| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GroupPage's 6620-line app.js carried Chat and Files wedged in directly, with
no way to add another group-level app without touching the shell itself. It
is now app.js (routing, non-group pages) plus nine focused files — apps.js
(the registry), chat-app.js, files-app.js, video-player.js, group-page.js
(the shell), group-settings.js, hub-client.js, icon.js and file-utils.js —
with docs/apps.md as the checklist for adding one (Videos/Music/Photos are
sketched there, not built).
Node side gained the matching enablement mechanism, mirroring
member_upload exactly: a roster setting, a signed apps_enabled op enforced
by _has_admin_authority, exposed in the handshake ack. Operators toggle
applications per group from Settings, which also gained a small reorder:
Invite, Pairing, Applications, Shared directories, Uploads, danger zone,
Your devices, Members.
Two bugs surfaced during the split, both missing an import across the new
file boundary and invisible to node --check or a module-load probe since
they only throw when the code path actually runs:
- group-page.js called onRefreshAuth on a stale-token handshake rejection,
but app.js never imported refreshAccessToken from hub-client.js — so a
brand new member (including a group's own creator) hit "Not a member of
this group" and the retry silently failed, throwing before it could
refresh the token.
- chat-app.js called getLocale() for message timestamps without importing
it from i18n.js. Opening Chat on a group with real messages threw mid-
render; uncaught, that appears to wedge Preact's render scheduler, so
every button on the page stopped responding until reload.
Caught the second class of bug with a proper no-undef audit across all
split files (a temporarily installed ESLint 9, since the system one is too
old to parse this codebase's syntax) rather than trusting grep. 827 tests
pass; 6 new ones cover the apps_enabled policy.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016SF6RKNBKg9qejmoMJ9ybA
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add a status panel at the top of the Node page — always visible, even
before an MNP connection exists — showing the meshbay-node systemd
unit's own state (via `systemctl --user show`, main process only) with
Start/Stop/Restart controls. This is the piece the rest of the page
cannot provide: it has to work while the daemon is stopped or crash-
looping, which the MNP-based sections require the daemon to already
answer.
While touching node lifecycle: `reload` and `restart-daemon` in the
CLI shelled out to pgrep + SIGTERM/SIGHUP and respawned the process by
hand, logging to a hardcoded /tmp path. That pattern already SIGHUPed
a developer's own running node by accident once (see the old
test_cli_dispatch.py comment). Both now delegate to
`systemctl --user reload|restart meshbay-node`, which the unit already
supports correctly (ExecReload=, Restart=on-failure).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016SF6RKNBKg9qejmoMJ9ybA
|
| |
|
|
|
|
|
|
|
|
| |
GroupPage keeps the same component instance across a group switch (no
key at the router), so currentPath/selected/filter were leaking from
the group just left. A directory that exists in one group but not the
other (e.g. "outputs") then showed an empty listing after switching.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016SF6RKNBKg9qejmoMJ9ybA
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Node daemon no longer blocks startup on slow directory scans — initial
indexing runs in the background so the node reaches "running" immediately
after transports are up. Fixes the wizard failing to detect the node when
large USB/NAS roots take minutes to scan.
Also: wizard key-linking deadlock resolved (main.js links during poll),
invite form stays in DOM during reconnects (disabled instead of destroyed),
pairing code bridges to renderer, and firewall docs for LAN casting added.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
| |
When the wizard calls node:start with {hubUrl, username}, the handler
writes a minimal node.toml and a random unlock.key if they don't exist.
The daemon then creates the keystore on first start using the unlock
file — fully non-interactive.
Existing configs are left untouched: if node.toml or unlock.key already
exist, provisioning is skipped and the node starts as before.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
| |
node.*, wizard.*, sidebar.node, settings_node.*, and group.retry
were present in en.js but missing from fr/es/de/it/nl/pl/pt-BR/zh-CN/ja.
All 10 locales now have identical key sets (test_key_sets_match_english passes).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Desktop client guides new users through the full onboarding sequence
without a terminal: detect local node → start daemon (systemd with
direct-start fallback for dev) → link node key to hub → create group →
attach directories → gek-init → operator pair.
- node:detect returns configured field to distinguish missing vs stopped
- node:start tries systemctl first, checks is-active, falls back to
spawning the binary directly when the unit crashes (dev mode)
- probeNode() extracts the shared config→token→fetch pattern
- SetupWelcome on empty HomePage links to /create-group
- CreateGroupWizard step 0 handles node detection and start
- platform.js exposes node.installed() and node.start()
- 12 setup.* i18n keys added to all 10 locales
- Integration test for node:start fallback logic
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Re-serve decrypted fMP4 video over HTTP on the LAN so a Chromecast can
play the stream. The relay runs in the Electron main process — same trust
boundary as downloads and MSE playback.
- cast-relay.js: HTTP server with BoxAccumulator (reassembles WebRTC
chunks into moof+mdat pairs), ring buffer, backpressure, finish() for
clean end-of-stream, fixed port range 19550-19553
- cast-chromecast.js: mDNS discovery (bonjour-service) + CASTV2 protocol
(castv2-client), connect/reload/disconnect lifecycle
- Seek-aware: relay restarts on every seek, Chromecast reloads new URL;
generation counter prevents stale async errors from killing active
restarts; landingPlayheadRef suppresses programmatic seeking events
- Device picker in video top bar with scan, device selection, copy-URL
fallback, and cast status indicator
- IPC bridge (main/preload/platform) for start/push/stop/finish/status/
discover/chromecastConnect/chromecastReload/chromecastDisconnect
- Phase 3 design doc for DLNA/Smart TV in docs/cast-smart-tv.md
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Create Group wizard (Electron-only) consolidates 6 steps across 4 interfaces
into a single multi-step page: group creation on hub, node attachment, root
selection via folder picker, GEK initialization, and auto-pairing — all in one
flow. Browser SPA keeps its current behavior unchanged.
Public group support (Option A — GEK for all groups):
- All groups have GEK regardless of visibility; open-join groups auto-admit
via TOFU when join_policy is "open"
- Key rotation blocked for public groups (API guard + UI hidden)
- Hub signaling allows WebRTC offers for nodes hosting open-join groups even
when the caller isn't a member yet
- attach_group writes join_policy to node.toml
- Daemon loads GEK for all groups, not just private ones
- Known-device path in join_request now auto-admits to open-join groups
Node loopback API bridge (Electron IPC):
- node:detect, node:call, node:pairing-code IPC handlers in main process
- Renderer never sees tokens, paths, or keys (session token = physical access)
- platform.js node namespace for UI consumption
- Loopback endpoints: roots CRUD, member-upload toggle, reload
Bug fixes:
- Root change detection: removed premature ctx["roots"] updates from add_root
and remove_root that prevented indexer retarget on reload
- Duplicate offline message: global fallback now gated on !group
- Signaling membership check: fallback to open-join groups for non-members
Sidebar groups sorted by last_activity_at (most recent first):
- New Group.last_activity_at column with Alembic migration
- POST /v1/groups/{id}/activity endpoint, called on connect and chat send
- Client-side sort + throttled hub updates (1/min)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Node admin panel (NodePage) now covers every CLI operation over MNP:
group attach/detach, roster, member unpin, GEK rotate, denylist, reload.
Daemon hot-loads new groups and tears down removed ones on config reload
instead of requiring a full restart. Group attach/detach via MNP or
local API triggers an automatic reload so the group is live immediately.
Fixed GroupPage hang on first visit to a newly created group: the JWT
issued at login didn't include the new group, the node rejected with
not_a_member, and the token-refresh path returned without re-triggering
the connect effect (Boolean(token) didn't change). Now bumps retryKey
after a successful refresh so the effect re-runs with the fresh token.
NodePage marks groups hosted by the node but absent from the hub with a
"not on hub" badge so stale groups are visible and easy to remove.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Hub/UI:
- Icon-only group tabs (chat, files, settings) with per-group default tab
- Transfer widget: filename becomes a clickable link to open completed downloads
- Pulse animation on transfer icon (pale→dark green) while active
- Download button feedback in FilePreview (spinner, auto-reset)
- Group mute toggle persists across navigation
- Login page autofocus, chat refocus after send
- Theme toggle closes menu, status badge and duplicate connecting removed
- Create-folder restricted to operators, download-path note removed
- User preferences API (CRUD) with Alembic migration
- Profile: email display/edit via PATCH /v1/users/me
- Settings: "Defaults" section for default tab selector
- All 10 locale files updated
Node:
- upload_dir in node.toml: separate filesystem path for uploads
- Root.direct flag: uploads land at root path, no subdirectory
- CLI --upload-dir flag on `group add`
- Admin UI accepts upload_dir
Client (Electron):
- shell.openPath bridge for opening completed downloads
- platform.js passes open callback from native save
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The stylised M replaces the letter M in "MeshBay" at the top left; `eshBay`
beside it stays text.
**The source artwork is now in git.** It was in `QE/`, which is gitignored, so
it would not have survived a clone — and the tree is about to be moved to
another machine. `assets/brand/` holds the original and the `convert` command
that derives the 10 KB nav asset from it (trimmed, 72px tall for a picture shown
at 30, so it survives a 2× screen). Verified: the documented command reproduces
the committed file byte for byte.
