| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Streaming an Xvid/MP3 .avi answered "Unsupported video codec" — a refusal, on
a file ffmpeg re-encodes at about six times playback speed on the machine that
reported it. Nothing about the source was wrong. The node simply never reached
its own re-encode path.
`probe_video` maps a source codec to an MSE codec string and knows four:
h264, hevc, vp9, av1. Everything else returns None, because there is no
MediaSource decoder in any mainstream browser to give a string to — MPEG-4
Part 2 (Xvid, DivX), MPEG-2, VC-1, WMV, Theora. `_stream_video_inner` read
that None as a verdict on the file and refused, while the re-encode sitting
twenty lines below it was gated on `raw_video_codec in
BROWSER_INCOMPATIBLE_VIDEO_CODECS` — a set containing "hevc" and nothing else.
So the whole ffmpeg fallback existed, worked, and was unreachable for every
codec that most needed it.
The setting that governs the fallback has documented the intended behaviour
since it was introduced: draft-v6 §2.11 says `transcode_incompatible_video`
covers "HEVC *and other browser-incompatible video codecs*". Only HEVC was
ever wired up.
Two questions were being answered by one value, and they are separated now.
"Is there a video stream at all" is the only thing this path genuinely cannot
serve, and the only refusal left. "Can it be copied" needs both an MSE string
to put in `stream_init` and a codec browsers decode; a source failing either
is re-encoded.
The operator's opt-out keeps meaning what it says, and it no longer means the
same thing for every source, because it cannot: HEVC has a codec string, so
`transcode_incompatible_video = false` falls back to a copy and the viewer's
own decoder decides (unchanged). MPEG-4 Part 2 has none, so there is nothing
to fall back to — a `stream_init` with no codec string is one the client
refuses before the first byte — and the stream is refused naming the setting.
"Unsupported video codec" is what sent this report to the file, and the file
was fine.
Verified against the reported file end to end: ffprobe reports mpeg4/mp3
720x404, the decision comes out `can_copy=False`, and the pipeline's exact
argv produces H264 High level 4.1 plus stereo AAC-LC — matching the
`avc1.640029,mp4a.40.2` that `stream_init` advertises and that the client puts
through MediaSource.isTypeSupported byte for byte.
test_stream_hevc_transcode.py becomes test_stream_video_transcode.py: it was
always about the policy rather than about one codec, and it now carries both
halves of it, with a synthetic Xvid/MP3 .avi built the same way as the HEVC
clip. Its module-level skip on libx265 went with it — an ffmpeg without x265
still encodes MPEG-4 Part 2, so that marker was skipping the reported defect
entirely on any box without it; it now gates the HEVC cases alone. Three
cases added, checked against the unfixed source. Hub and node suites 2269
passed, 4 skipped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019GXmScYB1uR29YCt74si9J
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The reconnect scenario added with the fix proved the composer follows
`_setDevicePk`, and it poked that method itself at both ends. That is a
narrower claim than it reads as: it says nothing about whether a reconnect
*reaches* it, and the harness's own `Host` stands in for group-page.js, so a
green probe did not mean the page joins the two.
Both halves are real now. The scenario calls `connect()` with the arguments
`_reconnectLoop` calls it with; it stops at signaling, because there is no hub
in the harness, and the identity has to be gone by then — connect() drops it
before it touches the network. The restore is the shipped `_announceDevice`,
answered by the stand-in node with a `device_hello_ack` as `_do_device_hello`
answers it, and the key it settles on is the one the following send seals and
signs with. Three assertions check the scenario went that way rather than
through a variable set by the test.
The seam the harness cannot drive gets its own check: the wiring exists, the
prop is in `commonProps`, and the callback is set *before* `connect()` — after
it, device_hello's answer is missed and the composer starts closed. That check
first passed with the wiring deleted, on the strength of a comment naming the
callback; it matches the assignment now.
Two things the harness turned up. `do_POST` answered every path, so the offer
connect() posts to the hub was swallowed as the measurement and put the
machine's own SDP, public address included, into the probe's output — it
answers `/log` and nothing else now. And a connect() that gives up before
`await channelReady` left that promise rejected with nobody attached, so
closing the peer connection printed "Uncaught (in promise) DataChannel closed"
on every failed reconnect attempt — noise in exactly the log a freeze is read
from.
