native-profiler-stop can't find the session created by native-profiler-start (iOS sim)

[csori99]

Hey, ran into this while profiling a React Native app on the iOS simulator and wanted to flag it.

native-profiler-start works fine (xctrace spawns, starts recording, returns a pid and a trace path). But native-profiler-stop always fails with No active native profiling session found. Call native-profiler-start first., so I can never export a trace through the tool.

Dug into it a bit:

• After calling start, the xctrace process stays alive. I watched it from a plain shell (no argent tool calls in between) for 6+ seconds and it kept running and recording.
• Calling stop after that still says "no active session", and the xctrace process gets killed right around then.
• So start and stop don't seem to share the same session state. stop looks like it resolves a fresh/empty session (profilingActive is false) instead of the one start created, and the real session gets disposed (dispose() SIGKILLs the live xctrace).

Only one tool-server.cjs and one argent mcp process running, so it's not a duplicate-instance thing. Happens whether or not there are other tool calls (gesture-tap, screenshot) in between.

Repro:

1. Boot an iOS simulator with an app running.
2. native-profiler-start (succeeds, recording).
3. interact or just wait a few seconds.
4. native-profiler-stop -> "No active native profiling session found".

Expected: stop finds the active session and exports the trace.
Actual: stop never finds it, and the recording gets killed.

Environment:

• argent 0.8.0 (also reproduced on 0.7.1)
• macOS, Xcode 26.4, xctrace 16.0
• iOS 26.4 simulator, React Native app (bridgeless, Hermes)

One side note in case it is related: running xctrace directly with the bundled Argent.tracetemplate records ok but won't finalize/package the trace on stop (it hangs, and the bundle is incomplete so export fails with "Document Missing Template Error"). Not sure if that is the same underlying issue or a separate xctrace/simulator thing on this machine, but the start/stop session mismatch above reproduces on its own.

The react profiler tools work great by the way, this is just the native side.

[csori99]

Did some more digging in the 0.8.0 source in case it helps narrow it down.

The native-profiler session blueprint never emits a terminated event, while react-profiler-session emits one on CDP disconnect. The native session also doesn't declare dependencies, so it shouldn't be torn down by a cascade either. And the registry caches RUNNING instances by URN, so start's session should still be cached and RUNNING when stop resolves the same urn (NativeProfilerSession:<deviceId>).

In practice though, stop resolves what looks like a fresh instance: profilingActive is false, no xctraceProcess, recordingExitedUnexpectedly is false. So it hits the "No active native profiling session found" throw instead of the partial-export branch. Something between start returning and stop running is putting the session node back to IDLE so it gets recreated on the next resolve.

One thing I noticed: native-profiler-start declares the session with services: () => ({ session: nativeProfilerSessionRef(...) }) (per-call), whereas react-profiler holds its session differently, which might be why react survives across calls and native doesn't.

Couldn't pin the exact teardown trigger from the outside, but the contrast with react-profiler-session seemed worth mentioning.

[pFornagiel]

Hi @csori99, thank you for the report and feedback.

When it comes to your digging - the contrast with react-profiler-session is something we are aware of, we have to interact with those mechanisms in different ways because of their characteristics.

We will try to look into how we can reproduce the issue. If you could, we would also appreciate if it was possible for you to provide:

• the logs from the ~/.argent/mcp-calls.log and ~/.argent/tool-server.log captured during the profiling session
• does the failing stop call return immediately, or hang first? (around 30s for example)
• whether it is possible to reproduce the issue with other versions of the simulator, e.g. 26.2

Also, as you mentioned using xtrace, would you be able to run xctrace export --input <path-to-the-.trace-bundle-from-native-profiler-start's-result> --toc , check whether the TOC looks complete and roughly how long it took?

[miguelartazos]

I’m seeing the same behavior on Argent 0.11.0 with Xcode 26.5 / iOS 26.5 Simulator: native-profiler-start succeeds, but native-profiler-stop reports no active session, and direct xctrace export ends with Document Missing Template Error.

Did you ever find a workaround or a reliable way to finalize/export the trace?

[miguelartazos]

I reran a clean repro and captured the details you asked for.

Environment:

• Argent: 0.11.0
• Xcode: 26.5 / build 17F42
• xctrace version 16.0 (17F42)
• Simulator runtime installed here: only iOS 26.5 (23F77), so I cannot test 26.2 on this machine yet.

One app-specific note: auto-detection finds CFBundleExecutable = Guiote, but xctrace only attaches successfully when I pass app_process: "Guiñote+".

Clean ~/.argent/mcp-calls.log slice:

{"ts":"2026-06-12T19:43:11.790Z","event":"tool_called","name":"native-profiler-start","args":{"device_id":"E10173DC-93B8-47F4-8483-9F51A082E492","app_process":"Guiñote+"}}
{"ts":"2026-06-12T19:43:13.579Z","event":"tool_result","name":"native-profiler-start","durationMs":1789,"isError":false,"result":{"status":"recording","pid":43668,"traceFile":"/var/folders/hz/53wmgfkn2k982zyvs37tfkdw0000gn/T/argent-profiler-cwd/native-profiler-20260612-194311.trace"}}
{"ts":"2026-06-12T19:43:27.266Z","event":"tool_called","name":"native-profiler-stop","args":{"device_id":"E10173DC-93B8-47F4-8483-9F51A082E492"}}
{"ts":"2026-06-12T19:44:00.905Z","event":"tool_result","name":"native-profiler-stop","durationMs":33639,"isError":true,"error":"[Tool:native-profiler-stop] No active native profiling session found. Call native-profiler-start first."}

So the failing stop call does not return immediately; it waits about 33.6s.

Clean ~/.argent/tool-server.log slice:

[2026-06-12T19:43:57.281Z] [native-profiler] xctrace did not respond to SIGINT; SIGTERM was used. Trace bundle may be incomplete.
[2026-06-12T19:44:00.904Z] [registry] toolFailed native-profiler-stop:
[Tool:native-profiler-stop] No active native profiling session found. Call native-profiler-start first.
    at stopNativeProfilerIos (/opt/homebrew/lib/node_modules/@swmansion/argent/dist/tool-server.cjs:68057:11)
    at Object.execute (/opt/homebrew/lib/node_modules/@swmansion/argent/dist/tool-server.cjs:69134:25)
    at process.processTicksAndRejections (node:internal/process/task_queues:103:5)
    at async Registry.invokeTool (/opt/homebrew/lib/node_modules/@swmansion/argent/dist/tool-server.cjs:54929:22)
    at async /opt/homebrew/lib/node_modules/@swmansion/argent/dist/tool-server.cjs:56388:22

xctrace export --input <trace> --toc result:

Export failed: Document Missing Template Error
real 0.63
user 0.30
sys 0.23

Trace bundle contents after failure:

native-profiler-20260612-194311.trace/Trace1.run/RunIssues.storedata
native-profiler-20260612-194311.trace/Trace1.run/event_data_15334.oa

So the TOC does not look complete; export fails almost immediately with Document Missing Template Error, although an event_data_<pid>.oa file is present.