[GH-ISSUE #4232] [Bug] LLDB denied launch loses observed exit and retains unknown target ownership #1664

New issue

Closed

opened 2026-10-16 21:08:44 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-16 21:08:44 +00:00

Owner

Originally created by @morluto on GitHub (Oct 15, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/4232

Observed failure

A denied LLDB launch can return a valid, already-exited SBProcess. REA reports cleanup_incomplete with native-target:unknown and retains its owned temporary runtime instead of returning the actual capability denial after confirmed cleanup.

Related audit: #4227. This is distinct from #2397 (identity failure after a running launch) and the closed #3185 (pre-launch rejection). The bridge has a concrete exited process here, rather than no process or an unverified live process.

Reproduction and control

Revision: bde139c5d, REA 6.5.0, macOS 26.3, Node 24.18.0, Apple LLDB lldb-2100.0.17.108 / Swift 6.3.2.

Compile the repository-owned tests/conformance/native/calls.m fixture with xcrun clang.
Copy it to a disposable workspace and ad-hoc sign with hardened runtime and without get-task-allow.
Invoke the public observe-native-calls CLI, or run npm run verify:native-calls.
Observe the LLDB attach denial. Before repair, the verifier expects capability_unavailable but receives cleanup_incomplete; the CLI reports unknown target identity and retains an owned rea-lldb-* runtime.

An independent probe of the same owned fixture observed SBError.Success() == false, SBProcess.IsValid() == true, a nonzero process ID, and GetState() == 10 (eStateExited). The error begins process exited with status -1 (attach failed (Not allowed to attach to process...)).

Root cause

bridge/native/rea_lldb_tracer.py emits target_process_created: true for the valid failed-launch process but publishes its PID only on the success path. src/native/LldbCallTracer.ts correctly refuses to infer exit from a missing identity, so the lost producer facts become retained unknown ownership.

Object validity, historical creation, identity availability, and observed exit are separate facts. A launch error alone must not imply that no process survived; conversely, an exact owned process handle already in eStateExited should not become an unknown live target because launch failed.

Expected repair and validation

Publish observed PID and explicit exit state on failed launch paths, preserving historical process creation.
For a failed launch that remains live, retire only the exact owned process handle under the existing ownership protocol and positively observe exit; preserve uncertainty when proof is unavailable.
Consume explicit exit proof in the provider, rather than falsely changing creation to false or inferring liveness from IsValid().
Preserve capability denial and cleanup diagnostics through CLI and MCP.
Exercise valid-but-exited, valid-and-live, missing identity, and unknown-exit controls.

A local patch passes the real verify:native-calls CLI/stdio MCP lane, including positive observation, duration/event limits, denied hardened-runtime launch, and the get-task-allow control. Python bridge regressions pass; integration and full checks are still in progress. No released fix is claimed.

Upstream contract references

LLDB SBProcess exposes object validity, process ID and state separately. SBTarget Launch returns a process with an error output. Context7 and DeepWiki were consulted for these distinctions; the live probe above is the runtime evidence.

