[Jackmeson1]

What was changed

Adds clock_gettime and clock_gettime_ns to the default workflow sandbox's runtime-only time restrictions. Calling either through the sandbox now raises RestrictedWorkflowAccessError; importing or referencing them and calling them during workflow-module import remain allowed.

Why?

The existing restrictions block clock readers such as time, monotonic and process_time, but omit these two functions. On 01af1f6, both can read CLOCK_REALTIME from a sandboxed workflow without raising.

This tightens the default restrictions: with default failure handling, an uncaught violation fails the workflow task. Only these two names are added; argument-dependent time functions remain outside this change. The sandbox remains best effort, as documented in the README. No replay divergence was reproduced.

Checklist

1. Related: none. This is a narrow fix to the sandbox restriction list.

2. How was this tested:

   Outside-sandbox validation on Linux x86_64 (Colab), using the Rust bridge built from this checkout and pytest 9.1.1:

   • The importer test covers time, monotonic, clock_gettime and clock_gettime_ns. Each call succeeds before the runtime transition and then must raise RestrictedWorkflowAccessError with the expected qualified name. Missing functions are skipped.
   • The existing worker-backed restriction test gets both CLOCK_REALTIME calls, guarded by platform availability. Its workflow converts restriction errors to ApplicationError so the test can assert the resulting workflow failure.
   • Against 01af1f6, both new clock cases fail with DID NOT RAISE RestrictedWorkflowAccessError, and the worker test fails with DID NOT RAISE WorkflowFailureError; the two existing-clock controls pass. Adding only one name leaves two failures.
   • I ran pytest tests/worker/workflow_sandbox: 35 passed, 1 skipped on 01af1f6; 39 passed, 1 skipped with the fix on Python 3.13.16 and 3.14.8; 40 passed on 3.10.22. I also ran time-skipping V2 on 3.13.16: 39 passed, 1 skipped.
   • I ran the full suite, pytest -n 4 --dist=worksteal, on Python 3.13.16: 1549 passed, 5 skipped on 01af1f6; 1553 passed, 5 skipped with the fix. The saved JUnit reports show four added cases and no changed statuses among existing tests.
   • I ran poe lint, which passed with ruff 0.15.22, pyright 1.1.403, mypy 1.18.2, basedpyright 1.34.0 and pydocstyle 6.3.0.
   • Windows/macOS target checks passed on the three changed Python files: pyright and basedpyright for Windows Python 3.10/3.14 and Darwin 3.14; mypy for win32 and darwin. These checks ran on Linux; Windows, macOS and ARM execution were not tested.

3. Any docs updates needed?

   Added changelog/breaking-changes/punctual-platypus.md. The README documents the general restriction behavior and best-effort limitation.

AI disclosure: this was prepared with AI assistance (Claude Opus 5.5).

🤖 Generated with Claude Code