The derived file is in `_ASSETS`, which is what the `/a/<hash>/` fingerprint is
computed from. A file missing from that list is a file whose change never moves
the URL, so a browser holding the old one never asks again — the comment on that
list says so, and a logo is exactly the kind of asset one would forget.
Its URL is resolved from `import.meta.url`, so the hub's fingerprinted path and
the application's `app://` scheme both come out right without either being named
in the source.
**The two sizes are independent, and that took four rounds to get right.** The
picture's height was written in `em` — a fraction of the lettering — so every
adjustment to the text resized the picture by the same stroke and the ratio
between them could never change. The operator spotted the coupling before I did
("si on change la taille du M, tu vas encore tout décaler"). The height is in
pixels now: `font-size: 1.375em` (22px) and `height: 30px`, two knobs that move
one thing each. The vertical nudge went with it — it was correcting a
misalignment I was causing myself.
Also: the wordmark blue goes one step down from sky-400 towards sky-500. It sat
brighter than the picture next to it, which made the two read as different
materials rather than one object.
A note on how this went wrong first: I checked the result on a mock HTML page I
wrote, not in the application. The picture was loading the whole time, but at
25px against a 29px line box it was *smaller* than the letters, so the wordmark
read as unchanged and the report was "none of what I asked for was done". Every
size since has been measured in the running window over CDP.
883 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A group where every member may add files stays the default. Some groups want a
library the operator curates, and until now the only way to get one was to
designate no upload root at all — which refuses the operator too.
**The node enforces it; the interface merely stops offering it.** The Upload
button in the Files toolbar and the paperclip in the chat composer both
disappear, which is a courtesy to the people who are not trying. The control is
`_do_file_upload` refusing with `member_upload_off`, so a member on an old tab,
or one speaking MNP directly, gets the same answer. There is a test for each,
and the enforcement test is in the node package rather than beside the UI one so
nobody reads the hidden button as the mechanism.
**Changing it is a signed operator instruction** — `OP_MEMBER_UPLOAD`, on the
same path as removing a member. An unsigned one would let any member turn it
back on and make the setting a suggestion. The transcript's subject is `on` or
`off`: what the operator is shown before signing has to name the outcome, not
the operation.
**It lives on the node**, in a new `group_settings` table in `roster.db`. Not
the hub, which has no business deciding who may write to someone else's disk.
Not `node.toml` either: that file is hand-written and full of comments recording
decisions, `ops.py` appends to it rather than round-tripping it through a
writer, and a setting toggled from a panel must not rewrite the operator's file
or need a restart. The value is cached in the group context because the upload
path is synchronous, and the signed operation updates both — storing it without
applying it would make the panel say one thing while the node did another.
**Absent means allowed**, at every layer: no row in the table, no key in the
context, no field in `handshake_ack`. An older node and an older client both
behave exactly as before, and upgrading never silently closes a group. Each of
those three has its own test, because they fail independently.
The operator is always exempt — otherwise turning it off locks them out of their
own node with a config file and a restart as the only way back. `is_node_admin`
was being computed in two places by then and is now one function, since two
copies of "is this the operator" is how the ack and the gate come to disagree.
A change reaches everyone already connected via `member_upload_ack`, so the
button goes without a reconnection. That message is both a broadcast and the
reply to the request that caused it, which is why the client does not return
early on it.
Docs updated for a cold start: draft-v6 §2.1b and change 9, a new "Where Phase
13 stands" section in CLAUDE.md recording what is built, deployed and still
missing, the module map row, and desktop-client-v1 §10b on the Settings tab and
where group settings live.
883 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
**The chat tab was 8px too tall, at every window size.** The panel is sized from
JS to `viewport - top - 16`, which puts its bottom 16px above the fold — but it
sits inside `.main`, which adds 24px of padding below it. Eight pixels of
document past the window, whatever the window. Measured at 700, 900 and 1200:
`scrollHeight` 708, 908, 1208.
This is the second one of these — the sign-in card was `.page-center` and
`.layout` each reserving `100vh - 52px` — so it is now measured in the suite
rather than reasoned about. `tests/harness/scroll_probe.py` renders the real
markup against the real stylesheet and **runs the real `fit()` lifted out of
`app.js`**: a copy of the formula in a test would go on passing after the
original changed, which is exactly the bug being guarded.
The fix does not encode 24 anywhere. The first pass runs as before, then the
leftover is measured and taken off, so anything added below the panel later is
absorbed the same way. Now `scrollHeight == innerHeight` at all three heights,
nothing below the fold, and the panel still fills the room it has — that last
one has its own test, because shrinking the chat to 240px would satisfy every
other assertion here and be useless.
The Settings tab was measured too and is **not** a bug: it fits at 1200px and
overflows only when its content is genuinely taller than the window.
**Group creation asked one question twice.** Visibility and admission were
separate selectors that could only ever be set together — picking Public reached
over and set the policy — and two of the four combinations are meaningless. The
API already refused public+invite with a 422, so the form could build a request
that could not succeed. Private+open was accepted and should not have been: a
group anyone may join that nobody can find is a listing with the listing
removed, since joining goes through the node and there is no link to pass
around.
So: one selector, "who can join", and the request derives the rest. The API now
refuses the other impossible pair as well, with a message that says which way to
resolve it. Six locale strings the visibility box owned are deleted rather than
left unread in ten files, and the two surviving descriptions now say what each
choice means for who can *find* the group — with the word "public" gone from the
page, nothing else would have said it, and someone would publish a group without
meaning to.
865 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
the widget
**"Zone sensible" said nothing.** The heading is now the name of the action in
it — "Quitter le groupe", or "Supprimer le groupe" for the owner, who sees a
different button. Worth stating because it is not quite what was asked for: a
fixed "Quitter le groupe" would have sat above a delete button for whoever owns
the group. The red goes with it; only the button is red, which is where the
warning belongs. `members.danger_title` is gone from all ten locales rather than
left behind unread.
**A folder name no longer ends in a slash.** The folder icon in the cell beside
it already says what it is.
**The transfers widget turns green while transfers run.** The badge counts them,
but a count has to be read; colour is what carries from across the room, which
is the point of a widget in the nav bar rather than on the page. Derived from
the live list on every render, so there is no state that can forget to clear
when the last transfer ends.
The class is set in `app.js` and coloured in `style.css` — either alone does
nothing and neither fails loudly, so there is a test for each half.
848 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
settings tab
**Joining a group could hang.** Reported after a first attempt that never
finished and a later one that worked — the shape of a network wait with no
deadline, and there were two.
Signaling here is non-trickle: the offer is not sent until ICE gathering says
it is done. A STUN server that is slow, filtered, or resolved through a DNS that
is not answering means `icegatheringstatechange` never reaches `complete`, and
`connect()` never returns. Same shape as the fullscreen denial fixed yesterday:
a promise that never settles leaves no error to find. Gathering now has four
seconds, after which the offer goes out with what it has — host candidates are
already there, which is enough on a LAN, and giving up instead would turn a slow
STUN server into a refusal to connect.
The second: `hub:fetch` in the desktop client had no timeout, so a host that
accepts a connection and then says nothing holds the request for as long as the
OS allows. `hub:probe` had one; the handler that carries signaling did not. Now
thirty seconds — longer than the hub's own fifteen-second signaling wait, so it
cannot abort a call that was about to succeed — and it says the hub did not
answer rather than "fetch failed".
**Resume positions belonged to the machine, not the account.** Stored as
`mb:pos:<file>`, so a second account signing in on the same computer was offered
"resume where you left off" in a film it had never opened. Wrong on its own
terms, and a small disclosure of what the other person watches, since the offer
only appears for files someone has actually been through. The account is in the
key now. Positions written before this are deleted rather than re-keyed: there
is no record of whose they were, and guessing hands them to whoever signs in
next, which is the bug.
**The staggered rules in the members table.** `display: flex` on the actions
`<td>` — a flex table cell stops being a table cell, so it no longer stretches
to its row and its bottom border is drawn wherever its own content ends.
Measured: in a row whose other cells were `top 76, height 40`, that cell was
`top 77, height 30`, its rule nine pixels above the rest. It is a table cell
again, held open by a zero-width strut so the owner's row — which has no remove
button — stays as tall as the others. Every cell now shares its row's top and
bottom exactly, at 420px and 900px.
**Members became Settings.** It was a list with three unrelated forms stacked
above it, laid out with inline styles on whichever element needed them, and the
group's own controls somewhere else entirely — leaving or deleting a group sat
in the page header beside the title. Now one tab in sections: invitations,
operator pairing, your devices on this node, leaving or deleting, and the roster
last, since it is the only part with no upper bound.
One consequence worth stating: the tab bar no longer waits for the node.
Membership is hub-side, and gating it on a live connection would have made
"leave this group" unreachable exactly when a node is down — which is when
someone most wants it. Files and chat still need the node and say so.
**A download button in the viewer**, beside the close button and in the same
style, for both the video player and the file preview.
844 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
**Full-screen was denied, and the denial was invisible.** The permission
handler was written from a true sentence — nothing here needs a camera, a
microphone or a location — and implemented as `callback(false)` for everything.
Chromium's own video controls ask for the `fullscreen` permission, so a film
could not be watched full-screen.
What made it hard to find, and what the test now pins: **a denied `fullscreen`
does not reject.** `requestFullscreen()` returns a promise that never settles.
No exception, no console message, nothing in the renderer that mentions a
permission — the button just does nothing. Measured rather than reasoned: the
probe reported `NEVER SETTLED` while the main process, instrumented for one run,
logged `PERMISSION ASKED: fullscreen`. After the fix the same probe reports
`granted` with `document.fullscreenElement` set.
The handler now enumerates what is *granted* — `fullscreen`, and nothing else —
so a camera, a microphone, a location, notifications and MIDI are still refused
and whatever Chromium adds next arrives refused rather than quietly allowed.
`Permissions.query` takes the other handler, so both now answer from the one
list instead of eventually disagreeing.
The old test asserted `callback(false)`, which is to say it locked in the bug.
It is replaced by three: what must stay denied, that `fullscreen` is granted,
and that both handlers read the same list.
**"Save automatically" opened a dialog.** The automatic path required a folder
to have been chosen first, and on a new profile nobody has chosen one — so the
very first download fell through to Save As, which is the one thing the setting
promises not to do. A browser does not make you pick a folder before it will
save a file; the system Downloads folder is the answer when there is no other.
Verified on a fresh profile with a home of its own: no dialog, 1024 bytes on
disk, destination reported as the default (`/home/…/Téléchargements` on this
machine, via the localized XDG directory).
A folder that *was* chosen and has since gone still asks. Silently redirecting
those files is worse than a dialog: someone who picked an external drive wants
to be told it is not there, not to find the film in their home directory a week
later. Settings shows the effective destination either way, and offers "forget"
only for a folder somebody actually chose.
813 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
**Downloads were going through RAM.** `_openDownloadTarget` tries a granted
folder, then a service worker, then its floor: collect the whole file in the
page and hand the browser a blob. Both of the first two are absent in the
desktop application — `showDirectoryPicker` does not exist, and Chromium refuses
a service worker on a custom scheme — so every download under 512 MB took the
floor. A gigabyte of film meant a gigabyte of RAM, and the only visible symptom
was a Save As dialog at the *end* rather than the start, which is what the
operator noticed and asked about.
The main process now streams to disk: it honours "save automatically" with a
folder chosen once and no dialog, never overwrites (a colliding name gets a
suffix), awaits each write so the renderer cannot outrun the disk and queue the
file in memory anyway, and unlinks a cancelled download rather than leaving a
truncated file that looks complete to whoever opens it next. Settings now offers
the native folder picker instead of saying downloads are unsupported.
Measured in the running application: the file on disk grows 256 KB → 512 KB →
768 KB → 1 MB as the chunks arrive, and an aborted download leaves nothing
behind.
**A permanent scrollbar on sign-in.** `.layout` and `.page-center` each reserved
`100vh - 52px`, and `.page-center` sits inside `main`'s 24px vertical padding —
so the page overflowed by exactly 48px at every window size. Found by measuring
in the app rather than reading the stylesheet: `scrollHeight` 819 against a 771
viewport, then the bottom edge of every element. The centring page brings its
own padding, so main's is dropped for it and the duplicated arithmetic goes
rather than growing a third term. Now `scrollHeight == innerHeight`, no
overflowing elements.
**The first-run screen was unstyled.** It used a class name I invented
(`auth-page`) that appears nowhere in the stylesheet, so it had no card and the
button sat against the input. It now uses the same `page-center` + `login-card`
markup as sign-in, which is where the 12px gap comes from. The sign-in link in
the nav is hidden until a hub is chosen — it led to a page that could not work.
809 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
All three came from the operator running the application on Ubuntu GNOME. None
would have been found by anything already in the suite.
**A second copy of the hub address.** `keyderive.js` carried
`const HUB = '' // same origin` — true of a page the hub served, false of one
loaded from a package, where the origin is `app://meshbay` and `/v1/users/register`
resolves against the application's own protocol handler. **Sign-up and sign-in,
the first two things anybody does, failed with "Not found."** The seam was
changed in `app.js` and in the signalling call and this was missed: the same
shape as the duplicate `MNP_VERSION` in `protocol.py`, a second copy of a
constant that is harmless until the context changes. `test_hub_address_seam.py`
refuses any file that decides where the hub is, and any `fetch('/v1/…')`
relative to the page origin.
**A window handler reading a variable another path reassigns.** Changing the hub
closes one window and opens another; `closed` arrives *after* the replacement is
assigned, so the outgoing window nulled the reference to the incoming one and
its `ready-to-show` crashed on it — a modal "A JavaScript error occurred in the
main process". Every handler now belongs to the window it was created with. The
CDP test wrote `config.json` in advance, so it never took the one path that
creates a second window; it does now, starting from an empty user-data dir.
**A first run that could not be undone.** The hub address was accepted on
anything URL-shaped and there was no way to change it afterwards — the prompt
only appears when none is set, so a typo meant editing JSON by hand. `https`
typed at a hub speaking `http` produced `TypeError: fetch failed`, which names
nothing. Now: the address is probed before being written, failures say which
URL and why ("does not speak https. If this hub is on your own machine, it is
probably http"), Settings can change it, and Electron's
"Error invoking remote method" wrapper is stripped from what a person reads.
Verified on the operator's desktop: **safeStorage really uses the GNOME
keyring** — Settings reports `gnome-libsecret`, and `secrets.bin` is written
0600 with Chromium's `v11` prefix, the marker for keyring-backed encryption
(the fixed-key fallback writes `v10`). Headless, the same code reports
`unavailable` and refuses to store rather than downgrading in silence, which is
now explained in Settings instead of shown as a bare word.
Unrelated but found while testing: `test_locales.py` assigned to
`globalThis.navigator`, which is read-only from Node 22. The client's build
already requires Node 22+, so the first CI machine configured for it would have
failed these tests for no visible reason. 809 tests pass on Node 18 and Node 24.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
D4. The passphrase stays the account's credential and its only recovery path;
what changes is that it is not asked for on every launch.
**The renderer never holds the device key.** It is generated, stored and used
entirely in the main process, which signs `meshbay:user_auth:<username>:<ts>` on
request. Same rule as the save dialog, for the same reason: the renderer is the
part of this application that parses decrypted content from nodes, which is
attacker-controlled input. And this key is *not* a per-node identity key — those
are generated per node and never leave that relationship, so nothing here
correlates a person across operators.
First run asks which hub, with no default. A client that picks its own hub is a
client that can be pointed at one, and the address is the whole of what this
application trusts a hub for — the interface comes from the package.
Verified against a hub running this code, not against the deployed one:
register 201 → passphrase login 200 → device register 201 → **device sign-in 200
with a real session** → `/v1/users/me` 200 → a stranger's key 401. The
signature was also checked directly against the hub's own Python verifier before
any of that.
Inside the running application, over the debugging protocol: the bridge reaches
the main process, the renderer calls the hub **through it** (200 — the CORS fix
working end to end), and a call to a host that is not the configured hub is
refused.
**Not verified:** safeStorage persisting the key. This session has no secret
service, and standing one up in xvfb did not succeed. The application behaves
correctly there — it *refuses* rather than storing unprotected, and now says so
in Settings, which is a real case rather than a hypothetical one since it is
exactly what a headless or minimal desktop looks like.
Worth remembering for next time: meshbay.org runs whatever was last deployed. It
answered 405 on the Stage-C endpoints and reported MNP 0.2 while the tree had
0.3, so a local `uvicorn meshbay_hub.app:create_app --factory` on SQLite is what
tests hub changes. Nothing was deployed to production for this.
799 tests pass; e2e.py passes end to end.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Electron 42 / Chromium 148, launched under xvfb. The packaged interface mounts
over `app://` with a secure context, `crypto.subtle` present, Argon2 WASM
loaded, and no console errors. Three statements in the design were wrong, and
only launching it found them.
**A CSP in a `<meta>` tag silently drops `frame-ancestors`.** Chromium says so
in the console. A policy carrying a directive that does nothing is worse than
one without it, so the policy is sent as a header by the protocol handler —
which is also the only thing serving the interface, so one source instead of
two.
**`secure: true` is not what makes the service worker register.** Chromium
refuses a worker on a custom scheme whatever its privileges: "The URL protocol
of the current origin ('app://meshbay') is not supported". The application has
no service worker and needs none — it saves through a native dialog, which is
the better of the two paths. `sw.js` stays in the package because the same files
serve the browser, where it is one of only three ways to write a large file.
What `secure: true` is actually for was measured at the same time: without it
**the whole of `crypto.subtle` is undefined**. The first probe loaded a `data:`
URL and every algorithm failed with TypeError, AES-GCM included — which is why
the probe was rewritten before believing its answer. X25519 and Ed25519 are both
present on Chromium 148, settling the version floor left open as O6.
**The renderer cannot call the hub.** Its origin is `app://meshbay` and CORS
refuses it. The hub has *no CORS middleware at all* — its API is reachable from
no web origin whatever — and that is worth keeping. Widening it for
`app://meshbay` would be worse than it looks: that origin is not a credential,
since any Electron application can claim the same scheme and host name.
So every hub call leaves from the main process, exactly as saving a file does,
and it refuses any origin that is not the hub the user signed in to.