Checked against the unfixed source both ways: with the clear removed from
connect() and the wiring removed from group-page.js, three cases fail; with
them back, 15 pass. Hub and node suites 2266 passed, 4 skipped.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019GXmScYB1uR29YCt74si9J
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The Chat tab froze about every other day — the textbox stopped taking clicks —
and it never recovered on its own: no timeout ends this one, only leaving the
group or restarting the client. A console dump of a session it happened in ruled
out everything it could and named nothing.
What that dump established was almost entirely negative, and that was the useful
part. No `Response timeout`, no `unsolicited`/`unrouted`/`with nothing waiting`
— so the 2026-08-30 routing defect, which produces this exact symptom for thirty
seconds, had not recurred. No `PC state: disconnected|failed`, no second ICE
cycle, no `Reconnected after N attempt(s)` — so the connection was alive and
untouched. The freeze was in the page, and no path that logs anything had run.
The composer is `disabled=${sending || cannotSend}`, and `cannotSend` was
`transport.connected && !transport.devicePk`, read off a **ref** during render.
`devicePk` is settled inside connect(), so every reconnect clears it and settles
it again; a ref changing re-renders nothing, and nothing else announced it. So
the panel went disabled on whatever unrelated re-render came next — a message
arriving — long after the identity was actually lost, and had no event that
would open it again. group-page.js never touches `status` after 'connected', and
`onReconnected` is claimed by video-player.js, so there was no second chance.
It was silent as well as sticky. `_announceDevice` had three exits that wrote
`devicePk` without a word: two early returns that left the *previous*
connection's value standing, and a reply that is not `device_hello_ack` — an
`error` reply does not throw, so the `.catch()` at the call site never saw it.
Reproduced in chat_send_probe.py, which mounts the real ChatPanel over the real
transport: with the old code, identity cleared leaves the composer open, an
arriving message latches it shut, and restoring the identity does not reopen it.
Every write to `devicePk` now goes through `_setDevicePk(pk, why)`, which logs,
traces and calls `onDeviceIdentity`; group-page holds the answer as state and
ChatPanel takes it as `deviceReady`. Defaulting that prop to `true` fails open —
a wiring mistake here must not be able to leave anyone with a dead textbox.
Two things found on the same path and fixed with it. `_send` throwing inside
_sendAndWait's executor left the pending entry and its 30s timer behind, so a
request that never reached the wire still logged a "Response timeout" half a
minute later. And the instrumentation this was meant to be diagnosed with
(3be8bd2) writes to localStorage behind ?trace=1, not to the console, so the
dump could not have carried it: the two lines that decide the composer's state
are now logged unconditionally, and MeshBayTrace gains `record` so the composer
writes into the same timeline as the channel events.
Hub suite 2264 passed, 4 skipped. chat_send_probe.py gains a `reconnect`
scenario and test_chat_send.py four cases, each checked against the unfixed
source.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019GXmScYB1uR29YCt74si9J
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A PDF preview showed the "this browser will not display the PDF inline"
fallback everywhere — in the desktop client since its first launch, and in
the browser since the hub started sending a CSP on 2026-09-01. It read as a
missing native feature because before that commit the hub sent no policy at
all, so Chrome had once worked and the application never had.
Two directives govern one feature. `files-app.js` decrypts the file in the
page and hands it to `<object type="application/pdf">` from a Blob; Chromium
loads that as plugin data (`object-src`, absent and therefore falling back to
`default-src 'none'`) and then renders it in an internal frame (`frame-src`).
Opening either alone changes nothing visible — the second refusal produces the
same fallback. `'self'` covers neither: a same-origin `blob:` URL is not
matched by it in either directive, measured in Chrome 152 against the deployed
page and in Electron 44 against the client's own policy.
`plugins` stays at its default `false`: the built-in viewer is not behind that
flag on Electron 44, verified by rendering one.
Widening `object-src` from `'none'` to `blob:` admits only what page script
minted itself, at a type this code sets — PDFium parsing bytes that came from
a node, which is what any browser does with the same file once downloaded.
