* perf(daemon): offload PNG decode/encode and screenshot pixel diff to a worker thread
PNG decode (inflateSync) and per-pixel screenshot diffing previously ran
synchronously on the daemon event loop, stalling all concurrent sessions
while multi-MB screenshots were processed.
- add src/utils/png-worker.ts worker_threads entry (rslib internal/png-worker)
handling one decode, encode, or diff-pixels job per message
- add src/utils/png-worker-client.ts async wrappers (decodePngAsync,
encodePngAsync, computeScreenshotDiffPixelsAsync) that lazily spawn the
worker, resolve it next to the current module in dev (.ts) and dist (.js)
like the companion tunnel entry, and fall back to the in-process
synchronous path when the worker is unavailable
- extract the unchanged pixel-compare loop into
src/utils/screenshot-diff-pixels.ts so both paths share identical logic
- route daemon call sites (screenshot-overlay annotate, compareScreenshots)
through the async wrappers; results stay byte-identical
https://claude.ai/code/session_01LXZXzxi55sZ11DSyqWyBA2
* refactor(daemon): apply review findings to PNG worker offload
- guard postMessage failures: a job-specific send error (e.g. DataCloneError)
now cleans up its pending entry and falls back to sync for that call only
- resolve the worker entry via a shared src/utils/internal-entry.ts helper
that returns null on any resolution failure (non-file import.meta.url)
- match repo precedent: spawn the worker with --experimental-strip-types
when the resolved entry is a .ts module
- report permanent worker degradation once (scoped diagnostic + process
warning with the failure reason) instead of silently going sync-only
- daemon lifecycle: pre-warm the worker at startup and terminate it during
shutdown with a 1s best-effort timeout (daemon-only)
- collapse the three async wrappers onto one kind-typed job runner with a
single unavailability channel (rejection, no null path) and drop the dead
mismatched-result guards
- derive the diff-pixels contract types from screenshot-diff-pixels.ts and
share toBuffer via the contract module
- serialize worker errors with normalizeError and reconstruct AppError
(code/message/details) client-side; the worker reuses decodePng so decode
failures are identical on both paths
- transfer result buffers back to the client when a view fully owns its
ArrayBuffer; clone pooled buffers to protect Node's shared buffer pool
- decode baseline/current screenshots concurrently in compareScreenshots
- move resizePngFileToMaxSize to src/utils/png-resize.ts and route its
decode/encode through the worker (daemon screenshot --max-size path)
https://claude.ai/code/session_01LXZXzxi55sZ11DSyqWyBA2
* fix(fallow): declare png worker entry and simplify resultTransferList
The Fallow audit flagged src/utils/png-worker.ts as unreachable from any
entry point and resultTransferList as over the complexity threshold.
- Declare src/utils/png-worker.ts in .fallowrc.json's entry list: it is a
worker_threads entry point loaded at runtime (mirrors src/daemon.ts /
src/companion-tunnel.ts and the internal/png-worker rslib entry).
- Flatten resultTransferList into filter/map over an extracted
ownsEntireArrayBuffer predicate, preserving the exact transfer rule
(byteOffset === 0, byteLength === owner.byteLength, real ArrayBuffer).
- Add direct unit coverage asserting fully-owned buffers are transferred
while pooled/offset views are not.
https://claude.ai/code/session_01LXZXzxi55sZ11DSyqWyBA2
---------
Co-authored-by: Claude <noreply@anthropic.com>
* feat: add e2e command perf benchmark harness + nightly CI
Adds scripts/perf, a cheap end-to-end perf benchmark that drives the built
CLI through an ordered Settings tour of ~24 commands for N rounds, on a fully
isolated daemon/state-dir and self-cleaning device, and emits JSON + Markdown
reports. Per-command timing comes from wrapping each batchable command in its
own single-step batch (daemon durationMs) plus wall-clock around the process.
Wires a scheduled + workflow_dispatch CI job (perf-nightly.yml) that reuses the
cached iOS XCUITest runner (setup-apple-replay) and the Android replay host, and
runs the CLI from source via --experimental-strip-types (no dist build).
* refactor(perf): drive the harness CLI via runCmdSync, not spawnSync
Review (P2): repo rule is to spawn processes through src/utils/exec.ts, not
node:child_process directly. Switch the perf harness's invokeCli to runCmdSync
(allowFailure so non-zero exits are recorded as samples) and add a maxBuffer
option to ExecOptions/runCmdSync (snapshot payloads exceed Node's ~1MB default).
* perf(harness): warm the runner after open so the first measured command is clean
The first interaction after open/relaunch pays the one-time iOS XCUITest runner
startup (~10s+ cold) and a per-relaunch first-AX-query settle cost (~4s). That was
landing on the first measured command each round (snapshot -i), inflating it ~10x
vs the next snapshot. Run an untimed warmup snapshot -i after establishSession, after
each round's reset-open, and after every freshRoot relaunch, so no measured command
absorbs runner startup. Noted in the report header.
* refactor(perf): address review + fix Fallow CI
- exec.ts: extract spawnRejectionError + commandCloseFailure helpers, deduping the
error/close handler clones (Fallow duplication ✗ that surfaced once the maxBuffer
change pulled exec.ts into the audit scope).
- .fallowrc: exclude scripts/perf/** (non-shipped benchmark tooling, like examples/
test-app) so its naturally-moderate functions don't trip the complexity gate.
- config.ts: drop unused exports CLI_BIN/DEFAULT_OUT_DIR; add readIntValue so
--n/--rounds/--warmup report the actual flag + reject non-integers clearly.
- harness.ts: extract toSample(); type sampleError param as CliResult.
- scenario.ts: ScenarioStep is now a discriminated union on execMode (removes step.step!/
step.args ?? []).
- comment/legend rewords (platform defaults are local-convenience/CI-overridden;
elements = node count). check:fallow now green; typecheck/lint/unit pass.
* perf(harness): downgrade sample ok when a batch step reports ok:false
Defensive belt-and-suspenders for the Codex review note: stop-only batch already
surfaces a failed step as a top-level failure (caught by invokeCli), but if an
on-error=continue mode ever keeps the batch ok while a step fails, don't silently
count that step as a successful sample — derive ok from the step's own result.ok.
Long-running AgentDeviceRunnerUITests sessions can inherit Xcode 26 screenRecording/deleteOnSuccess defaults and leave hidden testmanagerd attachments behind when sessions terminate outside a clean XCTest success path.
Pin an explicit AgentDeviceRunnerUITests.xctestplan with screenshots and keepNever attachment lifetimes, then normalize the generated per-session .xctestrun copy before test-without-building. The .xctestrun normalization protects stale cached artifacts and Xcode versions that ignore some test-plan attachment lifetime keys during build-for-testing.
This only affects XCTest's automatic attachments; user-requested agent-device recordings still use the existing record command paths.
Co-authored-by: Michał Pierzchała <thymikee@gmail.com>
Michał Pierzchała