className must appear once, with no spread after it, as the imported PhoneHit*
identifier bare, a classNames/clsx/cx direct argument, or a template span, and
the identifier must occur nowhere else in the expression. Tag enumeration uses
the same AST. expectPhoneHitClasses takes (string|null)[] (null = no class), as
the phone-hit-pages branch does, and is exercised by a test on a temp source.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A PhoneHit name inside a string, built by concatenation, behind ?:/&&/||/??, or
followed by a props spread that can override className no longer satisfies the pin.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Back/close buttons beside non-tappable titles take PhoneHit; the zoom out/in
buttons beside the zoom Chip take PhoneHitSpread24. Test pins each variant and
ties the sized ones to the drawn size.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Room<N> throws when no Spread of its size exists, and a test reads the compiled class lists to prove Room<N>
composes Spread<N> and PhoneHit. Props spread into an IconButton or Icon make the size undetermined (allow-list
entry required). <IconButton and <Icon match only before whitespace, > or /. iconButtonTags derives from iconButtons.
expectSizeTie fails on a file with no IconButton. Fixture tests live in phoneHitTestKit.test.ts.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The helpers moved to phoneHitTestKit.ts so each later group adds its own phoneHit.<group>.test.ts. A new
check ties PhoneHitSpread<N>/Room<N> to the px the button draws (IconButton size padding + Icon size); a size the
source does not state needs an explicit allow-list entry. It found RoomInput's reply-cancel (16 px icon, 24 px
button) carrying Spread32; it is Spread24 now. Room<N> takes its inline margin from the Spread class of the same
size, so the two axes cannot disagree.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Soul re-gate of 446306b1: swapped Spread24/26 definitions, display:none or overflow on the ::after, position:static on a
Spread rule and an aliased import all passed; all killed now.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Soul pass on fc92809e: RoomNavItem's buttons are 24 px, not 32 (36 px pitch, hit areas overlapped); the unpin button is 26 px
beside the Open chip. Six mutants survived (position: relative, pointer-events, z-index, title for className, aliased
import); all now killed.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
MessageEditor, RoomNavItem, MembersDrawer, ReactionViewer, RoomPinMenu, RoomViewTyping, CallChatView and JumpToTime
get PhoneHit (alone) or PhoneHitSpread32 (two neighbours in a row). phoneTapTargets.test.ts now lists them, plus
PasswordInput, Desk, SkyStateControl and PasswordLoginForm, which had no test, checks each class is imported from
styles/PhoneHit.css, and pins the ::after content (a mutant without it passed before).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Soul pass on 77edebff: the watchdog no longer counts interrupted entries from
a previous page (reload loop at startup), no longer reloads on sync ERROR/STOPPED
(js-sdk recovers offline by itself) or while offline, refuses instead of
throwing when sessionStorage is unusable, startClient is idempotent, and the
slow-send label says Resend may duplicate. resumeSync drops its health check
and recover hook: nudge only.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Sends have no timeout and sync polls do, so a dead connection leaves a send
SENDING forever with no request reaching the server while its echo looked
delivered. A watchdog (send pending past 20 s, sync ERROR/STOPPED for 30 s)
calls a guarded reload; what the reload interrupts is sent again under its
own txn id. Uploads and stickers go through the outbox. Resend takes a fresh
txn id while the SDK still holds the failed echo. The resume-sync recovery no
longer calls stopClient/startClient, which closes the rust crypto for good
("null pointer passed to rust" on later encrypted sends).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The SDK puts a local echo in the live timeline on sendMessage and marks it
NOT_SENT on failure; the timeline rendered it like any delivered message and a
reload dropped it, so a send killed with the page looked sent and vanished.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Soul pass findings: a failed localStorage write no longer loses the entry
(memory copy wins until a write succeeds), a stale draft is dropped when
its rewrite fails, storage reads/removes cannot throw, and pending entries
older than the page load read as interrupted with Resend available.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The draft lived in a memory-only jotai atom written only when RoomInput
unmounted, so a reload lost it. A sent message was cleared from the editor
at once and held nowhere durable. The draft is now written to localStorage
(debounced, flushed on pagehide/hidden) and restored on load; a send stays
in a per-account outbox under its txn id until the server confirms, shown
above the composer with Resend and Discard.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Soul pass findings on the stage 0-1 branch: the worker stored the token on every deployment, push configured or not
(now only while the browser holds a push subscription; the page re-sends its session after the toggle changes it);
the toggle read the browser's subscription, not the account's pusher (now both); a subscription under an old VAPID
key was reused (now replaced); a failing client list now still shows the notification.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>