refactor(ios): point xcodebuild at the scoped simulator set and delete the XCTestDevices redirect - #2963
Conversation
…on, not a redirect xcodebuild resolves a simulator in a custom CoreSimulator set through the DVTSimulatorSetLocation Xcode user default, passed as -DVTSimulatorSetLocation=<set>. Every runner xcodebuild phase (build-for-testing and test-without-building) now names the scoped set beside its -destination, so the runner never needs ~/Library/Developer/XCTestDevices. Xcode's first-launch cleanup deletes every device in XCTestDevices, and while it was a symlink into the user's set that cleanup deleted the user's simulators (#2935). Deleted: the XCTestDevices rename/symlink/reconcile machinery, the host-global xctest-device-set.lock, the session redirect handle and its releases, the simulator_set_redirect handoff refusal and the scoped-set adoption refusal. A simulator runner start puts back a symlink or XCTestDevices.agent-device-backup an older version left behind. A destination xcodebuild cannot find for a scoped set fails with details.reason simulator_set_destination_not_found, naming details.simulatorSetPath and details.xcodeVersion, in both xcodebuild phases. Closes #2935
close stopped the runner of a simulator in a scoped set only because that runner held the XCTestDevices redirect and its host-global lock for its lifetime (#1013). The redirect is gone, so a scoped-set runner is retained under idle-stop like a default-set one.
…t behind The legacy cleanup unlinked any XCTestDevices symlink on every simulator startup, so a host that links XCTestDevices to another volume lost the link on a default-set open. A symlink is now removed only when the agent-device backup sits beside it, or when it points at this runner's own scoped simulator set.
|
Size Report
Startup median (7 runs, lower is better):
|
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
3 issues found across 31 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="packages/platform-apple/src/runner/runner-device-set.ts">
<violation number="1" location="packages/platform-apple/src/runner/runner-device-set.ts:37">
P1: This migration misses legacy backup names that the previous redirect code explicitly supported. A host left with `.agent-device-xctestdevices-backup-*` can remain redirected, leaving the old `XCTestDevices` data-loss risk in place; scan and restore those legacy backups too.</violation>
</file>
<file name="packages/platform-apple/src/runner/runner-session.ts">
<violation number="1" location="packages/platform-apple/src/runner/runner-session.ts:257">
P2: Adopted legacy simulator runners skip this cleanup because adoption returns before this line. Move the cleanup before adoption, or make adoption perform it, so stale redirect state is removed during handoff.</violation>
</file>
<file name="src/platform-runtime-resource-cleanup.ts">
<violation number="1" location="src/platform-runtime-resource-cleanup.ts:88">
P2: Re-enabling warm retention for scoped-set simulators lets a later `open` reuse a runner whose build is bound to the previous simulator set. `ensureRunnerSession`/`resolveReusableRunnerSession` (packages/platform-apple/src/runner/runner-session.ts:150-158, 391+) reuse the retained session keyed only by `device.id` and never compare the set, and warm reuse skips the xcodebuild phase that would pass `-DVTSimulatorSetLocation`. If the same UDID is later resolved in a different set (a set cloned from the default set preserving UDIDs, or a copied device), the retained runner silently services the device with an artifact built for the old set — no `simulator_set_destination_not_found` error, no rebuild. Validate the set before reuse (e.g., compare `existing.device.simulatorSetPath` against the requested set and stop+rebuild on mismatch).</violation>
</file>
Tip: instead of fixing issues one by one fix them all with cubic
Re-trigger cubic
| device: DeviceInfo, | ||
| xctestDeviceSetPath: string = path.join(os.homedir(), 'Library', 'Developer', 'XCTestDevices'), | ||
| ): void { | ||
| const backupPath = `${xctestDeviceSetPath}.agent-device-backup`; |
There was a problem hiding this comment.
P1: This migration misses legacy backup names that the previous redirect code explicitly supported. A host left with .agent-device-xctestdevices-backup-* can remain redirected, leaving the old XCTestDevices data-loss risk in place; scan and restore those legacy backups too.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/platform-apple/src/runner/runner-device-set.ts, line 37:
<comment>This migration misses legacy backup names that the previous redirect code explicitly supported. A host left with `.agent-device-xctestdevices-backup-*` can remain redirected, leaving the old `XCTestDevices` data-loss risk in place; scan and restore those legacy backups too.</comment>
<file context>
@@ -1,392 +1,66 @@
+ device: DeviceInfo,
+ xctestDeviceSetPath: string = path.join(os.homedir(), 'Library', 'Developer', 'XCTestDevices'),
+): void {
+ const backupPath = `${xctestDeviceSetPath}.agent-device-backup`;
+ const backupExists = fs.existsSync(backupPath);
+ const isSymlink =
</file context>
There was a problem hiding this comment.
