In unit testing with wrapture the tests leaned on a timeline and a tape to assert on what happened, and I skipped over what those actually are. This post is about the recording side of wrapture: what gets recorded, what one event holds, how a test reads the record back, and the whole-tape views that answer questions about the flow between calls rather than about any one of them.
The example is a resource leak, because it is the kind of bug the recording model was made for. Code that acquires a connection has to release it on every path out: the normal return, the early return, and the exception. The path that forgets is the one nobody looks at, and it does not fail. Nothing raises, nothing returns the wrong value, the test passes, and the pool runs dry a week later in production. The failure is an absence, and asserting on an absence needs a record of what did happen, on the real objects, including objects minted mid-call that the test never held.
The code under test
A stand-in for any pooled resource. Database.connect() mints a Connection, and a connection answers queries until close() sets its closed flag:
class Connection:
