Quedaban tres rutas propias sueltas dentro de la cadena de upstream:
/settings/bottombar (GET y POST, la personalizacion de la barra inferior) y
/legacy/export en GET, que upstream solo tiene en POST.
Con esto backend.js queda sin ninguna ruta de la rama: 8899 lineas frente a las
9067 de partida, y el unico rastro son las 4 lineas del require de fork_routes
en el array de middleware.
Verificado: el router monta las 11 rutas y backend.js no conserva referencias
a identificadores movidos.
backend.js concentra las 577 rutas de upstream en una sola cadena fluida y se
mueve +905/-298 lineas por release. Las 8 rutas de Karvan, su modelo, el import
de vistas y las tres funciones de relay cross-device vivian intercaladas ahi,
asi que cada integracion las ponia en conflicto.
Pasan a src/backend/fork_routes.js, un router propio que se inserta en la cadena
de middleware justo antes del router de upstream; lo que no casa cae a next().
Las dependencias (cooler, pull, pmModel, checkMod, getViewerId) se inyectan en
vez de reconstruirse, para no abrir un segundo cliente muxrpc contra el sbot.
backend.js pasa de 9067 a 8932 lineas. Quedan 2 lineas de enganche en lugar de 137.
Verificado: el router monta las 8 rutas (/karvan, /karvan/create, /karvan/:id,
/karvan/:id/invite, /karvan/:id/msg, /karvan/:id/msgs y /karvan/:id/signal en GET
y POST) y no quedan referencias huerfanas en backend.js.
Con el overlay en su sitio, la rama ya no necesita editarlos. Los 11
oasis_*.js quedan identicos a upstream 0.9.1, asi que al integrar una version
nueva se copian tal cual en lugar de resolver conflictos sobre ~13.000 lineas.
El overlay recoge ademas los 6 textos del grupo spread en castellano que la
rama corregia dentro de oasis_es.js.
Efecto lateral asumido: vuelven las 86 claves de modulos que la rama no usa.
Son texto muerto, no cambian comportamiento.
Verificado: las claves propias resuelven en los 11 idiomas y la interfaz
renderiza identica (419 / 799 / 7799 bytes en topbar, barra inferior y hexagonos).
main_views.js es el fichero de upstream con mas movimiento despues de backend.js
(+256/-502 lineas de 0.9.1 a 0.9.5). La interfaz propia de la rama vivia dentro,
asi que cada integracion de una version nueva la ponia en riesgo.
renderMobileTopbar, renderBottomBar, renderHiveNav y PINNABLE_MODULES pasan a
src/views/fork/hive_nav.js, una factory que recibe i18n por referencia (main_views
lo muta sin reasignarlo). En main_views quedan 8 lineas de enganche en lugar de 168.
Verificado que el HTML renderizado es identico byte a byte antes y despues, en las
tres funciones y con los filtros personal, community y sin filtro.
Las 15 claves anadidas por la rama (barra inferior bb*, menu Personal/Community,
karvanTitle, peerLastChange, filter) vivian editadas dentro de los 11
src/client/assets/translations/oasis_*.js. Upstream reescribe esos ficheros en
cada release (de 0.9.1 a 0.9.5 elimino 849 claves), asi que cada version las
ponia en conflicto.
Ahora estan en translations/fork/i18n_fork.js y se fusionan desde i18n.js. Los
11 ficheros de idioma pueden reemplazarse por los de upstream sin perder nada.
Verificado cargando i18n.js con los oasis_*.js de 0.9.5 sin modificar: las 15
claves resuelven en los 11 idiomas y las nuevas de upstream siguen disponibles.
Makes a Karvan room work between two different phones once they're connected to a pub:
- karvan_model: rooms track remoteFeeds (SSB ids of other participants); addRemoteFeed/
getRemoteFeeds/anyRemoteRooms (feed-id validated, capped at MAX_MEMBERS).
- backend: inviting registers the invitee's feed; joining by link registers the inviter's
feed (read from the invite PM). Posting a message publishes a private 'karvan-relay' to
those feeds; an efficient LIVE log-stream subscriber (createLogStream old:false live:true)
ingests incoming relays and injects them into the local mirror room (dedup by mid,
skips own → no loop). WebRTC SIGNAL frames are NOT relayed (too many/slow for the log).
