t.py", line 3, in badfunc return 1/0
APSW has significantly enhanced debuggability. More details are available than just what is printed out when exceptions happen like above. See augmented stack traces
APSW has execution and row tracers. pysqlite has no equivalent to execution tracers and does have data adaptors which aren’t the same thing as a row tracer (for example you can’t skip rows or add a new column to each row returned). pysqlite does have a row factory but you can easily emulate that with the row tracer and Cursor.getdescription().
APSW has an apswtrace utility script that traces execution and results in your code without having to modify it in any way. It also outputs summary reports making it easy to see what your most time consuming queries are, which are most popular etc.
APSW has an exception corresponding to each SQLite error code and provides the extended error code. pysqlite combines several SQLite error codes into corresponding DBAPI exceptions. This is a good example of the difference in approach of the two wrappers.
The APSW test suite is larger and tests more functionality. Code coverage by the test suite is 99.6%. pysqlite is good at 81% for C code although there are several places that coverage can be improved. I haven’t measured code coverage for pysqlite’s Python code. The consequences of this are that APSW catches issues earlier and gives far better diagnostics. As an example try returning an unsupported type from a registered scalar function.
APSW is faster than pysqlite in my testing. Try the speedtest benchmark.
pysqlite has an adaptor system that lets you pretend SQLite stores and returns more types than it really supports. Note that the database won’t be useful in a non-pysqlite context (eg PHP code looking at the same database isn’t going to recognise your Point class). You can implement something similar in APSW by intercepting Cursor.execute() calls that suitably mangles the bindings going to SQLite and does something similar to the rows the iterator returns.
pysqlite lets you work with a database that contains invalid Unicode data by setting a text factory that deals with the text data.
APSW does not let you put non-Unicode data into the database in the first place and it will be considered invalid by other tools reading the data (eg Java, PHP). If you somehow do manage to get non-Unicode data as a SQLite string, you can cast it to a blob:
for row in cursor.execute("select CAST(column as BLOB) from table"): # row is buffer (py2) or bytes (py3) here deal_with_binary_data(row)