`platform.apiFetch()` is `fetch` in a browser and the bridge in the application,
so no caller has to know which it got. `transport.js` reaches it through a
global because it is a classic script, not a module — the alternative was a
second fetch path, which is how two callers of one hub start disagreeing about
how to reach it.
Verified from inside Electron: the main process gets 200 from
/v1/hub/version, the renderer is refused by CORS, and **a script served by the
hub is refused by the policy** — T3's mitigation demonstrated rather than
asserted.
Build note, written into the README because it will bite the next person:
**Ubuntu 24.04's nodejs 18 cannot install Electron at all** — the download
script `require()`s an ESM module, which Node gained in 22. Node 24 LTS,
checksum-verified against nodejs.org, is what this was built with.
package-lock.json is committed; builds use `npm ci`, not `npm install`.
799 tests pass, e2e.py still passes end to end. The session harness needed a
platform stub: it lifts `hubFetch` out of app.js as text and runs it, so the
adapter is now part of the environment it models.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage D, and the honest half of it.
D1 — the seam (done, and verified)
----------------------------------
`static/platform.js`. `HUB` becomes `platform.hubBase()` and the transport is
built with the same base, so one address has one source. In a browser it returns
'' and every path stays relative to the origin that served the page — the
acceptance criterion for this split was "the browser SPA behaves identically",
and it does. `platform.js` joins `_ASSETS`, or a change to it would not move the
content hash and a cached browser would never ask for it.
D2 — the shell (written, never launched)
-----------------------------------------
**There is no npm on this machine. Electron was never installed and
`packages/meshbay-client/` has not been run once.** That is stated here rather
than discovered later.
What is there: a main process serving the packaged interface over a privileged
`app://` scheme (`secure` and `standard` are not cosmetic — without them the
service worker refuses to register and streamed downloads break silently), a
preload exposing an enumerated bridge that never passes a filesystem path, a
window with `sandbox`, `contextIsolation` and no node integration, navigation
away from the package refused, and a CSP where the hub is reachable over
connect-src and is not a script source. The hub address arrives as a process
argument because `platform.hubBase()` runs before anything can await.
`test_desktop_shell.py` pins each of those by reading the source — the treatment
`test_downloads.py` already gives the three browser save paths. It catches a
property being removed and proves nothing about the application running. Two
were checked by breaking them.
The interface is *copied* into the package by `build/sync-ui.js` from the hub's
static directory, and `ui/` is gitignored: a silent fork is the only real way to
end up maintaining the interface twice.
D3 — partial
------------
The bridge, and the part worth having now: safeStorage's backend is reported
rather than assumed. On Linux it falls back to a fixed key when no keyring is
running, silently — someone who believes the OS is holding their keys is told
when it is not. The native key lifecycle belongs with D4 and needs a running
application to mean anything.
D8 — partial, and a real defect found
--------------------------------------
`meshbay-node.spec` installed the SYSTEM template — the one carrying `User=%i` —
into `%{_userunitdir}`. A user unit already runs as its owner and cannot carry
`User=`; systemd refuses the file, so the packaged unit could never have
started. Nothing noticed because nobody had built and installed the RPM.
Two units now: the template to `%{_unitdir}`, and a new `meshbay-node-user.service`
that a person enables themselves without a password — which is what lets the
desktop client install a node without asking for one. It carries ExecReload, so
`meshbay-node reload` does not have to stop a service somebody is streaming from,
and documents the drop-in for a drive outside the home, RequiresMountsFor
included.
798 tests pass; e2e.py still passes end to end. Nothing here was built or
launched: no npm, no rpmbuild.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage C. Identity keys are per node, so a browser and a desktop client are two
keys on one account there — and the node refused the second where it accepted
the first. Without this, an account created natively could never be opened in a
browser without an operator code per node, and "a native client must not prevent
web use" would have been dead on arrival.
Device linking (node)
---------------------
`identities` is keyed by `(user_id, pk_ed25519)` instead of `user_id` alone. The
old shape did `INSERT OR REPLACE`, so a second device overwrote the first
silently; SQLite cannot change a primary key in place, so the table is rebuilt.
Existing pins are carried over — verified against a live roster with 10 of them,
nobody re-pairs.
A new device files a request bound by `sha256(code ‖ its own keys)`, and a key
the node **already pinned** countersigns it. The hub cannot: it has stored no
user keys since 2026-08-14, which is what makes this safe to do without an
operator in the loop.
**The code never reaches the node.** It lists this account's pending requests
with their stored hashes; the approver recomputes and keeps the match. A node
offering fabricated keys would have to produce a hash over a code it has never
seen. Nothing rests on a human comparing digits — that ritual was dropped in
12.1 as "correct, unusable as the default" and must not return by the back door.
The design document had the approver look a request up *by* its hash, which is
circular: computing it needs the keys being asked about. Corrected in both.
Revocation marks rather than deletes, because a deleted row is a key the node
would happily pin again — which is the laptop somebody just reported lost. Your
last device cannot be revoked: coming back would need an operator's code.
Hub — the only change in the whole plan
---------------------------------------
`POST /v1/users/auth` signs in with a device Ed25519 key, on the same pattern as
`/v1/nodes/auth`, plus `/v1/users/devices` to register, list and retire. New
`user_devices` table with an Alembic migration, because `create_all()` is not
one.
This is **not** the key directory that was H3, and the tests say so: nothing
reads it but the hub, no group key is ever wrapped for one, and it is a
different key from the per-node identities. What it does cost is metadata — the
hub now knows how many devices an account has and when each last signed in.
Also `client.minimum` / `client.recommended` in `GET /v1/hub/version`: an
installed client meets a newer hub the day the interface ships in a package, and
that is cheap now and awkward to retrofit.
Browser
-------
The `key_changed` refusal becomes `unknown_device` and offers a linking code
instead of telling someone to find their operator. The Members panel lists this
account's devices here, approves one by code, and retires one.
773 tests pass. `e2e.py` gained a step that links a device end to end against
the live deployment — file, list, recompute, countersign, then open the group
with the new keys and no code — and it also gained `recv_type`, because a step
that assumes the next message is its own answer reads an ack left by the step
before.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage A — a group's content is a set of named roots
---------------------------------------------------
`shared_dir` becomes a list of {name, path, kind}. The name is the directory's
basename, derived once at add time and *stored*: recomputing it would
re-identify a whole library the day someone renames a folder on disk. Duplicate
names are refused case-insensitively and no root may contain another — both
compared with NFC folding, because most of these directories live on exFAT or
NTFS where `Films` and `films` are one directory.
Every index path carries its root name, in a one-root group as much as in a
five-root one. One path shape has to be got right once; two have to be kept
right for ever.
**A root that goes away freezes; it never empties.** Unmounting a volume makes
watchdog report every file under it as deleted, or presents an empty directory
to the next scan. Acting on either propagates deletions for a whole library to
every member, as though the owner had erased it. So a deletion is acted on only
once its root is confirmed readable, and availability is tracked per root — one
unplugged drive leaves the others serving. 12 tests, verified to fail against an
indexer without the check.
Events are not trusted to be complete either: ReadDirectoryChangesW drops them
under load and inotify on a FUSE mount misses changes made outside it. A
periodic reconciliation sweep is the only thing that recovers a missed event.
MNP 0.2 → 0.3 (additive). The hub needs no change: SwarmSource carries a content
hash, a node id and an endpoint — no paths, no filenames — and private groups
register nothing (H7).
Stage B — one implementation behind every front door
----------------------------------------------------
C1 and C6 were both "a second path into the node with its own weaker
handshake". Two implementations of `revoke` with two authorization checks is
that shape one size down. `meshbay_node/ops.py` holds each operation once,
takes the daemon state, and knows nothing about HTTP, argv or MNP. The loopback
API is one `_op(...)` line per endpoint; the MNP handlers call the same
functions. test_ops.py asserts the shape rather than trusting it.
Phase 14 is finished on top of it — `group list`, `gek init|rotate`, `reload`
(SIGHUP), `denylist show|clear`, `file list|rm`. **No operator action requires a
browser any more.** Plus `gek_rotate` and `member_unpin` as operator-signed MNP
operations: rotation is the half of revocation that revocation cannot do, since
the ex-member holds the current key, and the node generates the replacement
with its own CSPRNG — no key material crosses the wire, which is what the C5b
rule is actually about.
Two bugs found by running it rather than by testing it
------------------------------------------------------
GroupIndex is keyed by **content hash**, so the same bytes at two paths are one
entry — which is also why a scan reports ten files and indexes nine.
Reconciliation compared paths, so it decided the second path was a missed event
every 60 s, rewrote the entry and pushed an index update to every connected
peer. Seen in a live node's log.
`meshbay-node reload` crashed on first use with `subprocess` unimported: the
module compiles fine, which is the "syntax, not names" trap already recorded for
the SPA. test_cli_dispatch.py now walks every verb and refuses to let one be
added to the parser without an entry there.
Also corrected: protocol.py declared a second MNP_VERSION of "0.1" while the
wire carried "0.2" — harmless only because nothing imported it. And
_do_dir_create/_do_dir_delete referenced an undefined `filename` on their error
path.
