test: keep ACP process-tree fake PIDs from aliasing the worker PID

[rynfar]

Summary

Fixes the recurring Test Server 2 failure in AcpSessionRuntime.processTree.test.ts > terminatePosixOwnedProcessTree > rotates more than 64 live parents without scanning retained tombstones (ACP process tree 100 retained owned processes: 1088, 11088, and the same error for 1001/11001 and 1047/11047).

Root cause

This was a fixture PID collision. It was not a timing problem, and no merged change introduced a production regression.

• The test seeds the fake controller with server() at the real process.pid and adds 130 fake parents at PIDs 1000–1129 plus children at +10000 in the same Map keyed by PID.
• On fresh ubuntu-24.04 runners the vitest worker's real PID often falls in 1000–1129. When it does, the fake parent replaces the server entry. controller.identity(process.pid) then returns that parent, so current.pgid/current.sid become the parent's group.
• terminatePosixOwnedProcessTree intentionally never signals process.pid or anything in the current process group or session. It skips that parent and its child, and the final residual check reports exactly N, N+10000. The production behavior is correct.
• The failing PIDs changed from run to run because the worker PID depends on how many processes started before it on the runner. Changes in shard and job composition move that number, which is why the failures started after unrelated merges.

it.live and the 10 ms sleeps don't matter here. Because grace: 0, the outcome depends only on the PID.

Evidence

Reproduced by setting process.pid in a temporary copy of the file (removed before committing):

forced process.pid	before	after
unchanged / 1130	pass	pass
1001	retained owned processes: 1001, 11001	pass
1047	retained owned processes: 1047, 11047	pass
1088	retained owned processes: 1088, 11088	pass
11088	retained owned processes: 1088, 11088	pass
1000, 1129, 11000, 11129	(in collision range)	pass

The before messages match the CI errors exactly.

Fix

The test now allocates its synthetic parent, child, and tombstone PIDs above Linux PID_MAX_LIMIT (2^22; macOS PIDs stay below 100000), so they can never alias a real worker PID. Every assertion is unchanged, including snapshotCalls === 10, fixture.processes.size === 1, and the ≤64 work bound per pass. No production code changed.

Validation

• vp test run src/provider/acp/AcpSessionRuntime.processTree.test.ts ×50: 50 passed, 0 failed
• AcpSessionRuntime.test.ts + AcpSessionRuntime.processTree.test.ts: 37/37 passed
• vp run -F t3 typecheck: clean; scoped vp lint / vp fmt --check: clean

Impact: test-only. No change to runtime, providers, wire contracts, or clients.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ The exact PR base did not have a successful artifact. Baseline uses the latest successful main measurement shown below.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	4.9 KiB	4.9 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.7 KiB	3.7 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.4 KiB	20.4 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	4.9 KiB	4.9 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.7 KiB	3.7 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	20.8 KiB	20.8 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: cda6633 · PR result: 6387524 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.