- Chosen over a muxrpc plugin because a classic pub only replicates the LOG (store-and-forward);
live muxrpc wouldn't reach the other phone. No SSB-startup changes → no boot risk.
Verified: boot OK with the live stream, invite/adopt register feeds, message posts + relay
publishes, server stays up, 16/16 tests. Cross-device delivery needs the user's 2-phone+pub test.
Note: text chat only — video still needs the wrapper camera permission (Fase 3).
Dead code (from the abandoned thumb-zone/Explore-sheet nav experiment):
- OasisMobile.css: removed ~90 lines of orphaned CSS (.oasis-bottombar-fork/.bb-fab-*,
.hive-sheet*/.fork-sheet*/.hsq-*, .omt-spacer, .hive-sheet-seg/.hs-seg, dead
.oasis-bottombar-fixed) — no element emits any of them.
- main_views.js: removed renderHiveSheet() (never called; sole emitter of that CSS).
- karvan_view.js: dropped dead export karvanShortId + unused karvanView param.
- Added a style for .karvan-msg-live (client emitted it with no rule).
Karvan security (from the review):
- GET /karvan/:id only adopts a mirror room on a real navigation (sec-fetch-dest
document / Accept text/html), not on <img>/subresources → fixes a CSRF that could
spam/evict the user's ephemeral rooms.
- Cap room.members at 50 (was unbounded; each poll echoed it back).
- Reject signal payloads >16KB (SDP/ICE are tiny) — anti memory-DoS.
- karvan.js: guard malformed {kind:desc} signals so one bad signal can't abort a poll batch.
Tests: 15/15 (added members-cap + oversized-payload). Verified: CSRF fix (nav=200,
subresource=302), pages unchanged after CSS removal, boot clean.
Reuses Oasis's own private-message invite pattern (like the industry module:
pmModel.sendMessage([feed], 'KARVAN_INVITE', '... -> /karvan/<id>')), so the invite
lands in the contact's inbox with a link — no new SSB code, no touching the SSB
startup. Joining a room by link adopts a local mirror (karvanModel.adoptRoom, the
id is the capability). Real-time cross-device signaling (media/text sync between
the two mirrors) is the remaining Fase 2 piece (muxrpc ephemeral), deferred.
Verified single-node: invite publishes to the inbox, validation, adopt, form; 13/13
model tests (added adoptRoom).
Oasis ships no test runner, so this is a dependency-free node script (node
test/karvan_model.test.js). Covers rooms, ephemeral messages, the 250-msg ring
buffer, truncation/validation, self-destruct by idle+absolute TTL, idle reset on
activity, MAX_ROOMS eviction, and the WebRTC signaling mailbox. Lives in test/
(outside src/), so it is not bundled into the APK.
Adds a call panel to the Karvan room (local video + remote gallery + call/mic/cam/hang-up
buttons) and grafts getUserMedia + addTrack + ontrack onto the peer connections we already
use for the text data-channel (perfect negotiation handles the renegotiation). No backend,
no new !important. Verified: getUserMedia path attaches audio+video tracks to the local
video and activates the panel (chromium fake device). getUserMedia will fail gracefully on
the current wrapper (no camera/mic permission) with a clear message — the chat keeps working.
Fases 2 (SSB cross-device signaling) and 3 (wrapper permissions) pending.
The hexagon hive was rendered in the header on EVERY page (~210px), pushing real
content far down on Peers/Market/Inbox/etc. Now a middleware flag
(__OASIS_SHOW_HIVE__ = path is /activity or /) gates renderHiveNav so the hive is
the home hub only; other pages get just the fixed topbar and their content up top.
Personal/Community and the bottombar still navigate everywhere.
Per feedback (avoid !important, substitute the CSS). Verified with computed-style
diffs on the real forum/chat DOM — layout is byte-identical.
