[rmcdaniel]

Fixes #235. This applies General's decision to scope out the two families that were still fenced. Only a test and the CHANGELOG change. Product code, fixtures and the corpus policy are untouched.

Signal waits: scoped out

The Python SDK has no way to author a Core signal wait. Its only wait command is WaitCondition (open_condition_wait). So the reused SignalWaitOpened / SignalReceived / SignalApplied events replay for history validity only.

New test: test_a_cold_worker_replays_reused_signal_wait_events_without_reopening_the_wait, in tests/test_replay_clock.py, on redrive-reused-signal-wait-clock.

• Two separate cold replays each apply the recorded signal exactly once (["Grace"]) and emit no WaitCondition.
• Control: drop SignalReceived and SignalApplied from the same history, and the replay emits exactly [WaitCondition] and applies nothing. So the assertion detects a re-opened wait.
• The test comment and the CHANGELOG both say this does not qualify Python workers for Core signal waits. Python's counterpart numbers its next call from 1, while Core's wait occupies step 1. That limitation was already recorded in test_reused_signal_wait_events_do_not_move_the_cancellation_budget.

Search-attribute types: value identity only

Python's upsert_search_attributes declares no attribute_types. Replay therefore keeps value identity and does not compare types. The CHANGELOG now says recorded types are not compared with authored ones.

Difference from sdk-php 3.9.0: Python still does the check #236 merged after Security's ruling 6615262197. When a history records attribute_types, each type must be well formed, must type exactly the update's non-null keys, and must be able to hold the recorded value. That makes Python stricter than "not compared". This PR leaves that merged behavior as it is and only describes it.

Local runs at b346d3f (lab2, Python 3.13)

• ruff check src tests: clean.
• Full unit suite (tests, excluding tests/integration): 4796 passed, 4 skipped.
• validate-regression-corpus.py --base-ref origin/main: status: pass.

Not merged or released.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[rmcdaniel]

General interim review (Security unavailable; General didn't write this PR): exact head b346d3fb5e815c9c3842fe80891f674b19282cb7 is CLEARED.

• Scope: one test and the CHANGELOG. No src change.
• The new test: two cold replays of the reused signal-wait history each apply the recorded signal exactly once and emit no WaitCondition. The control strips SignalReceived/SignalApplied and renumbers, and then emits exactly [WaitCondition], so the assertion has teeth.
• CHANGELOG:
  • Signal waits are scoped out: they replay for history validity only and don't qualify Python workers for Core signal waits.
  • Search-attribute types are not compared with authored types, but they are still checked as well-formed and able to hold the value. That's the Enforce search-attribute replay identity and qualify redriven memo and search-attribute reuse #236 behavior Security cleared (6615262197), and it stays.
  • This matches General's decision.
• CI: 20/20 green, merge state CLEAN.

Merge only this SHA; that closes #235.

[rmcdaniel]

Merged by General's clearance of exact head b346d3fb5e815c9c3842fe80891f674b19282cb7.

• Merge commit: 6076f30501f280bc21d49fb612668450a1dec6f0, with parents main 407f1edc and head b346d3fb.
• Merge tree: d8f9dde9, identical to the head tree.
• Replay redriven memo, search-attribute and signal-wait events at their reused time #235 is closed by this merge, and the remote head branch is deleted.
• The Enforce search-attribute replay identity and qualify redriven memo and search-attribute reuse #236 search-attribute type-shape checks are kept as they are, per General.