Abort test requires a sandbox-independent execution fixture
Open 💬 1 comment Opened Aug 20, 2026 by rebroad
What is happening?
The focused codex-core abort test can fail when the test runner cannot create its sandbox namespace. In that case, the shell process never reaches the interruption path; the test records the sandbox startup error instead of an aborted command.
Expected behavior
The test should exercise interruption and verify the elapsed-time output independently of whether the runner provides the default sandbox implementation.
Verification
Commit 9ec70ca31c adds a test-only full-access execution fixture for the abort tests. The existing elapsed-time assertion then passes without changing production behavior or the model tool-call payload. An upstream implementation should provide an explicit execution fixture or equivalent test isolation.
1 Comment
Correction: the original failure was caused by the test runner being unable to create the sandbox namespace, not by the runtime selecting the wrong aborted-output format. The shell never reached the abort path.
Commit
9ec70ca31cadds a test-only full-access execution fixture; the existing elapsed-time assertion then passes without changing production behavior or the tool-call payload. I am leaving this issue open pending upstream review of whether the test should use an explicit execution fixture.