Tests: each policy is pinned to carry `blob:` in both directives (each fails
if either token is removed), and the two policies are now held identical
directive by directive apart from the two deliberate differences — the comment
claiming they were the same had already drifted and nothing checked it. The
CSP source parser in test_desktop_shell.py read `//` comment lines as
directives, which is the "parse directives, not text" mistake this file
already records; it skips them now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XauykfBvRrpy6RYbF6F7Wu
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported the day MNP 3.0 shipped: playing a track answered "Too many files open
at once without a transfer. Download this one instead of previewing it."
§3.4.1's bound of two was reasoned about *viewers* — a photo viewer shows one
photo, a preview modal one document, and the second is for prefetching the next.
It forgot the music player, which warms a read-ahead window: `prefetchDepth()`
returns 5 on Wi-Fi and 3 otherwise, so playing an album has six files in flight
and the fourth was refused. Browsing a group is never subject to a transfer slot
— that is a stated requirement, not a tuning parameter — and a constant nobody
had checked against the client broke it.
Twelve now: six for the music read-ahead at its widest, two for a photo viewer
and its own prefetch in the same session, the rest as headroom. Generosity is
cheap here and refusal is not — this is a fairness control among cooperating
clients, not a security boundary, so a client that lies gets twelve files at a
time instead of its member cap, bounded and audited, while refusing a legitimate
read breaks the requirement outright.
And the number is now derived rather than chosen: a test reads `prefetchDepth()`
out of the shipped player and fails if the node's bound no longer covers it, so
widening the client's read-ahead breaks the build instead of reaching a person.
Checked by widening it: "the music player reads 21 files ahead and the node
admits only 12".
Three cases that hard-coded "two then refuse" now set their own limit — they are
about the mechanism, and the shipped number moves with the client.
Node suite 1210 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported from Chrome, with a screenshot: four downloads with no pause button
and an upload beside them with one, and nothing anywhere saying why.
The reason is real. Without a granted download folder the browser writes
through the service worker — a download it already owns, which cannot be paused
without stalling it somewhere we can neither see nor resume. An upload writes to
the node, which keeps the position, so it is always pausable. But that was
stated only in a Settings line nobody reads on the way to a download, and a gap
where the row above has a button is not an explanation.
So a download that cannot be paused now shows a dimmed pause icon where the
button would be, carrying the reason and the remedy in its tooltip. Not a
button: there is nothing to click, and a disabled one invites the click anyway.
And only where the advice can be taken. Firefox and Safari have no folder to
choose — the streamed path is the only target they have, which is what §6.5 of
~/next/improve-downloads.md costs out — so telling someone there to choose one
would be advice they cannot follow. Nothing is drawn.
Hub suite 866 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported while testing the flag day: pausing an upload put it under "Finished".
"Finished" was defined by exclusion — everything that is not running, queued or
preparing — so it swallowed `paused` the day pausing shipped. A transfer
somebody stopped on purpose then sat beside the ones that are actually over,
offering a resume button in the section of things that cannot be resumed, and
dropped out of the badge, which announced less activity than there was.
Paused is now its own group, in all ten catalogues, and counts as active: it is
not over, the person means to come back to it.
The three filters are lifted out of `app.js` and executed rather than described
in the test, and one case asserts that every status lands in exactly one group
— a state added later that falls into none is a transfer the panel simply does
not show, which is how this one got in.
The same report also said the three running downloads lost their pause buttons
when the upload was paused. That part is **not** explained and **not** fixed:
the store returns `pausable` true and status `running` for all three (new test),
closing an upload lease pumps only the upload queue, and the button's condition
is a pure function of those two. All three say the buttons should have stayed,
so an observation is missing rather than a cause.
Hub suite 864 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage 4 of ~/next/improve-downloads.md, the flag day. Leases become compulsory
and a 2.x peer is refused at the handshake.
**The bound on leaseless reads (§3.4.1) did not exist, and it is what makes the
rest mean anything.** Browsing a group is never subject to a transfer slot —
that is an operator decision and a requirement: a member must be able to browse
a group at capacity exactly as they browse an idle one. But "not leased" cannot
mean "unbounded", or a client that simply omits `tr` transfers outside every cap
and the caps are decoration. A session may now read two distinct files at once
without a lease: one because a viewer looks at one file, two so that prefetching
the next photo stays possible. A count of files and not a byte budget, because a
RAW photo is 60-80 MB and is browsing while a 40 MB archive is a download, and
no size threshold separates them. Thumbnails, posters and cover art never reach
this check at all — they resolve out of the node's own cache.
It is a fairness control among cooperating clients, in the company of
`max_concurrent_streams`, and is not a defence against a member determined to
saturate a node's disk. That member is a member, and the answer to them is
`member revoke`.
**MNP_VERSION and MNP_MIN_SUPPORTED both move to 3.0**, on both sides. The
messages are additive; the requirement is not. An opt-in switch would leave a
leaseless branch reachable on every node, which is finding C6's lesson — a
transport that accepted a bare JWT — one feature later.
**The desktop client now checks before it connects.** The SPA is served by the
hub and picks up a new client on reload; the application ships its own
interface, so an un-updated one would sign in, list groups, and fail every
connection with `version_too_old` — a refusal in a protocol vocabulary with
nothing anyone can act on. It asks `/v1/hub/version` for `client.minimum` and
says so plainly instead. An unreachable hub is deliberately *not* "too old": a
captive portal or a closed laptop must not make starting the application
impossible.
**Every package is aligned on 0.13.0.** `meshbay-client/package.json` had
drifted to 1.0.0 while the Python packages were on 0.12.0 — invisible until
something compared those numbers, and then load-bearing: an installed client
announcing 1.0.0 sorts above a 0.13.0 minimum and walks through the gate meant
to stop it. That is stated in the code rather than left to be rediscovered; it
is acceptable exactly once, because the operator is updating every client, node
and hub by hand for this flag day. A new test fails if two packages ever
disagree again, and another fails if the hub would refuse the client the tree
builds.
Node suite 1209 passed, hub suite 861 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported from Chrome: a download started while an upload was running took
thirty seconds to begin, every time. The console named it exactly —
/_mbdl/mtty5btz-sbmgdegx 404 ()
[MeshBay] the worker did not answer the download within 15s (attempt 1)
A 404 from the hub means the request reached the *network*: the worker looked,
found no entry for that id and let it through. So the worker was alive and
controlling the page, and the message handing it the stream had simply never
been processed.
`pending` lives in the worker's memory, and a worker with nothing to do is
terminated within tens of seconds. A WebRTC upload gives it no events at all,
so minutes of uploading leave it dead; the stream posted to it is lost,
silently, and the iframe then wakes it with nothing to find.
`mbdl-ping` already existed for this exact reason -- sent every ten seconds
*while* writing, because a streaming response does not count as activity.
Nothing sent one before *starting*. So a download now wakes the worker and waits
for the pong, and `sw.js` answers `mbdl-ready` once it has actually stored the
entry, which the page waits for before navigating: confirmed rather than
assumed. A worker that predates the ack sends nothing and the page navigates
anyway, which is what it did before.
This cause was measured and wrongly dismissed hours earlier, with an idle probe
that made the worker work between its own attempts -- it never actually slept.
A measurement that does not reproduce the conditions refutes nothing. The
harness now models a worker that is asleep: a ping wakes it, and anything else
posted while it sleeps is lost, which is what made the failure silent.
`test_backpressure_is_real` read the first `worker.postMessage` in the function
to check that the readable half is transferred rather than copied. The wake-up
put a ping in front of it, so it began inspecting a call that carries only a
port -- and kept passing. It now checks every post, each bounded by its own
call, since the keep-alive ping transfers nothing at all. Same shape as the
upload-seal contract this morning: a guard that reads "the first" stops
guarding the moment something is inserted before it.
Hub suite 851 passed. Both new cases checked against the unfixed source.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage 8 of ~/next/improve-downloads.md, second half, plus the gap it exposed in
stage 7.
