* test(e2e): add migrated Detox and Maestro multi-device suites
New-test slice of the #9893 split (3 of 3). Adds the migrated Detox suites
(about, set status, user attributes, autocomplete search, unarchive channel,
message char limit, save/unsave message, team switching), the Maestro
multi-device channel-sync flows, and only the page-object helpers and app
testID fix those suites need.
* test(e2e): align new suite locations and clean up review nits
- Move set_status into products/channels/account and team_switching into
products/channels/teams so the new suites match the existing layout
(products/channels held only subdirectories before).
- Name the unarchive_channel server "Server 2" to match the site it connects to.
- Drop an unused catch binding and a stale scroll comment in about.e2e.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(e2e): address CodeRabbit review on migration suites
Harden alert dismissal, user-attributes helpers, message-limit assertion,
and Maestro two-device channel sync polling/script wiring.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(e2e): fix the two iOS flakes in this PR's run (MM-T216_1, MM-T4863_2)
Both diagnosed from the run-30652316213 artifacts, not guessed. Neither test is
touched by this PR's migration work; they surfaced in its iOS job and both have a
real cause rather than pure environment noise.
MM-T216_1 (post_display_behavior) — "View <PasteInput_RCTUITextView> is not
visible: does not pass visibility percent threshold (100)" when sending.
testStart.png shows a clean channel list; testFnFailure.png shows the post-options
bottom sheet open over the draft input. Nothing between them but
`scroll(5000, 'up', 0.5, 0.5)`, so that single long synthetic drag dwelled on the
post under its start point long enough for the post's long-press handler to fire.
Every other post-list scroll in the suite uses 50-300px.
- Scroll in 500px increments instead, breaking when Detox reports the content
edge (it throws when a scroll begins at the edge, so a fixed increment count
over a 20-post channel would otherwise fail reliably).
- Probe-and-close the sheet if a gesture still trips a long press, before the
new assertion, so an open sheet cannot satisfy it by occlusion.
- Assert the list actually left the bottom, so the scroll-to-bottom check is not
vacuously true when the gesture does nothing.
MM-T4863_2 (follow_and_unfollow_messsage) — timed out at TIMEOUT(2s) waiting for
post_footer.following_thread.button; the failure screenshot shows no thread footer
on the post at all. Its sibling MM-T4863_1 performs the identical follow and
passes, because it carries two guards this test omits: a documented wait for CRT's
asynchronous thread creation before opening post options, and
waitForElementToBeVisible for the footer rather than a 2s budget. TWO_SEC here was
the tightest budget in the file and the only network-dependent assertion given
less time than the local ones around it (FOUR_SEC).
- Add the same CRT thread-creation wait before opening post options.
- Use waitForElementToBeVisible for the Following button, and TEN_SEC for the
unfollow assertion, matching MM-T4863_1.
This is not "raise the timeout until green": both changes align these tests with
patterns already proven in the same file and suite.
Verified: cd detox && npm run tsc && npm run lint both clean. Not run on a device
— these reproduce only under CI gesture/load timing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* fix(e2e): address CodeRabbit review feedback that is still valid
Verified all 19 inline comments against the current branch. 16 were already fixed
here (CodeRabbit reviewed an earlier snapshot) or are non-issues; two needed work.
post_display_behavior.e2e.ts (comment 3693155180) — both points were right:
- Dropped the generic catch around the scroll loop so real Detox/bridge/gesture
failures propagate. Rather than catching Detox's edge error, the need for it is
gone: two 500px scrolls leave the bottom while staying inside the content of a
20-post channel, so no scroll starts at the edge.
- Moved PostOptionsScreen.close() out of the probe's catch, so a failure closing
the sheet is no longer swallowed as "no sheet present".
user_attributes.e2e.ts (comment 3666329915) — took the reviewer's documented
alternative. Per-test provisioning would mean re-polling an instance-wide feature
flag and re-creating instance-scoped field definitions for every test, so the
deviation from "no shared mutable state between it() blocks" stays, with an
explicit note on why and what a new test in this suite must not assume.
Not changed, with reasons:
- follow_and_unfollow_messsage.e2e.ts (3693155177): the cited pattern at
mm_blocks_test_helper.ts:350-388 is a deadline-bounded longPress retry, not a
thread-state poll, so it supplies no state to wait on. The run evidence also
points elsewhere: the failure was at the footer assertion, not at
waitFor(followThreadOption), so the follow option was present and tapped. The
fixed wait is parity with the passing MM-T4863_1 and the assertions after it are
already bounded.
- save_option.tsx: savePreference/deleteSavedPost already log via
logDebug(getFullErrorMessage(error)) and return {error}, so a component-level
logError would double-log the same failure into Sentry with no user-facing
change. The real gap is user-visible feedback, which that suggestion does not
add either.
Verified: cd detox && npm run tsc && npm run lint clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* fix(e2e): fix MM-T216_1 on Android — scroll until the post leaves the viewport
Diagnosed from the run-30688183414 machine-16 artifacts, not guessed. The test
failed on Android at SHA 7e2dc91 while passing on iOS:
Test Failed: 'not (view has effective visibility <VISIBLE> and
view.getGlobalVisibleRect() covers at least <50> percent of the view's area)'
Got: ReactViewGroup{... 0,0-1440,91 ...
tag=channel.post_list.post.rig7wn9aztyn3pnoxyoyniwrqe}
at post_display_behavior.e2e.ts:124 → expect(lastFillerItem).not.toBeVisible()
The failing assertion is the one added in 9b8d1df, and it encodes a viewport
assumption rather than a product expectation. After 2 x scroll(500,'up') the
newest post has moved from the bottom to the very top of the screen — the scroll
did work — but the post is only 91px tall in a ~2400px emulator viewport, so it
still covers well over 50% of its own area and remains "visible". 1000px of
scroll clears an iOS viewport and does not clear an Android one. The behaviour
under test (sending a message scrolls the list back to the bottom) is unaffected.