Declined, with evidence. No commit on main ever wrote a .agent-device-xctestdevices-backup-* name: git log --oneline -S "agent-device-xctestdevices-backup" origin/main -- packages src returns only #354 (the original redirect), #1062 (a module split) and #2598 (a lock change). In #354 itself the only writer used ${XCTestDevices}.agent-device-backup (no -* suffix); the -backup-* form was only ever read, referring to a pre-squash commit on the #354 branch that never landed on main. No released agent-device version can leave a host in that state, so there's nothing for the daemon-startup scan to find.
| // an external xctestrun that never launches are different steps, and a caller told "developer disk | ||
| // image" should not have to know which one this run happened to take. | ||
| try { | ||
| if (device.kind === 'simulator') restoreLegacyXctestDeviceSetRedirect(device); |
There was a problem hiding this comment.
P2: Adopted legacy simulator runners skip this cleanup because adoption returns before this line. Move the cleanup before adoption, or make adoption perform it, so stale redirect state is removed during handoff.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/platform-apple/src/runner/runner-session.ts, line 257:
<comment>Adopted legacy simulator runners skip this cleanup because adoption returns before this line. Move the cleanup before adoption, or make adoption perform it, so stale redirect state is removed during handoff.</comment>
<file context>
@@ -247,16 +247,14 @@ async function startRunnerSessionWithLease(
// an external xctestrun that never launches are different steps, and a caller told "developer disk
// image" should not have to know which one this run happened to take.
try {
+ if (device.kind === 'simulator') restoreLegacyXctestDeviceSetRedirect(device);
xctestrunArtifact = await measureRunnerStartupStep(
startupTimings,
</file context>
There was a problem hiding this comment.
Originally fixed at 3dd240c by moving the restore ahead of tryAdoptRunnerSessionFromLease in startRunnerSessionWithLease, so one call covered both routes.
Superseded at 9e623ba: the restore left the runner-start path entirely and moved to a one-time daemon-startup step (see the top-level reply, design question). Neither adoption nor a build can skip a step that no longer runs on that path, so this is moot rather than fixed-in-place. The commit that added the adoption-ordering call and its test deleted both.
| !params.hasScreenRecording && | ||
| !params.hasLease && | ||
| !params.device.simulatorSetPath | ||
| !params.hasLease |
There was a problem hiding this comment.
P2: Re-enabling warm retention for scoped-set simulators lets a later open reuse a runner whose build is bound to the previous simulator set. ensureRunnerSession/resolveReusableRunnerSession (packages/platform-apple/src/runner/runner-session.ts:150-158, 391+) reuse the retained session keyed only by device.id and never compare the set, and warm reuse skips the xcodebuild phase that would pass -DVTSimulatorSetLocation. If the same UDID is later resolved in a different set (a set cloned from the default set preserving UDIDs, or a copied device), the retained runner silently services the device with an artifact built for the old set — no simulator_set_destination_not_found error, no rebuild. Validate the set before reuse (e.g., compare existing.device.simulatorSetPath against the requested set and stop+rebuild on mismatch).
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At src/platform-runtime-resource-cleanup.ts, line 88:
<comment>Re-enabling warm retention for scoped-set simulators lets a later `open` reuse a runner whose build is bound to the previous simulator set. `ensureRunnerSession`/`resolveReusableRunnerSession` (packages/platform-apple/src/runner/runner-session.ts:150-158, 391+) reuse the retained session keyed only by `device.id` and never compare the set, and warm reuse skips the xcodebuild phase that would pass `-DVTSimulatorSetLocation`. If the same UDID is later resolved in a different set (a set cloned from the default set preserving UDIDs, or a copied device), the retained runner silently services the device with an artifact built for the old set — no `simulator_set_destination_not_found` error, no rebuild. Validate the set before reuse (e.g., compare `existing.device.simulatorSetPath` against the requested set and stop+rebuild on mismatch).</comment>
<file context>
@@ -85,8 +85,7 @@ export const platformResourceCleanup: PlatformResourceCleanup = Object.freeze({
!params.hasScreenRecording &&
- !params.hasLease &&
- !params.device.simulatorSetPath
+ !params.hasLease
);
},
</file context>
There was a problem hiding this comment.