**Asking where to resume.** The node identifies an upload by (member,
directory, filename), so a client resuming one has to name the file — and
`transfer_open`, the obvious place to ask, travels in clear. Naming it there
would undo exactly what sealing this path bought in MNP 2.0: before it, the same
file was ciphertext leaving a node and plaintext arriving at one. So the
question is asked inside the seal that already exists, as an ordinary
`file_upload` with no bytes and `chunk_index: -1`. The node writes nothing,
creates no state, reserves no name, and answers with `resume_from` in the sealed
ack. A node that predates it refuses the index, which the client reads as "start
from the beginning" — the behaviour it had anyway — and the wait is bounded so
one that answers neither does not strand an upload.
The probe is answered after every check the write path makes, so it cannot ask
questions about a directory the caller may not write to, and it answers only
about the member who asks: otherwise one member could measure another's
progress on a file they never sent, and worse, resume it.
**Pausing an upload.** Reported: no pause button on an upload, even in the
desktop app. Stage 7 built pause around the download path — a target declares
whether it can be stopped — and an upload has no local target to ask. It was
also refused by design, since a transfer handed a lease it cannot re-create must
not be offered a button that would drop its slot for good. Uploads now ask for
their slot rather than being handed one, and say they are pausable outright: a
File is seekable and the node keeps the position. Resuming re-probes rather than
trusting the client's own memory, so it works across a reconnect too.
**And the slot they hold.** `_do_file_upload` never called `slots.touch(tr)`.
Chunks are not gated by the lease, so the file arrived — but the node reclaimed
a grant nobody appeared to be using after thirty seconds, twice, then abandoned
it, and the widget follows the lease. Measured from the journal: a 3.5 GB upload
read "waiting, 0 ahead" for a minute and a half while it was transferring. The
download twin of this was fixed on 2026-09-08; the same omission was still here,
invisible until uploads took a real lease.
`test_the_upload_itself_is_sealed` now checks every message `uploadFile` sends
rather than the first. Adding the probe put a second one in front of the one it
was written for, and it would have kept passing while guarding nothing.
Node suite 1202 passed, hub suite 850 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage 8 of ~/next/improve-downloads.md, first half. Two defects that are the
same defect seen from two sides.
An upload's progress lived on the session, keyed by `rel_dir/filename`. A
dropped connection threw it away and the client's next chunk was refused with
`not_started`: an upload interrupted at 99% could only be started again from
zero, on a link flaky enough to have interrupted it once. It now lives in the
group context, keyed by member as well -- a shared directory means two people
can be sending IMG_1234.jpg at the same moment and neither may inherit, or
overwrite the position of, the other's.
What the lost state left behind was a `.part` nothing would ever finish, delete
or look at again. It is not an index entry, so it is invisible to every member
and to the operator's own file list: one abandoned film is a gigabyte of their
disk, kept for ever. That leak predates this branch.
A `.part` is deleted only when **both** hold: no upload is writing it, and
nothing has been written to it for 24 hours. Waiting costs disk; being wrong
costs somebody their upload, and is not reversible -- so a read-only root is
never walked (it cannot have received an upload), an unavailable one is never
walked (an unmounted drive reporting "nothing found" is how a careless janitor
deletes a library), and a file whose mtime is in the future is left alone (a
clock that went backwards is not evidence). The reaper matches whole paths and
the state records the path it is writing, rather than both sides rebuilding one
from a root name -- two implementations of one rule whose failure mode is
deleting a live upload.
The rules are in `uploads.py`, pure logic with no asyncio and no transport, the
same shape as `transfers.py` and for the same reason.
23 cases, four of them checked against the unfixed source. Node suite 1195
passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported from Chrome: connecting to a group triggered a page refresh within
seconds, taking the WebRTC session down with it.
The boot check added with the bypass repair asked its question by *performing a
download* -- a four-byte stream through a hidden iframe. Chrome rations the
downloads a page may start without a user gesture to about three, measured: on
a first visit three consecutive attempts went served, served, refused. So the
check competed with the person's own downloads for that budget, and its answer
depended on how much of the budget was left. On a healthy page it concluded the
worker could not serve, and reloaded.
The same mistake the repair was written to fix, from the other side: paying a
capability to obtain a diagnostic.