740 tests pass; QE/deploy/e2e.py passes end to end against the live deployment.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported: after an hour of watching a film, every action answers "token expired
or invalid", with signing out and back in as the only way on. Reopening the tab
the next day did the same.
The access token lasts an hour and the refresh token thirty days, and nothing
used the second one. `hubFetch` reported a 401 like any other error, and
watching a film is precisely an hour in which the hub hears nothing at all,
because the video travels over WebRTC. So the token aged out with no request to
notice, and a tab reopened the next morning presented a stale token with a
perfectly good refresh token sitting beside it in localStorage.
Underneath was the reason it could not be recovered from. The hub *rotates*:
the refresh endpoint revokes the token presented, returns a replacement, and
treats a revoked one presented again as theft, revoking the whole family. The
client kept only the access token out of that response. So the refresh token was
spent on first use and the second attempt did not merely fail — it destroyed the
family. Which is exactly the reported symptom.
Renewal now happens on a margin, on returning to the tab, on mount, and on a 401
with the request replayed. Concurrent renewals share one request: two 401s
racing would otherwise present the same refresh token twice, and the hub cannot
tell that from theft, so the remedy would have been worse than the fault. A
refusal signs out cleanly rather than leaving a session that fails every call
while looking signed in.
The lifetime goes to four hours, which is not what makes long sessions work —
renewal is — but is what someone has to notice by if renewal itself breaks. An
hour was less than a feature film. Twelve was considered and declined: it widens
the window in which a leaked token cannot be turned off, and it lets the renewal
path go a whole day between uses, which is how it came to be broken here without
anyone noticing. Production sets this in its own hub.toml, so both moved.
The tests run the shipped code against a hub that enforces rotation, because a
stub that accepted the same refresh token twice would have passed against the
broken client. Checked that dropping the rotated token reproduces the revoked
family, so the guard is guarding something.
Also widens the orphan-setter rule to ignore `setX` functions declared in the
module: `setAuth` is not a hook setter, and a rule that cries wolf is one
somebody eventually silences. Verified it still catches a real orphan.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported: on mobile you see only the right-hand edge of the panel, without the
content. Measured, before anything was changed:
320 px viewport -> panel at -138..192, 138 px off the left
360 px -> -98..232
412 px -> -46..284
The panel is 330 px wide and anchored to the right edge of its button — but
that button is not at the right edge of the screen, since the bell and the user
menu come after it. What falls off is the left-hand side, which is where the
file names are, so what stayed on screen was a strip of progress bars belonging
to nothing.
Narrowing it would not have helped: the overflow comes from where the right
edge is pinned, not from the width. Below the existing 768 px breakpoint the
panel is anchored to the viewport instead, full width on a phone and capped at
420 px on a tablet, where stretching two filenames across 750 px would be
silly. Desktop keeps its 330 px against the button.
The interesting part is how it was found. The responsive tests read numbers out
of the stylesheet and said, in their own docstring, that a layout could not be
measured because the suite had no browser. It has one now — Chrome, from the
video work — so layout_probe.py renders the real stylesheet at a given width and
returns rectangles. `width: 330px` was never the thing worth asserting on.
An iframe carries the viewport, because a headless window will not go below
about 500 px, and one browser measures every width: launching one per test put
three minutes on the suite against twenty-six seconds for all of them. Checked
that the new tests fail with the rule removed — three of them do — and that
they pass with it back.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clicking the scrubber now restarts the stream there. A seek inside what is
already buffered never reaches the node. Dragging is debounced at 350 ms,
because `seeking` fires continuously and each one acted on kills an ffmpeg and
spawns another.
`SourceBuffer.mode` changes from 'sequence' to 'segments', and each stream sets
`timestampOffset` to where the node says it started — otherwise a stream
beginning at forty minutes is buffered at zero and the scrubber lies about
everything. Proved in Chrome before any of it was written: abort, remove, offset
600, append, and playback resumes at 600 with `readyState` 4.
A seek makes the buffer discontinuous, and everything reading `buffered` then
has to mean a particular range. "The last one" stops being "the one playing":
the read-ahead would report a full buffer across a gap while the player starves,
and the eviction would take out everything between the first range and the
playhead, including what is showing. Both work from the range around the
playhead now. `reinitAt` also clears the buffer outright rather than keeping two
ranges, and calls `abort()` first — ffmpeg was killed mid-fragment, so the
parser holds half of one and the next header would land on top of it.
The position is kept in localStorage, per file: no protocol, no storage anyone
else has to keep, and nothing new learns what you watch. Not below thirty
seconds, not past 97% of the film, and a pill offers the beginning back.
Two things this had to get right and did not at first, both found by running it
rather than reading it. Between asking for a seek and its `stream_init`,
everything on the channel belongs to the film being left — same file, so
`file_id` cannot separate them, but ordering can. And a segment that arrived is
no longer in flight whatever is then done with it: discarding one before
decrementing the window leaked a slot every time, and `reinitAt` is
asynchronous, so a whole window could arrive while it waited. The player then
believed eight segments were in flight, granted nothing further, and the node
waited for credit that could not come — a race, which is why the same seek
worked twice and hung on the third.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The scrubber was drawn the length of the film — `ms.duration` has always been
the real duration — and then `onSeeking` quietly clamped every target back into
whatever happened to be buffered. The bar invited a click and refused it.
`stream_req` gains a `start`. The session's previous stream is retired by the
path that already exists for switching films, and ffmpeg is spawned again with
`-ss` **before** `-i`: an index lookup rather than decoding and discarding up to
the point, which is milliseconds on a 500 MB film instead of tens of seconds.
Measured over real MNP: 0s -> 492 MB, 600s -> 418 MB, 3000s -> 179 MB.
A seek at or past the end is pulled back, because ffmpeg would produce nothing
and the player would wait for segments that are never coming.
`stream_init` reports the position actually used. It has to: ffmpeg restarts its
output timestamps at zero however far in it seeks — `-copyts` does not change
that for this input, measured — so the client is the one that puts the fragments
back on the film's timeline, and it cannot guess by how much. The value is also
not what was asked for, since `-c copy` lands on the keyframe at or before it.
The diagnostics that found the rest of this are here too: a seek, a first init
and a re-init are each one line at INFO, which is rare enough to keep on. The
five-second client report stays at DEBUG.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A 500 MB film loaded about 100 MB and hung on "buffering" for good. 100 MB is
not a number in our code: it is where the browser stops. ffmpeg remuxes with
`-c copy`, so the bytes on the wire are the file's own, and credit granted per
append meant taking them as fast as the network allowed — which for a film is
very much faster than watching it. The SourceBuffer ceiling arrived in the
first minute.
Past it every append was refused, and the refusal was unrecoverable: a refused
append fires no `updateend`, `updateend` was where credit was granted, so the
node sent nothing and no segment arrived to retry the append. Every wakeup the
pipeline had was downstream of the append that had just failed. Playback
continuing — the one thing that frees room — woke nothing at all.
Credit now follows the buffer instead of the writes. `pump()` is the only
place it is granted, it keeps `STREAM_WINDOW` segments in flight while less
than `BUFFER_AHEAD_S` of film is held past the playhead, and it is driven by a
one-second clock and by playback progress, never by arriving data. Buffering
by time makes a two-hour film cost what a two-minute clip costs.
A window rather than a debt, and this took a second measurement to get right:
accumulating a credit per append and releasing the balance when the buffer
finally drained sent six megabytes in one burst, overshot by a minute of film,
then said nothing for forty-six seconds. Measured in Chrome against real
fragmented MP4.
Two smaller things found on the way. `updateend` fires for `remove()` as well
as `appendBuffer()`, so crediting from it paid the node for the player's own
evictions. And a viewer that is deliberately far enough ahead grants nothing
for minutes, which the node read as a closed tab — it now sends `stream_more`
with n=0, which grants no room but proves someone is there.
The first version of the test modelled the credit loop and passed while the
player still hung: a model written by whoever wrote the fix agrees with it by
construction. `tests/harness/mse_harness.mjs` lifts the real functions out of
app.js as text and runs them against a SourceBuffer that has a ceiling. What is
modelled is the browser.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Chat opens on the newest hundred messages, loads fifty older on demand with the
reading position anchored — the distance from the *bottom*, since everything
above the viewport just grew — and follows new messages only when the reader was
already at the end. Day separators, sender grouping, an unread marker, and a
jump-to-latest pill. Messages are keyed by id: index keys plus prepending makes
Preact reuse the wrong bubbles.
A presence dot per group in the sidebar, three states, each backed by something:
the hub's registry, or a connection this browser made or failed to make. Never
colour alone — red and green are the pair colour-blind readers cannot separate —
so each dot carries a title and an aria-label.
Profile is split out of Settings: identity, node link, pinned node identities and
account deletion. Mixing them put an irreversible button two scrolls under a
theme picker. The create-group page loses its centred 520 px card, which left
190 px of margin either side, and its two button panels become a radio group —
a button conveys no chosen state to a screen reader, and side by side they read
as two independent actions rather than one either/or.
The Files toolbar shows its actions as icon buttons the moment Select is on,
disabled when they do not apply rather than appearing and vanishing. On a phone
the right-hand group could not wrap and ran 130 px off the screen.
