There are some issues with the error handling in pysqlite_connection_backup_impl():

1. ValueError is returned if the target connection equals source connection. Should be OperationalError, IMHO.

2. The aforementioned check is already performed by sqlite3_backup_init(), so we should just let SQLite take care of it and let _pysqlite_seterror() set the error if sqlite3_backup_init() returns NULL. This will also take care of 1.

3. The following comment seems to be wrong; errors are set on the connection object, not on the backup handle:
   /* We cannot use _pysqlite_seterror() here because the backup APIs do
   not set the error status on the connection object, but rather on
   the backup handle. */

After sqlite3_backup_finish(), we can just check the return code, and call _pysqlite_seterror() on the connection and return NULL. The mentioned comment can be removed.

Resolving these issues will save 18 lines of code, and make the backup function easier to maintain.

Berker?

I'm not sure about 1) because if target == source it means a user error. OperationalError is usually used for non-user errors.

3. sounds good to me if that part is already covered by tests and they passed.

I'm not sure about 1) because if target == source it means a user error. OperationalError is usually used for non-user errors.

Yes, my bad. ProgrammingError would be better.
target == source results in SQLITE_ERROR, BTW.

I'll throw up a PR for 3., then.

The unit test suite shows one case of improved "exception text". I'd say it's an improvement.

$ ./python.exe  # with python/cpython#68774 applied
>>> import sqlite3
>>> c1 = sqlite3.connect(":memory:")
>>> c2 = sqlite3.connect(":memory:")
>>> c1.backup(c2, name="non-existing")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
sqlite3.OperationalError: unknown database non-existing

$ python3.10  # latest alpha from python.org
>>> import sqlite3
>>> c1 = sqlite3.connect(":memory:")
>>> c2 = sqlite3.connect(":memory:")
>>> c1.backup(c2, name="non-existing")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
sqlite3.OperationalError: SQL logic error

I'd even consider a cryptic error message a minor usability bug and backport it to at least 3.10.

3.9, I presume? Yeah, I guess so.

Regarding 1: After thinking about it, I see nothing wrong with raising ValueError error. No need to change current behaviour.

3.9, I presume?

Yes, I was going too fast :)

New changeset c1ae741 by Erlend Egeberg Aasland in branch 'master':
bpo-43265: Improve sqlite3.Connection.backup error handling (GH-24586)
c1ae741

Thank you.