Fixed at 3dd240c. Identity is now simulatorAddressFor (core/simctl.ts), the existing owner of "a udid with the set that holds it" — no new keying scheme. It now stores the resolved set path, so two addresses are equal exactly when they name the same simulator; every existing consumer re-resolves the path, so this doesn't change their behavior (new test in core/tests/simctl.test.ts).
Reuse: resolveReusableRunnerSession stops a registered session whose device isn't isSameRunnerSimulator and starts fresh, before the external-xctestrun early return, so external runners are covered too.
Lease/adoption: RunnerLease.simulatorSetPath is written by buildRunnerLease(device) (no longer optional-by-omission), and adoption refuses with typed reason simulator_set_mismatch on a set difference.
The slot stays keyed by udid (map key, lease file/lock) because xcodebuild cleanup and lease paths are per-udid — two concurrent runners for one udid in two sets would kill each other's xcodebuild. It's matched, not re-keyed.
Tests: runner-session-lifecycle.test.ts "the same udid in another simulator set starts its own runner instead of reusing this one"; runner-adoption.test.ts "a runner leased for the same udid in another simulator set is never adopted" (A->B, A->default, default->A); runner-device-set.test.ts "one udid in two simulator sets names two simulators". Mutation check by hand: disabling either check fails its test.
… set in destination errors - Reuse and lease adoption now require the same simulator: one udid in one resolved simulator set. The lease records the scoped set; a runner leased or registered for another set is stopped and restarted, never reused. - The runner derives a simulator's set through simulatorAddressFor, which now resolves the set path, instead of restating the simulator predicate. - simulator_set_destination_not_found names the set and the selected Xcode in its message. The Xcode version comes from the toolchain fingerprint owner, so external and adopted sessions report it too; the artifact no longer carries its own copy. The hint points at the error instead of printing field names. - The legacy XCTestDevices restore runs before adoption, so both startup routes pass through it, and a step another daemon already took counts as done instead of failing the startup.
|
At 18078a3 this replaces the XCTestDevices lock, symlink and handle with one argument helper used at both xcodebuild spawns, for net −328 production lines. I found no blocking defect. CI is green, and there are no conflicts. One design question first: would the legacy-redirect cleanup be smaller as a single step when the daemon or the Apple backend starts, instead of a call on every runner start that goes through a re-export in Related to that: could a symlink left by an older agent-device, with no backup next to it, survive a runner start and a daemon kill (runner-device-set.ts#L41)? And should the restore at runner-session.ts#L257 ignore ENOENT and EACCES, so that a cleanup error cannot fail a runner start? The live Xcode 26.2 and 27.1 runs are taken from the PR body. The Not blocking: the session tests could import |
…restore steps into helpers
…de 26.2 output The simulator_set_destination_not_found fixtures were invented text. They are now the output Xcode 26.2 (17C52) printed for the runner's own build-for-testing and test-without-building argv, with the destination naming a scoped-set udid and -DVTSimulatorSetLocation naming a set that does not hold it, and for the same build without the key. Destination lines that identify the host's hardware and simulators are left out.
…at daemon startup The restore ran on every simulator runner start and reached the session through a runner-xctestrun re-export. It is now one daemon-startup step on the platform owner lifecycle, after daemon.log publication. Without a device to compare against, it removes any symlink at XCTestDevices, as released versions did on every scoped runner start, and restores the agent-device backup. Each repair and any failure is recorded in daemon.log; a failure stops neither the daemon nor a runner start, because the runner no longer reads XCTestDevices.