Streamed downloads no longer freeze after one chunk. `registration.active` says
a worker exists, not that this page is controlled by it — and an uncontrolled
page's requests never reach its fetch handler, so the worker took the stream and
was never asked for it, leaving `writer.write()` waiting on backpressure that
would never lift. The page now requires control and the worker confirms it
actually served the request before the sink is trusted.
Fixed on the way: `setActionsOpen` outlived the state it belonged to and threw
on every Files action; the chat scrollbar stopped short of the bottom; the
owner's row sat lower than the rest; About showed a version hardcoded two
releases ago.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
French, Spanish, Brazilian Portuguese, Simplified Chinese, Japanese,
German, Italian, Dutch and Polish, all in the formal register. `hub`,
`node` and `GEK` stay in English: they name the CLI, node.toml and the
docs, and translating them would cut the interface off from everything an
operator reads and types.
Catalogues move out of i18n.js into locales/, one file per language,
fetched with a dynamic import. A visitor downloads their language plus
English as a fallback — about 36 KB rather than the ~180 KB that ten
inlined catalogues would have cost everyone. i18n.js keeps only the
loader, so the first render now waits for initLocale().
Three things the old code got wrong, none of them visible until there was
a second language:
- Resolution trimmed a tag to its base before matching, so a browser
reporting pt-BR looked for a `pt` catalogue that does not exist and fell
back to English. Matching is now exact first, then by base language.
- Counted strings were single strings, so Polish could not express
1 plik / 2 pliki / 5 plików at all. t() selects through
Intl.PluralRules; en.js gains the same treatment, which incidentally
fixes "1 files".
- Interpolation used String.replace, which reads `$&` in the replacement.
A file named rap$&sody.mp3 rendered corrupted in its own delete dialog.
Five strings were still hardcoded in app.js — the group name placeholder
and the four visibility/join-policy descriptions — and are now keyed.
test_locales.py holds the nine translations to the shape of en.js: same
keys, a counted string stays counted everywhere, every plural entry covers
each category Intl actually produces for that language, and the
{placeholders} survive translation. Verified failing first, against a
catalogue with a key removed, a placeholder dropped and the Polish `few`
form deleted.
The language menu also grew from one entry to ten, which overran a short
viewport inside a dropdown that clipped instead of scrolling.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
**Closing the viewer left the node working.** Nothing told it to stop:
the player dropped its handlers, which only made the browser deaf. ffmpeg
kept running and held one of the node's two transcode slots until the
credit timeout expired two minutes later — which is why the next video
answered "server busy". `stream_stop` ends it at once, and the viewer
also drops its queue, ends the MediaSource and revokes the object URL on
the way out, any of which could be holding megabytes of decrypted video.
While there: `file_chunk` replies were matched to their requests by
arrival order, which was true by luck rather than by construction. The
reply now names the file it belongs to and is matched on that and the
chunk index; a chunk nobody is waiting for is dropped instead of being
handed to whatever request happens to be oldest.
**The administration panel counted its own history.** A deleted account
is tombstoned so the connection log stays readable, and every count and
list treated that row as a user — including a group's member count, and
the member list of the group itself. They do not any more.
**Where a node is.** `endpoint_hint` is what a node believes its address
to be, learned from a STUN server and sent to us: useful for reaching it,
and a claim. The announcement that carries it is signed with the node key
over a fresh timestamp, so the address that request *arrives from* is the
address of whoever holds that key — that is now recorded on the node row
and shown in a Nodes tab, next to the hint, with the difference spelled
out. Clients get the same treatment: `webrtc_offer` is logged with the
address the hub saw when a browser starts a peer connection.
Verified against the live deployment: the node's row reads 90.112.206.172
after a restart, and in e2e a stopped stream goes quiet in one message
and the next one starts immediately instead of being refused.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
**Removing a member.** The owner can do it from the Members tab, and it
is two halves in the order that fails safe: the node stops serving the
group key first (an operator-signed request, so a paired browser only),
then the hub drops the membership row. The other order would leave
someone able to reach a node that still serves them.
It is a membership, not an account. The user row is never written: their
other groups, their files and their pinned identity survive, because one
group's owner must not be able to erase someone from the hub. It is also
per group — a node hosting two loses them from one — and it does not take
back the key they already unwrapped, which is what rotating the GEK is
for. The confirmation and the panel both say so.
**Downloads and streaming through the disk, in both browsers.** The audit
this started as found two ways to put gigabytes in a tab.
Firefox and Safari have no File System Access API, so every download
there was collected in memory. A service worker fixes it: the page keeps
the writable half of a transferred stream, the worker answers a made-up
URL with the readable half and a Content-Disposition header, and the
browser writes it to disk as it arrives, with real backpressure. The
worker caches nothing and falls through on every request that is not one
of these downloads. A zip announces no Content-Length, since the archive
is larger than the files in it and a length we miss truncates the file.
Video was worse and affected both browsers. The node pushed ffmpeg's
whole output as fast as it was produced while the player consumed a
segment at a time, so the queue held the film — and appending all of it
hit the SourceBuffer's cap, where the handler logged the error and
dropped the segment, leaving a hole in the middle of the film with
nothing to show for it. Streaming is credit-based now, 24 segments of
256 KB in flight, verified against the live node: three credits, three
segments, then silence until more are granted. The player evicts what is
more than a minute behind the playhead and retries a refused segment
rather than dropping it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
of it
Two bugs in what shipped last, and both were mine.
Automatic mode still opened Save As, because with no folder granted the
code fell through to the file picker — while the documentation said it
would use the browser's own download folder. It does that now. Over
512 MB it still asks, since getting there means holding the file in
memory and a tab will not survive a 40 GB blob; Settings is where to stop
it asking again.
Selecting two files downloaded one. They were started without awaiting,
so each asked the browser for a save dialog at once, and a browser allows
exactly one — the rest were rejected and the errors went nowhere. They
are awaited one at a time now, which serializes the dialogs and not the
transfers: each call returns as soon as its transfer is registered.
Then the adjustments. The transfers widget offers Open on a finished
download that went into a granted folder — the bytes go to a new tab, and
that is the whole of what a page can do: no browser lets one start a
desktop application or show a file manager, so the folder half of that
request cannot be built and the guide says so.
The Files toolbar was four controls of three different heights in a row.
It is three groups now — what you can add, where you are, what you can do
with what is here — on one baseline, with icons from the set and a gap
between the dots and the word Actions. Chat comes first among the tabs
and is the one you land on. The three Discover entries in the sidebar
have icons. And a link in a chat message becomes a link: built as an
element and never as markup, http and https only, so `javascript:` is not
one message away from running here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Downloading a selection of twenty files meant twenty Save As dialogs,
which is the wrong answer for the feature that had just been built.
Settings → Downloads now offers saving automatically, and that is the
default; asking every time stays available for people who want it.
The correction worth recording: a web page cannot be given a filesystem
path and cannot read one either. There is no ~/Downloads to configure and
nothing to type, on any operating system — which is also why none of this
will need changing on Windows. What a browser grants is a handle to a
folder the user picked in a dialog, so that is what the setting keeps:
picked once, stored in IndexedDB, re-confirmed once a session because the
grant comes back as a claim rather than a permission. Where no folder has
been granted, and in Firefox and Safari where none can be, files go to
the browser's own download folder — which on most machines is the folder
that was meant all along.
Automatic saving has one risk a dialog does not: it can silently replace
a file. It does not — a taken name gets a suffix before the extension,
`clip (2).mp4`, so a download folder does not fill up with files the
system no longer recognises. That, and the default, are what
test_downloads.py pins.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
per-row menus
Downloads and uploads were state inside GroupPage. Leaving a group
unmounted the component, its cleanup closed the DataChannel, and a
half-written file was all you had — which is also why only one thing
could be in flight at a time.
They live in a module-level store now. A group page hands its transport
over on the way out rather than closing it, and the last transfer using
it closes it; signing out is the one thing that cancels everything,
because those transfers are moving data on a token about to stop being
ours. The store is plain JavaScript with no browser globals, so
test_transfers.py runs it under Node and pins the parts that are timing
and lifetime rather than markup: that a cancel stops the work instead of
greying out a row, that a stalled transfer reads as stalled rather than
reporting its own historical average, and that a released transport is
closed by the last transfer and not before.
The widget by the bell shows each transfer with its rate and a cancel
button, so the Files panel no longer carries progress bars — you can
watch a 40 GB archive from the chat, or from another group.
Selection replaces the per-row menu: a Select toggle puts checkboxes on
files and folders, and ⋮ Actions acts on what is ticked. Ticks survive
walking into another folder, so a selection can span directories.
Downloads start together and run together. Videos offer Play only — View
did the same thing, which is the sort of duplication that makes people
wonder what the difference is.
Uploads had to become parallel-safe for any of this to mean anything:
their acks were matched by arrival order, so two at once credited each
other's progress. The node names the file in every ack, so they are keyed
by name now — with the same file twice refused, since the node keys its
own upload state that way too.
Two mistakes worth recording. The selection column went into the body
rows and not the header, because that edit matched nothing and I had not
made it assert; the columns were misaligned until a screenshot showed it.
And the Actions menu opened leftwards from a button at the right edge of
the toolbar, half of it off-screen.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Two things a Files panel needs and did not have.
