I'm having trouble figuring out the proper way to do atomic commits.  My
intuition is that (for autocommit=False), everything passed to the
execute() method between calls to commit() (on a single given Cursor) gets
batched into a single atomic transaction.  Is that right, or should I be
trying to execute() everything in a big SQL statement?  Thanks.

K.

You are correct - by default connections have autocommit set to False, so
everything you do is batched up, so to speak, until you call
Connection.commit (or Cursor.commit in recent source builds).

This makes coding very handy since you put the commit at the bottom of a
function.  If an exception occurs and the connection is closed (which
happens automatically if you don't keep references to it), the transaction
is rolled back.