|
Addressed the review at 18078a3. New head 9e623ba. Design question (single startup step): Moved the restore off the runner-start path entirely. It now runs once, at daemon startup, through a typed Symlink surviving a daemon kill with no backup: Now removed at the next daemon start, whatever it points at — matching what released v0.21.14 did on every scoped runner start. Each removal logs Ignore ENOENT/EACCES so cleanup can't fail a start: Since the restore left the runner-start path, it can no longer fail a runner start at all. At daemon startup every error is caught and recorded as Captured (not invented) xcodebuild output: Captured live on Xcode 26.2 against a real scoped-set simulator: build-for-testing (scoped + default set) and launch, all exit 70, Non-blocking (re-export, xcodeVersion fallback): Both done — re-export deleted with the design-question fix; the hint stays a static string, and the message now says "Xcode (version unreadable)" when the version can't be read within its 5s budget. Gates (head 9e623ba): typecheck, lint, format:check, layering, fallow — all pass. Live: daemon-startup restore verified for link+backup, link-with-no-backup, and EACCES, all non-fatal. A scoped runner start with the per-start restore removed still resolves and interacts correctly ( Replies to each inline comment follow. |
|
Summary of this round New head: Gates on 9e623ba: typecheck pass, lint pass, format:check clean, Live (Xcode 26.2, real device, isolated state):
Thread replies posted on all 10 inline comments (7 fixed-in-place, 1 fixed-then-superseded by the daemon-startup design, 2 declined with evidence). Top-level reply above answers the design question and the two open questions from the review. |
|
Reviewed at 9e623ba. The cleanup now runs once at daemon start instead of on every runner start, so the design question from the last review is answered. I found no blocking defect in the delta. The restore is best effort: a missing path, a permission error, or a link with no backup does not fail daemon start, and it only unlinks a symlink, so the target keeps its data. The call goes through the platform lifecycle seam, and the Apple code stays in the root composition file. The One behavior question for a decision: the delta drops the guard from 18078a3 (a backup exists, or the link points at our scoped set). Daemon start now removes any symlink at CI: Smoke Tests, Repo Guards and Coverage were still running at review time, and none had failed. Smoke Tests starts the daemon and the iOS runner, so it covers the changed path; a failure there needs a look against this PR. Not blocking: |
thymikee
added
the
ready-for-human
There was a problem hiding this comment.
1 existing issue remains and 4 new issues found across 37 files (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="website/docs/docs/commands.md">
<violation number="1" location="website/docs/docs/commands.md:207">
P3: The destination error does not always include the selected Xcode version; when the version probe fails, it reports `(version unreadable)`. Qualify this documentation claim to avoid promising a detail that may be unavailable.</violation>
</file>
<file name="src/daemon/server/daemon-runtime.ts">
<violation number="1" location="src/daemon/server/daemon-runtime.ts:562">
P1: This cleanup runs after daemon readiness is published, so a client can start a default-set runner while the legacy `XCTestDevices` symlink still exists. Restore the redirect before exposing the servers, while buffering its diagnostics until after publication.</violation>
</file>
<file name="packages/platform-apple/src/runner/__tests__/runner-startup-failure-reasons.test.ts">
<violation number="1" location="packages/platform-apple/src/runner/__tests__/runner-startup-failure-reasons.test.ts:177">
P3: The shared `assertFailureEnvelope` check was relaxed from exact equality to a prefix `startsWith` for every row in the table, not just the scoped-set rows whose message now appends a suffix. Rows whose failure message stops matching the old exact text (appended detail, reworded base, wrapped text) now pass silently; the exact expected message is only pinned in the two dedicated tests. Keep a per-fixture exact expected message so the other rows retain their regression guard.</violation>
</file>
<file name="packages/platform-apple/src/runner/__tests__/runner-early-exit-diagnosis.test.ts">
<violation number="1" location="packages/platform-apple/src/runner/__tests__/runner-early-exit-diagnosis.test.ts:220">
P3: These assertions pin only the message tail via `endsWith`, so the middle of the user-visible sentence (`xcodebuild found no simulator <udid> in ...`) can drift without failing. The full message is deterministic here (udid, set path and Xcode version are all known), so assert the complete `error.message` to catch a wording regression in the sentence the user sees.</violation>
</file>
Requires human review: Auto-approval blocked because this review re-detected 1 unresolved issue already reported by Cubic.
Fix all with cubic | Re-trigger cubic
| createOwnerScopedDeviceClaimReconciler(createDaemonRecoveryPlatformScope()), | ||
| baseDir, | ||
| ); | ||
| await restoreLegacyXctestDeviceSetForDaemonStartup(logPath); |
There was a problem hiding this comment.
P1: This cleanup runs after daemon readiness is published, so a client can start a default-set runner while the legacy XCTestDevices symlink still exists. Restore the redirect before exposing the servers, while buffering its diagnostics until after publication.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At src/daemon/server/daemon-runtime.ts, line 562:
<comment>This cleanup runs after daemon readiness is published, so a client can start a default-set runner while the legacy `XCTestDevices` symlink still exists. Restore the redirect before exposing the servers, while buffering its diagnostics until after publication.</comment>
<file context>
@@ -559,6 +559,7 @@ export async function startDaemonRuntime(
createOwnerScopedDeviceClaimReconciler(createDaemonRecoveryPlatformScope()),
baseDir,
);
+ await restoreLegacyXctestDeviceSetForDaemonStartup(logPath);
// Arms the initial idle-reap timer: a daemon that starts and never
// receives a request must still be able to reap itself.