**Removing a directory** is privileged, where creating one is not: it
acts on a name other members are using, on the operator's disk. It is
refused unless the directory is empty, and that rule is the safety
property — whatever the browser sends, this cannot destroy content. The
check runs twice, once before the challenge and once after the signature
comes back, because a file can land during the round trip. A file also
accepts its uploader's key; a directory has no uploader, so only the
operator's key will do.
**Downloading a folder** produces a zip built in the browser, written
straight to disk as the chunks arrive. An archive of a group folder is
routinely tens of gigabytes, so nothing is held: peak memory is one chunk
plus a small record per file. The node is not involved at all — it serves
the same encrypted chunks as any other download, holds no temporary
files, and cannot be asked to compress anything.
zipstream.js is store-only. Group content is video and images, already
compressed, so deflate would spend CPU on every byte to save nothing, in
the thread that is also decrypting. Sizes and CRCs go in a data
descriptor after each file because a stream cannot seek back to patch a
header, and zip64 kicks in per entry past 4 GiB and for the archive
itself. Because none of that can be checked from the Python side of the
house, test_zipstream.py runs the real module under Node and reads what
it produces with zipfile — CRCs, UTF-8 names, zip64 records and all. The
archives also pass `unzip -t`.
Firefox and Safari have no File System Access API, so there is nowhere to
stream to: the fallback builds the archive in memory and says so, with
the size, before starting rather than after failing.
One mistake worth recording: the first version of deleteDirectory passed
the node's own answer as the value to check the challenge against, which
turns the comparison into a tautology. It checks the path we asked for.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The uploader read a 48 KB slice, sent it, and waited for the node to
acknowledge it before reading the next one. That caps throughput at one
chunk per round trip regardless of available bandwidth, and it is worse
than the arithmetic suggests: the sender is idle for almost the whole
time, so SCTP's congestion window never opens either, and the transport
stays slow even when the link is not.
Measured against the real node over a 100 ms path (netem on loopback):
48 KB chunks, one at a time 0.16 MB/s
48 KB chunks, 32 in flight 3.47 MB/s
On loopback with no latency both are ~32 MB/s, which is why nothing here
ever caught it: the local end-to-end run cannot see a round-trip problem.
transport.uploadFile() now keeps a window of chunks in flight and matches
acks by arrival, with the node's own ordering rule as the guard — a
DataChannel is ordered and reliable, and the node refuses any chunk that
is not the one it expects next. It pauses when the channel's buffered
amount gets high, so the progress bar keeps reporting what the node has
taken rather than what the browser has queued. Both callers, the Files
panel and chat attachments, go through it.
The end-to-end harness grew an opt-in benchmark behind MESHBAY_BENCH=1
that removes its own files afterwards, and it taught me something about
the harness rather than the code: it took an unsolicited index_sync push
for an upload ack, because unlike app.js it had no place to put one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A description could only be set the moment a group was created, so every
group made before anyone thought of one stayed blank for good. The owner
can now edit it from the group's page, and PATCH /v1/groups/{id} takes it.
That endpoint takes the description and nothing else, deliberately. The
name, the visibility and the join policy are the terms members joined on;
a private group that can quietly become public is not the group they
agreed to be in. Changing those needs a decision about who gets told, not
a field on a form — there is a test saying so.
Separately, the legacy operator key is gone. `admin_pk_ed25519` in
node.toml named the operator before the roster existed and was kept so
that an existing deployment would keep working; nothing uses it, and a
second source of node authority is not something to carry around out of
politeness. Authority is the roster, read fresh on every check.
It is removed rather than ignored: a config that still names the key gets
a warning at startup pointing at the file. Dropping it in silence would
refuse invites and file deletion with a signature error that looks like a
bug somewhere else — which is exactly how finding M3 presented.
Two tests were verifying admin operations by naming a key in the context,
which was the legacy path. They now pair an operator into a roster, the
way an operator does. The authority test anchored on the deleted function
and passed vacuously once it disappeared; it states the invariant against
the verifier and the daemon instead.
Also defined .btn-secondary, used in four places and styled in none.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Four small things, two of which were the same bug wearing different hats.
The index cache seeded the Files panel and then raced the live index:
IndexedDB is async, so a fast node could hand you the real list and have
it overwritten a moment later by the cached one. That is the "choses
bizarres". The panel now shows what the node says, or says it cannot
reach the node — no third state that looks like data but is memory.
The cache stays, written on every sync, and the search page is the only
thing that reads it. Search across groups has no other source: it cannot
connect to every node to answer a keystroke. So it now says how stale
each hit is — "synced 2 hours ago", per group — and a line under the
results explains that opening a group refreshes what search knows about
it. Deleting a file also rewrites the cache now; it used to refresh the
table and leave the cache holding a file that no longer existed, which is
why search kept offering it.
The invite form is shown only once this browser holds an operator key.
Invites are signed with it and the node checks the signature against its
roster, so an unpaired browser could fill the form in and fail on submit.
An owner who is not the node's operator is told to ask the one who is.
Group descriptions now show on the home cards, as they already did in
Explore.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
pairing form
Moving the invite form above the member list cut both out of MembersPanel and
pasted them into AdminPage, where `doInvite`, `members`, `adminId` and
`inviteCode` do not exist. A standard member saw an empty Members tab, the
group owner saw only a pairing form, and the hub's own Users tab referenced
four undefined names.
The pairing form outstaying its welcome is a second bug and an older one.
`is_node_admin` compares the connecting account with the account that owns the
node — it says nothing about whether *this browser's key* was ever paired, which
is the thing pairing changes and the thing that lets you sign an invite. So the
form showed for an operator who paired months ago, accepted a fresh code,
reported success, and stayed exactly where it was. The node already reports the
roster role in `join_result`; the transport keeps it, and the form appears only
when this identity is not an operator key yet.
Also dropped a clause from the pairing hint: the code never passing through the
hub is worth saying, the theory behind it is not.
test_spa_ordering.py gets three checks for this class of bug — a cut-and-paste
between components is invisible to every other test we have.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The site drew its icons with colour emoji, which means a different
drawing on every operating system and never the same line weight twice —
except the Administration entry, whose shield is a text-default glyph and
so rendered as a thin outline. That one was the odd one out and also the
one that looked right.
Fifteen inline SVG paths now replace it: menu, bell, shield, globe, gear,
sun, moon, power, lock, envelope, door, clip, check, chevron, close. They
are stroked in currentColor and sized in em, so an icon takes the colour
and size of the text beside it and needs no rule of its own — the caret
in the dark nav bar and the gear in a light dropdown are the same file.
The explorer keeps its emoji. There the icon says what kind of file this
is and the colour is doing real work; a wall of identical grey outlines
would be a worse file list.
One bug fell out of the change: .nav-notif was painted with --text, the
page's text colour, on a nav bar that is nearly black. The emoji bell
carried its own colours and showed anyway; a stroked one was invisible
until it was given --nav-text.
Checked by rendering the real stylesheet in headless Chrome, light and
dark.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Four things were wrong, and they compounded: a busy chat produced one row
per message, muting a group did nothing at all, there was no way to clear
the list, and the one person guaranteed to know about a message — its
author — was told about it.
The author bug was a name mismatch across two processes. The node sent
chat_notify without saying who wrote the message, so the hub used the
node's own token subject, which is the operator's account. The skip
therefore matched the operator and no one else: everybody was notified of
their own messages, and the operator was notified of nobody's. The node
now names the author and the hub reads that field.
Muting lived in the browser's localStorage and nothing ever read it, so
the checkbox was decoration. It is a column on group_members now, checked
where the notification is created — a notification nobody wants is not
written at all.
Chat keeps a single row per (user, kind, group) whose date moves and whose
read flag clears, so a conversation is one line saying when it last spoke.
Clicking it opens the group and dismisses it; joining a group dismisses
its invitation; and DELETE /v1/notifications clears the lot.
The hub deploy now runs alembic. create_all() only creates missing tables,
so group_members.muted never arrived on the running hub and /v1/groups/mine
answered 500 — worth catching in the script rather than in a browser.
Verified end to end against the deployed hub and node: the author receives
nothing, the other member receives exactly one, carrying its group_id.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Both go through the same erasure, so there is one description of what happens
rather than two that drift.
Gone: credentials, email, node key, group memberships, notifications, refresh
tokens, node registrations. The username is released.
Kept, on purpose and stated in the UI: the row itself, emptied, and the IP log
that points at it. Those logs exist for a year to answer legal requests, and a
log that can no longer say whose connection it recorded keeps the data while
losing the only thing it is for. So the account becomes a tombstone rather than a
hole in the table.
Out of reach, also stated: files uploaded to nodes, and the identity keys nodes
pinned. Those are on machines the hub does not command, and only their operators
can remove them — `member unpin` and a delete on their own disk. Saying so in the
confirmation matters more than the button.
Owning groups blocks deletion, with the list. Cascading would delete other
people's groups out from under them; the account holder can hand them over or
delete them first, deliberately.
Self-deletion re-checks the passphrase. A live token may be a borrowed laptop or
a tab left open, and it is not consent to something irreversible. Admin deletion
requires admin rather than moderator: suspension is the reversible moderation
tool and stays one click away.