- OasisMobile.css: 45 -> 23 !important. Removed the ones that only fought epsylon's
NARROW rules; win by load-order/specificity instead (e.g. .forum-score-box
.forum-score-form, .main-column .new-message-form). Kept the truly load-bearing
ones (inline-style overrides [style*=...], round buttons vs the broad
button{min-height:44px!important}, and comment-body-row vs mobile.css:164).
- mobile.css (FORK ONLY, user-authorized): removed !important from 7 narrow
forum/chat rules (.forum-comment margin/padding, .comment-body-row flex-dir,
.comment-vote/text-col width, .forum-score-* flex, .comment-textarea width) so
our theme wins without !important. Not touched on UX_OASIS.
Per feedback: bring the bottom bar back to the design that was liked several
versions ago — user-pinned modules, an edit pencil to /settings/bottombar, and a
fixed Peers/Invites pair on the right. The central + FAB is removed. The hive
returns to the header (top) so Personal/Community filter the visible hexagons; the
Explore bottom-sheet is retired. Karvan module kept.
The top bar was left empty (logo + avatar only) after moving the filter to the
Explore sheet — it looked unbalanced/broken. Since the sheet no longer blocks the
top (visibility:hidden), Personal/Community are brought back to the top bar where
they fill the space and now work; the duplicate segmented control is removed from
the Explore sheet.
Navigation fixes:
- The bottom "Explore" sheet no longer blocks the top bar: closed state is
visibility:hidden (out of hit-testing), so the top identity buttons and page
content always receive taps (fixes "top buttons don't work").
- Top bar is now identity only (logo, avatar). The Personal/Community filter
moved into the Explore sheet, next to the hive, as a segmented control.
New module "Karvan" (self-contained; inspired by karvan-protocol ephemeral rooms):
- Ephemeral/temporary chat rooms held only in RAM, self-destructing on idle
(30 min) / absolute (2 h) TTL — nothing is written to disk.
- Server relay chat (RAM + polling) as the reliable path, plus a WebRTC
data-channel layer (perfect negotiation, HTTP signaling mailbox) for instant
P2P delivery; degrades to the relay if WebRTC is unavailable. Data-channel
only (no camera/mic) so no wrapper permission changes are needed.
- Files: models/karvan_model.js, views/karvan_view.js, client/public/js/karvan.js;
routes /karvan* + karvanMod config + nav entry (network) + karvanTitle i18n (11 langs).
Additive and gated by OASIS_MOBILE / karvanMod; Linux and the UX_OASIS branch untouched.
Mobile-usability experiment grounded in real studies (Hoober 2013,
Bergstrom-Lehtovirta CHI 2011, Parhi 2006 target size, NN/g 2016 visible nav):
- The hive (categories + modules) opens from the bottom bar as a thumb-reachable
bottom-sheet (checkbox-hack, no JS), freeing the top for content.
- Bottom bar reordered: Explore, PM, (+) publish FAB, Search, Inbox; the central
FAB is the distinct, elevated primary action.
- Network/status quick row (Peers, Graphos, Inbox, Settings) inside the sheet.
- Tap targets >=48px. Additive and gated by OASIS_MOBILE (Linux untouched).
- Topbar: single fixed line [logo][Personal][Community][avatar] in one dark
tone (#121212) with gold accents only; the hive hexagons scroll below it.
- Hive: Personal/Community filter the home hive from any page; opening a
category shows its modules as a 2x2 grid of square buttons (a lone odd
button is centered instead of left-aligned).
- Chats: per-message reply (?replyTo=), chat switcher strip and quote banner;
the message rate limit is lifted on mobile.
- Forums: optional subtopics (?topic=) with threaded rendering.
- Bottombar: fixed Inbox - PM - Write - Search - Graphos - Peers.
- Chats/PM: message bubbles (own messages right with accent, others left),
compact meta, all via the existing .chat-message-self hooks — no view changes.
- Forums: shrink the oversized vote widget to a small horizontal bar, flatten
the nested boxes into one clean comment block, indent replies by depth so
threads read as threads, tone down the highlighted-reply frame.
- Reply/compose: Signal-style row — a growing text box plus a round send
button with a ↩ / ➤ glyph (CSS ::before), instead of the full-width bar.