</file context>
| ``` | ||
|
|
||
| - `--ios-simulator-device-set <path>` constrains simulator discovery and simulator command execution via `xcrun simctl --set <path> ...`. | ||
| - The XCTest runner's `xcodebuild` phases resolve a scoped simulator in the same set through `-DVTSimulatorSetLocation=<path>`; `~/Library/Developer/XCTestDevices` is never redirected. If the selected Xcode no longer resolves the simulator that way, the runner start fails with `details.reason: "simulator_set_destination_not_found"`, and the error names the set and the selected Xcode version. |
There was a problem hiding this comment.
P3: The destination error does not always include the selected Xcode version; when the version probe fails, it reports (version unreadable). Qualify this documentation claim to avoid promising a detail that may be unavailable.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At website/docs/docs/commands.md, line 207:
<comment>The destination error does not always include the selected Xcode version; when the version probe fails, it reports `(version unreadable)`. Qualify this documentation claim to avoid promising a detail that may be unavailable.</comment>
<file context>
@@ -204,7 +204,8 @@ agent-device devices --platform android --android-device-allowlist emulator-5554
- `--ios-simulator-device-set <path>` constrains simulator discovery and simulator command execution via `xcrun simctl --set <path> ...`.
-- The XCTest runner's `xcodebuild` phases resolve a scoped simulator in the same set through `-DVTSimulatorSetLocation=<path>`; `~/Library/Developer/XCTestDevices` is never redirected. If the selected Xcode no longer resolves the simulator that way, the runner start fails with `details.reason: "simulator_set_destination_not_found"`.
+- The XCTest runner's `xcodebuild` phases resolve a scoped simulator in the same set through `-DVTSimulatorSetLocation=<path>`; `~/Library/Developer/XCTestDevices` is never redirected. If the selected Xcode no longer resolves the simulator that way, the runner start fails with `details.reason: "simulator_set_destination_not_found"`, and the error names the set and the selected Xcode version.
+- On macOS, daemon startup puts back a `~/Library/Developer/XCTestDevices` that an older agent-device left redirected into a scoped set: it removes any symlink at that path, restores `XCTestDevices.agent-device-backup` when it exists, and records each step in `daemon.log`. A failed restore is recorded there too and does not stop the daemon or a runner start.
- `--android-device-allowlist <serials>` constrains Android discovery/selection to comma or space separated serials.
</file context>
| - The XCTest runner's `xcodebuild` phases resolve a scoped simulator in the same set through `-DVTSimulatorSetLocation=<path>`; `~/Library/Developer/XCTestDevices` is never redirected. If the selected Xcode no longer resolves the simulator that way, the runner start fails with `details.reason: "simulator_set_destination_not_found"`, and the error names the set and the selected Xcode version. | |
| - The XCTest runner's `xcodebuild` phases resolve a scoped simulator in the same set through `-DVTSimulatorSetLocation=<path>`; `~/Library/Developer/XCTestDevices` is never redirected. If the selected Xcode no longer resolves the simulator that way, the runner start fails with `details.reason: "simulator_set_destination_not_found"`, and the error names the set and includes the selected Xcode version when it is readable. |
| ): void { | ||
| assert.equal(envelope.code, 'COMMAND_FAILED'); | ||
| assert.equal(envelope.message, 'xcodebuild build-for-testing failed'); | ||
| assert.ok(envelope.message.startsWith('xcodebuild build-for-testing failed')); |
There was a problem hiding this comment.