A deleted account's access token stops working at once — the status check already
refuses anything but "active", which the tests now pin down, because refresh
tokens being gone would otherwise leave up to an hour of usable session.
Tests: 8 covering what survives and what does not, plus a db_session fixture for
assertions that cannot honestly be made through the API.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Correction to the previous commit. Uploads went wherever the member happened to
be looking, which spreads chat attachments through the tree and makes the
destination a client-supplied path — surface that had to be defended. Everything
a member sends now lands in `uploads/` at the root of the shared directory:
visible, one place, easy for the operator to look into or empty.
Chat attachments go there too, so the separate out-of-tree thumbs directory is
not needed and is not built. They were already ordinary uploads; now they are
ordinary uploads that land somewhere sensible.
The destination is chosen by the node, so a client naming somewhere else changes
nothing — the traversal surface simply is not there on this path. safe_subdir()
remains for dir_create, where the path genuinely does come from the client, and
keeps its tests.
One shared directory means name collisions are ordinary rather than adversarial:
every camera produces IMG_1234.jpg. The node finds a free name — "IMG_1234 (2).jpg"
— and reports it in the ack, because a chat message has to point at the file that
was actually written and not at someone else's. Nothing is ever replaced, which
is the property the per-user quarantine existed for (C5a) and the one the tests
assert; they fail if the free-name search is removed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The per-user quarantine is gone. `.uploads/{user_id}/` was the fix for C5a, and
it worked, but it made the shared directory something nobody could organise:
every file landed under a uuid nobody recognises. Files now go where the member
is looking, most often the root.
What the quarantine actually bought is kept, and is now what the tests assert
rather than the location:
- an existing file is never replaced. That was the real defect — overwriting a
file also made the attacker its recorded uploader, and therefore able to
delete it through the uploader path
- the name allowlist is unchanged
- the destination is confined under the shared root
That last one is new surface: the directory arrives from the client. safe_subdir()
is the single place that decides, with two independent guards — every segment
against the name allowlist, and the resolved result under the root — because one
of them will eventually be refactored by someone who does not know why it is
there. Ten traversal cases are covered, and they fail if both guards go.
Also adds `dir_create` (any member may organise a shared directory; audited like
anything that writes to the operator's disk) and makes the node report its real
directory list in index_sync — folders were inferred from file paths, so a new
empty one, or one that had been emptied, simply did not exist as far as the UI
was concerned.
Two C5a tests changed their assertions deliberately, as C5b's did before: they
encoded the quarantine path, which is the thing being removed. The property they
existed for is asserted more directly than before.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Three of the ten UI items, the ones that needed no decision.
Upload showed a disabled button and nothing else — on a large file that is
indistinguishable from a hang. It now has the same bar downloads have, filled
from chunks the node has acknowledged rather than bytes read locally, and says
"indexing" for the pause at the end: the node re-indexes on a filesystem event,
so there is a real wait there with nothing to poll.
PDFs open in the overlay through the browser's own viewer. The file is decrypted
in the page as any other preview is, and shown from a blob: URL — nothing leaves
the tab, and no external viewer is involved.
The invite form sits above the member list, where the action is rather than after
the thing it acts on.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
One keypair was copied to every node its owner joined, so cracking the bundle on
any single node yielded the identity used on all of them: their content on other
operators' machines, and the ability to sign as them anywhere. That lateral reach
was the part of C4 worth attacking.
Each node now gets its own keypair, generated the first time its owner joins it
and left with that node alone. An operator who cracks what sits on their own disk
holds a key that is a stranger to every other node — and on their own node, one
that unlocks nothing they did not already hold: they serve the content, the index
and every byte of it by design.
Nothing changes for the user. A first contact with a node already needed that
operator's code, and the key is created in the same step; a second browser still
recovers it from the node with the passphrase alone. Two operators can also no
longer tell they host the same person by comparing keys.
BREAKING, and deliberately without a compatibility path — the deployment is wiped
for the next demo:
- users.pk_ed25519 / pk_x25519 dropped (migration a7c31f9e40b2)
- registration no longer sends or stores a key
- PUT /v1/users/me/keys and regenerateKeys() gone; rotation is now
`member unpin` plus a fresh code, decided on the machine that pinned it
- /pubkeys returns an account id and a node's linking key. It was the directory
H3 read, and nothing wraps for it any more
- the pk_user JWT claim is gone
That last one closed a live defect the inventory turned up: the node recorded
pk_user as the uploader's identity and authorized deletion against it, so a hub
issuing a token naming its own key could delete anyone's uploads on any node.
Attribution now uses the key the node itself pinned.
A simplification falls out. Registration generates nothing, so a scripted signup
is a real account: `demo.py bootstrap` takes a wiped hub and node to a working
demo with no browser, which was impossible while keys were born in one.
Also fixes, found by running it on a wiped deployment: the key handed back on a
join now belongs to the group the connection is for, not the group named in the
invitation — an operator pairs node-wide but redeems the code while opening a
group, and expects to read it.
Tests: 343, including the two that state the property — a key pinned by one node
is refused at another, and someone else's code does not admit it. Verified end to
end against a wiped hub and node: bootstrap, pair, invite, join, download,
stream, second browser, revoke.
Design: docs/per-node-identity-v1.md
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Argon2id memory 64 → 128 MB. Memory is the lever, not time: it caps how many
guesses a card can hold at once, so the ceiling on one high-end GPU moves from
roughly 4k to roughly 2k guesses/s and its 24 GB fits ~187 lanes instead of ~375.
Measured through the vendored build: 640 ms, against 322 ms at 64 MB.
While measuring the real cost of a sign-in, found the SPA deriving the bundle key
twice — once for the key pair kept for the session, then again inside
decryptBundle() for the local bundle. At these parameters that is 0.6 s of pure
waste. Measured now, end to end:
auth_key (PBKDF2 600k) 239 ms
bundle v1 (PBKDF2 600k) 240 ms legacy, until every bundle is upgraded
bundle v2 (Argon2id 128MB) 650 ms
-----------------------------------
sign-in 1 129 ms (889 ms once no v1 bundles remain)
Once per sign-in, and only then: reopening a group, downloading, streaming and
reloading the page all reuse the key, which lives in IndexedDB from login.
Also bounds two waits in the node's hub WebSocket, found because the node went
silent again mid-deploy. It had reconnected after the hub restart, sent its auth
frame, and waited for a reply that never came — `ws.recv()` had no timeout, so a
hub that accepts a socket and then says nothing for a few seconds while starting
up parks the task forever: node running, logging nothing, invisible to everyone.
The auth exchange now times out at 15 s, connect at 15 s, and a refused auth
retries with a fresh token instead of ending the task for good.
QE harness signs in once per account and reuses the token — several clients there
stand for several browsers of one person, and what tells them apart is which keys
they hold, not which token, while the hub quite rightly rate-limits repeated
logins from one address.
Tests: 341, plus the live workflow.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Two corrections to yesterday's judgement, in the order they matter.
**The toggle is gone.** Asked to make the remote key backup optional, I shipped a
setting whose "off" position meant: no second browser, ever, and clearing your
storage destroys the account. I wrote the warning that says so without drawing
the conclusion. A control whose only effect is to break the ordinary case is not
a control, and removing an exposure by removing the feature is not a fix. Every
browser backs its keys up again, unconditionally.
**The exposure is fixed where it actually lives: the KDF.** The keypair bundle
rests on every node whose group its owner joins, protected by the passphrase
alone (finding C4). It used PBKDF2-SHA512 at 600k — compute-only, which is
exactly what a GPU eats. Measured on this machine: PBKDF2 600k costs 241 ms and
Argon2id 64 MB/t=3 costs 322 ms, near enough the same honest work, except only
one of them forces an attacker to find 64 MB per guess.
So the bundle key is now Argon2id 64 MB / t=3 / p=1, via a vendored WebAssembly
build (no external host — the CSP forbids one, and 12.2 will tighten it further).
Parameters chosen by measurement through that build: 19 MB is OWASP's floor at
118 ms, 256 MB is 1.3 s and too slow for a phone, 64 MB sits where a login should.
What this buys, stated honestly: cracking a bundle yields the owner's identity
keys, and with them content on OTHER nodes and the ability to sign as them — not
the content on the operator's own node, which they host in the clear by design.
Argon2id raises that price steeply; it does not remove it, and a weak passphrase
still loses. Hence the floor raised to 12 characters and ~60 bits in the same
breath, which can only be enforced client-side: with the password split (T1) the
hub never sees a passphrase.
Migration is automatic and invisible. Bundles carry an "MBK2" marker; the old
form is still readable, and is re-encrypted the first time a browser backs it up.
Both keys are derived at sign-in, because which one a bundle needs is only known
once it is read and the passphrase is deliberately not kept around.
Two implementations of the KDF now exist — the browser's WASM and argon2-cffi in
QE — so a parity test holds them byte-identical. A disagreement would not look
like an error; it would look like an account nobody can open.
keypair_bundle_delete stays, without a UI. It is the mechanism behind withdrawing
your data from a node, exercised end to end, and it will belong to a deliberate
"forget me on this node" action rather than a setting that quietly disables
multi-device.
Verified against the live deployment: the full workflow passes, including
recovering keys on a second client from the passphrase alone.
Tests: 341.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|