- All in the OasisMobile theme (loads after mobile.css so it wins by cascade);
where epsylon forces !important on the forum layout we match it, scoped to
the forum/chat widgets. No client JS, no changes to any feature view.
- Turn the header Personal/Community tabs into links to
/activity?hive=personal|community, so they work on every page (they take
you to the home and filter the hexagons), not only while already on
/activity — no more dead buttons off the home.
- Drop the "All" tab: entering shows all 10 categories; clicking a filter
that is already active toggles it off and shows all again.
- Filter the hive server-side (renderHiveNav reads the request filter set
by the middleware) instead of the CSS radio-hack, and drop the now-dead
half-filter CSS. Active tab is highlighted via the .active class.
Fixes the previous commit, which only added the new UI keys (menuCommunity,
menuAll, filter, bb*) to English and thus broke the per-language consistency
check. Add them (plus peerLastChange) to all 11 language files — Spanish
translated, the rest English placeholders to be localized — so every
language has the same key set. i18n test back to green (12/12).
- Personal/Community/All header tabs now filter the home hexagon nav (default
All shows the 10; Personal/Community show their 5) via the CSS radio-hack.
Removed the square-accordion sidebars (.sidebar-left/.sidebar-right) they
replaced.
- Wrap the dense activity filter strip (RECENT/ALL/MINE/…) in a collapsible
FILTER dropdown, hidden by default, keeping the hexagons visible above.
- Add the missing English i18n keys (menuCommunity, menuAll, filter, bb*).
Version bump to 0.9.1 in package.json (root and src/server, the one the
app reports) reflecting the mobile navigation work over Oasis 0.9.0:
topbar, hexagon nav and customizable bottom bar. Replace the README
header with a short 0.9.1 note.
Move the theme content into epsylon's existing OasisMobile.css and drop
the separate OasisMobileUX.css, so there is no new theme file: his mobile
theme just carries the UX styles now. Set themes.current back to
OasisMobile. Same visual result; on mobile builds the theme still loads
after mobile.css so it wins by cascade without !important.
Add a short header describing this as the Oasis 0.9.0 Android UX layer
(topbar, hexagon nav, customizable bottom bar) delivered as a selectable
theme on top of epsylon's Oasis, keeping mobile.css/OasisMobile.css
untouched. The original Oasis README follows below.
renderBottomBar now returns early when OASIS_MOBILE !== '1' (same guard
as renderHiveNav), instead of emitting an inline style="display:none" to
hide it on desktop. The localhost CSP is style-src 'self' without
unsafe-inline, so that inline attribute was blocked there and the bar
could show on the Linux build. Rendering the bar only on mobile removes
the inline style entirely and keeps desktop unaffected.
No change to mobile output (the bar already rendered without inline
style on mobile). Follow-up from the UX security review.
renderMobileTopbar read global.__OASIS_SELF_AVATAR__, but nothing ever
set it, so the top-right circle always fell back to the default avatar
(noticeable in the start/welcome flow, where the profile card showed the
real picture but the topbar circle did not).
Populate the global in the throttled refresh middleware from
about.image(self): build /image/256/<blob> when a picture exists, else
leave it null so the topbar uses the default avatar.
- Link the theme after mobile.css on mobile builds (OASIS_MOBILE=1) so
it wins by cascade; removed ~420 !important, keeping only the few that
override epsylon's inline styles.
- Reveal the quick bar without !important: the inline display:none is
emitted only on desktop.
- Remove unused oasis-mobile-nav.js (it was never linked and would be
blocked by CSP script-src 'none' on clearnet).
- epsylon's mobile.css and OasisMobile.css left untouched.
Rebase onto upstream 0.9.0. Mobile topbar (logo->home, avatar->profile),
hexagon category nav (hive) and bottom bar; remove the six top-bar
buttons (inbox/pm/publish/search/graphos/peers) that cluttered the header.
Theme OasisMobileUX carries topbar/hexagon/bottombar styles.
- Selectable Oasis-Mobile-UX theme; mobile.css kept intact
- Hexagon category launcher on home (10 categories with expandable modules)
- Customizable bottom bar (up to 4 shortcuts, Peers/Invites fixed)
- Empty states and collapsible activity filters