P3: The shared assertFailureEnvelope check was relaxed from exact equality to a prefix startsWith for every row in the table, not just the scoped-set rows whose message now appends a suffix. Rows whose failure message stops matching the old exact text (appended detail, reworded base, wrapped text) now pass silently; the exact expected message is only pinned in the two dedicated tests. Keep a per-fixture exact expected message so the other rows retain their regression guard.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/platform-apple/src/runner/__tests__/runner-startup-failure-reasons.test.ts, line 177:
<comment>The shared `assertFailureEnvelope` check was relaxed from exact equality to a prefix `startsWith` for every row in the table, not just the scoped-set rows whose message now appends a suffix. Rows whose failure message stops matching the old exact text (appended detail, reworded base, wrapped text) now pass silently; the exact expected message is only pinned in the two dedicated tests. Keep a per-fixture exact expected message so the other rows retain their regression guard.</comment>
<file context>
@@ -173,7 +174,7 @@ function assertFailureEnvelope(
): void {
assert.equal(envelope.code, 'COMMAND_FAILED');
- assert.equal(envelope.message, 'xcodebuild build-for-testing failed');
+ assert.ok(envelope.message.startsWith('xcodebuild build-for-testing failed'));
assert.equal(envelope.details?.reason, fixture.reason);
assert.ok(
</file context>
| assert.equal(error.details?.simulatorSetPath, SET_WITHOUT_UDID); | ||
| assert.equal(error.details?.xcodeVersion, STUBBED_APPLE_TOOLCHAIN.xcodeVersion); | ||
| assert.ok( | ||
| error.message.endsWith( |
There was a problem hiding this comment.
P3: These assertions pin only the message tail via endsWith, so the middle of the user-visible sentence (xcodebuild found no simulator <udid> in ...) can drift without failing. The full message is deterministic here (udid, set path and Xcode version are all known), so assert the complete error.message to catch a wording regression in the sentence the user sees.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/platform-apple/src/runner/__tests__/runner-early-exit-diagnosis.test.ts, line 220:
<comment>These assertions pin only the message tail via `endsWith`, so the middle of the user-visible sentence (`xcodebuild found no simulator <udid> in ...`) can drift without failing. The full message is deterministic here (udid, set path and Xcode version are all known), so assert the complete `error.message` to catch a wording regression in the sentence the user sees.</comment>
<file context>
@@ -166,50 +175,76 @@ test('a session that never probed the device publishes no disk-image claim (#268
+ assert.equal(error.details?.simulatorSetPath, SET_WITHOUT_UDID);
+ assert.equal(error.details?.xcodeVersion, STUBBED_APPLE_TOOLCHAIN.xcodeVersion);
+ assert.ok(
+ error.message.endsWith(
+ `simulator set ${SET_WITHOUT_UDID} with Xcode ${STUBBED_APPLE_TOOLCHAIN.xcodeVersion}`,
+ ),
</file context>
1 participant
Summary
For a simulator in a scoped set (
--ios-simulator-device-set), both runner xcodebuild phases now get-DVTSimulatorSetLocation=<set>. With that, xcodebuild resolves and runs the test inside the user's set, so nothing has to swap~/Library/Developer/XCTestDevicesany more. The XCTestDevices redirect is deleted: the symlink and backup, the host-wide lock, the session redirect handle, thesimulator_set_redirecthandoff and adoption refusals, and the close-retain exclusion for scoped sets.This removes the data-loss path from #2935 by construction. On Macs whose Xcode shims are armed, the shims run
xcodebuild -runFirstLaunch, which cleans XCTestDevices. XCTestDevices now never points at user data, so those cleanups can't delete it. Scoped-set runners also work on those Macs now, where #2947 would have refused them.Evidence for the key: static presence in every Xcode from 14.0 through 27.1, and live runs on 26.2 and 27.1 (see the #2935 redesign comment). If a future Xcode drops the key, the runner fails with a typed
simulator_set_destination_not_founderror that names the set path and the Xcode version, and never falls back to a redirect. A one-time cleanup at daemon startup undoes a redirect that an older version left behind. It only touches state that agent-device wrote.Closes #2935. Supersedes #2947. 31 files. Production +171/−499, net −328. Tests +288/−879.
Validation
Tested head
9e623baff2:check:affected --runpasses (3758 tests, 497 files, command-docs 12/12). Also passing: typecheck, lint,format:check, layering, and fallow with no issues in changed files (one pre-existing inherited finding, unchanged).apple-runnerproject: 622/622 tests.open,snapshot, a runner-backedpress,snapshotandcloseall pass.-runFirstLaunchcleanups, includingsimctl --set XCTestDevices delete allduring the build.Library/Developer(EACCES) — the daemon stays up in all three, and each outcome is recorded in daemon.log. A scoped runner start with no per-start restore call still resolves and interacts correctly, with the hostXCTestDevicesdirectory untouched throughout.simulator_set_destination_not_foundmessage text is now captured from real Xcode 26.2 output (three xcodebuild invocations, all exit 70), not invented.