Originally created by @morluto on GitHub (Oct 15, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/4232 ### Observed failure A denied LLDB launch can return a valid, already-exited SBProcess. REA reports `cleanup_incomplete` with `native-target:unknown` and retains its owned temporary runtime instead of returning the actual capability denial after confirmed cleanup. Related audit: #4227. This is distinct from #2397 (identity failure after a running launch) and the closed #3185 (pre-launch rejection). The bridge has a concrete exited process here, rather than no process or an unverified live process. ### Reproduction and control Revision: `bde139c5d`, REA 6.5.0, macOS 26.3, Node 24.18.0, Apple LLDB `lldb-2100.0.17.108` / Swift 6.3.2. 1. Compile the repository-owned `tests/conformance/native/calls.m` fixture with `xcrun clang`. 2. Copy it to a disposable workspace and ad-hoc sign with hardened runtime and without get-task-allow. 3. Invoke the public `observe-native-calls` CLI, or run `npm run verify:native-calls`. 4. Observe the LLDB attach denial. Before repair, the verifier expects `capability_unavailable` but receives `cleanup_incomplete`; the CLI reports unknown target identity and retains an owned `rea-lldb-*` runtime. An independent probe of the same owned fixture observed `SBError.Success() == false`, `SBProcess.IsValid() == true`, a nonzero process ID, and `GetState() == 10` (`eStateExited`). The error begins `process exited with status -1 (attach failed (Not allowed to attach to process...))`. ### Root cause `bridge/native/rea_lldb_tracer.py` emits `target_process_created: true` for the valid failed-launch process but publishes its PID only on the success path. `src/native/LldbCallTracer.ts` correctly refuses to infer exit from a missing identity, so the lost producer facts become retained unknown ownership. Object validity, historical creation, identity availability, and observed exit are separate facts. A launch error alone must not imply that no process survived; conversely, an exact owned process handle already in `eStateExited` should not become an unknown live target because launch failed. ### Expected repair and validation - Publish observed PID and explicit exit state on failed launch paths, preserving historical process creation. - For a failed launch that remains live, retire only the exact owned process handle under the existing ownership protocol and positively observe exit; preserve uncertainty when proof is unavailable. - Consume explicit exit proof in the provider, rather than falsely changing creation to false or inferring liveness from `IsValid()`. - Preserve capability denial and cleanup diagnostics through CLI and MCP. - Exercise valid-but-exited, valid-and-live, missing identity, and unknown-exit controls. A local patch passes the real `verify:native-calls` CLI/stdio MCP lane, including positive observation, duration/event limits, denied hardened-runtime launch, and the get-task-allow control. Python bridge regressions pass; integration and full checks are still in progress. No released fix is claimed. ### Upstream contract references [LLDB SBProcess](https://lldb.llvm.org/python_reference/lldb.SBProcess-class.html) exposes object validity, process ID and state separately. [SBTarget Launch](https://lldb.llvm.org/python_reference/lldb.SBTarget-class.html) returns a process with an error output. Context7 and DeepWiki were consulted for these distinctions; the live probe above is the runtime evidence.

bsakaguchi closed this issue

2026-10-16 21:08:44 +00:00

bsakaguchi commented

2026-10-16 21:08:46 +00:00

Author

Owner

@morluto commented on GitHub (Oct 15, 2026):

Local repair passed the real CLI/stdio MCP native-call verifier on Apple LLDB lldb-2100.0.17.108: normal exit, duration/event limits, denied hardened-runtime launch and the get-task-allow control. The bridge retains historical process creation, publishes the observed PID, and supplies explicit positive exit proof; the provider consumes that proof before attempting generic OS termination recovery. A failed Kill without an observed exit remains unresolved.

Bridge producer tests passed (13), including already-exited, live failed-launch retirement and failed-Kill/unknown-exit controls. LLDB boundary tests passed (10), including missing-identity negative controls and finalization cancellation. The combined lifecycle/schema regression run passed 114 tests across 14 files, and the full repository check passed. No released or globally installed fix is claimed: patches remain local and uncommitted.

@morluto commented on GitHub (Oct 15, 2026): Local repair passed the real CLI/stdio MCP native-call verifier on Apple LLDB `lldb-2100.0.17.108`: normal exit, duration/event limits, denied hardened-runtime launch and the get-task-allow control. The bridge retains historical process creation, publishes the observed PID, and supplies explicit positive exit proof; the provider consumes that proof before attempting generic OS termination recovery. A failed Kill without an observed exit remains unresolved. Bridge producer tests passed (13), including already-exited, live failed-launch retirement and failed-Kill/unknown-exit controls. LLDB boundary tests passed (10), including missing-identity negative controls and finalization cancellation. The combined lifecycle/schema regression run passed 114 tests across 14 files, and the full repository check passed. No released or globally installed fix is claimed: patches remain local and uncommitted.

bsakaguchi commented

2026-10-16 21:08:46 +00:00

Author

Owner

@official-rex commented on GitHub (Oct 15, 2026):

I can review and finalize this fix. Kindly assign me this issue; I look forward to your response.

@official-rex commented on GitHub (Oct 15, 2026): I can review and finalize this fix. Kindly assign me this issue; I look forward to your response.

bsakaguchi referenced this issue

2026-10-17 21:01:32 +00:00

[GH-ISSUE #2383] [Tracker] Audit analysis and debugging coverage for the TIOBE top 100 languages (October 2026) #794

bsakaguchi referenced this issue

2026-10-17 21:12:35 +00:00

[PR #1583] [MERGED] feat(process): bound a finalization interval before forced capture timeout #2988

bsakaguchi referenced this issue

2026-10-31 00:54:30 +00:00

[PR #1664] [MERGED] fix(javascript): exclude overwritten Electron API identities #4735