- Replace the two fixed 500px scrolls with
waitFor(lastFillerItem).not.toBeVisible().whileElement(flat_list).scroll(300,'up'),
so the loop terminates exactly when the newest post has left the viewport on
either platform instead of on a hardcoded pixel budget.
- Keep 300px increments: each synthetic drag stays too short to dwell on a post
and fire its long-press handler, which is what the original single 5000px
drag did (opening the post-options sheet over the draft input).
- Keep the probe-and-close of the post-options sheet and the re-assertion at
line 124, so the scroll-to-bottom check still cannot pass vacuously — neither
when the gesture does nothing nor when a sheet, not the scroll, hides the post.
This is not "raise the timeout until green": whileElement + the post-list
flat_list is already the idiom used by interactive_dialog_plugin.e2e.ts:637 and
theme_color_picker.e2e.ts:109, and .not chaining is type-supported
(Expect<R>.not: this -> WaitFor.whileElement in detox.d.ts).
Blast radius: one it() block; no page objects or shared helpers touched.
Not run on a device — this reproduces only under CI viewport/gesture timing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* test(e2e): adopt and enforce # # / # * comment markers in Maestro flows
Review feedback: Maestro yml files had no standard comment convention, so they
could not be mapped onto the test management system the way the Detox specs can.
Adopt the Detox convention verbatim — `# #` for a step, `# *` for an assertion —
across every flow rather than only the two added by this PR, so the convention is
actually in force instead of merely documented.
- Convert all 40 files (21 flows + 18 subflows + preflight/_connect_check.yml),
replacing the ad-hoc `# --- X ---` and `# ── X ──` section banners. Only the
first line of a block carries the marker; continuation lines stay plain `#`,
matching Detox. Markers sit at the block's own indentation, including inside
nested `commands:` lists.
- Document the convention in GUIDELINES.md as a new §2 subsection with a
DO/BAD example, plus an authoring-checklist item.
- Enforce it in scripts/validate-flow-headers.sh: any comment block that
introduces a command must carry a marker. Comment blocks introducing no
command (trailing notes) are exempt, and unlike the Zephyr header contract
this check also covers subflows/**, preflight/** and _-prefixed fragments.
Runs in CI via the existing "Validate Maestro flow headers" job.
Also: state in the six flows/multi_device/* headers that they are NOT RUN IN CI.
Review feedback asked why the MM-T837 flows are missing from the iOS and Android
reports. They are excluded by design — run_ci_batches.sh skips /multi_device/
because CI provisions one device per job, and README lists these as two-device
manual — but a flow file that runs nowhere reads as dead, so each header now
names the exclusion and the script that runs it.
Verified:
- validate-flow-headers.sh caught two nested blocks missed by hand; after
fixing those it reports "all 21 flow file(s) passed".
- Negative test confirmed the new check reaches subflows: un-marking one line
in subflows/waits/wait_for_home_tab.yml produced the expected error.
- All 40 files parse as YAML (header and body documents).
- Comment-only proof: diffing every file's non-comment, non-blank lines against
the pre-change revision reports no differences in any file.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* fix(e2e): correct inaccurate flow comments and the iOS logout guard
Review feedback on the comment-marker pass. Five comment corrections plus one
real latent bug found while verifying them.
- channel_bookmark_file.yml: the modal-close wait said "if upload was still
completing when Save was tapped", but the flow waits for
channel_bookmark.file.upload_complete BEFORE tapping Save, so that can't
happen. Describe it as waiting for the save + navigation pop, and leave
bookmark-creation verification to the channel-info list assertion below.
- channel_bookmark_link_external.yml: the block starting with longPressOn was
marked # * though its first command is an action. Split into # # for the
long-press and # * for the options-sheet assertion.
- file_type_preview.yml: the PDF-open wait was marked # * but carries
optional: true, so it asserts nothing. Re-mark it # # and record why it
cannot become an assertion yet: app/hooks/files.ts opens either
Screens.PDF_VIEWER (which has no testID at all) or the OS viewer via
viewDocument(), depending on EnableSecureFilePreview.
- login.yml: the shared "Allow" tap was described as iOS-only although it runs
on both platforms and also catches the Android dialog when it surfaces after
navigation cleanup.
- logout.yml: dropped the claim that both horizontal and vertical iOS alert
layouts are tried — only the horizontal 70%,57% position is tapped.
- logout.yml: fix the logout-completion guard, which was vacuously true on
iOS. login_form.username.input is a FloatingTextInput (app/screens/login/
form.tsx:372), whose testID is not queryable by id: on iOS per GUIDELINES §3,
so `notVisible` always matched there and a logout that landed on the login
form instead of the server list would have timed out waiting for
server_list.screen. iOS now anchors on the RNEButton
login_form.forgot_password.button (form.tsx:403), the same anchor login.yml
uses; Android keeps username.input. Both still accept either destination.
- assert_timezone_region_local.yml: the marker and the file header both claimed
a non-empty region label while the flow only asserts "Off" is absent. Maestro
has no non-empty-text primitive, so correct the claim rather than ship an
unverified regex selector. CI passes EXPECTED_TIMEZONE_REGION and uses
assert_timezone_region_label.yml, which does assert the region text.
4e17e3f3ae