The replacement costs nothing and asks nothing. Measured on Chrome, at document
start, before anything registers:
first visit controller false, registration false
ordinary reload controller true, registration true
hard reload controller false, registration true
Being uncontrolled while an active registration already exists names a
hard-reloaded document exactly, so that is now the whole of the evidence. A
first visit is uncontrolled too and is not a bypass -- the worker is installing
and will claim the page in a moment -- which is precisely the case that was
reloading.
Four cases, each checked against the unfixed source, including that priming
performs no download at all. Hub suite 847 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported from testing 7a: pause worked in the desktop app and no button
appeared in Chrome. That is the design working — without a granted download
folder, "save automatically" means the service worker, and that target is a
download the browser already owns — but nothing anywhere said so, and choosing
a folder looked like a question of where files land.
So the Settings line now says what it costs not to choose one, in all ten
catalogues. It renders only where a folder can be chosen at all, which is
exactly the browsers the advice applies to.
Also pins the tier table the pause button is drawn from: a granted folder, a
save dialog and the desktop sink can be paused, a service-worker stream cannot.
Four cases through the real `_openDownloadTarget`, and one more that reads the
value off the real `downloads.js` rather than a stub of it -- the first version
of these stubs did not carry the field at all, so the cases would have passed
while checking nothing.
Hub suite 847 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Stage 7a of ~/next/improve-downloads.md: pausing within a session, on the
targets that can actually do it. Resuming across a reload is 7b.
A paused transfer holds **nothing**. Its slot goes back to the node the moment
it stops and resuming rejoins the queue at the tail, because anything else lets
one member close a node by pausing four downloads and going to lunch. So the
lease is taken inside the run loop rather than before it, and pause is refused
outright for a transfer that could not ask for another one.
Resuming is exact rather than approximate: the pipeline stops between two
chunks and never inside one, so what is on disk is always a whole number of
chunks and `fromChunk` is a verified position. The failure mode being avoided
is a file that looks complete and is quietly corrupt.
The target has to survive it, so a pause no longer reaches the `abort()` that a
failure does -- that would delete Electron's `.part` or the file just created in
the granted folder, leaving nothing to continue. And the in-memory fallback
keeps its accumulated chunks rather than starting a second array.
The button is drawn only where the target says it can. A service-worker stream
says no, in its own code and for its own reasons: the browser is already writing
an HTTP response into its own download folder, not feeding it stalls that
download where we cannot see or resume it, and an idle worker is terminated
within seconds. Firefox and Safari therefore keep cancel and get no pause, which
is the decision recorded in §6.5.
Cancelling a paused transfer ends it. A paused run is parked on a promise;
without waking it the row said "cancelled" over work that had not stopped and a
target that was still open.
Six cases, each checked against the unfixed source. Hub suite 842 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The repair worked but arrived too late to help: about thirty seconds after a
hard reload, by which time four downloads had been started and hung, and the
page reloading under them read as an unexplained refresh.
Two delays, both removed.
`_claimController` waited its whole control budget before asking for the claim.
A page that is uncontrolled while an active worker exists will never be claimed
on its own -- a document fetched by a hard reload is exactly that shape -- so
the fifteen seconds were spent waiting for something that was not coming. The
claim is now asked for first; waiting is the fallback, not the opening move.
Measured in the harness: 6042ms of a 6000ms budget before, milliseconds after.
And a download that starts while the self-test is still running now waits for
it rather than racing it. Otherwise the click spends both its attempts failing
on a path that is about to be repaired, which is what put four frozen rows on
screen.
Hub suite 836 passed. Both new cases were checked against the unfixed source.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Downloads on Firefox failed with "the worker did not answer the download within
15s", every time, for one operator, while the same profile driven from here
succeeded every time. Their own test sequence found it: a freshly started
browser downloaded four files out of four, twice; one Ctrl+F5 and every
attempt afterwards failed; restart, fine again; Ctrl+F5 before any attempt and
the very first one failed.
A document fetched by a hard reload is loaded with the service worker bypassed.
It can still be claimed afterwards, so `navigator.serviceWorker.controller`
comes back and every check in `_claimController` passes — but the navigations
that document starts keep missing the worker, and the hidden iframe a streamed
download needs is a navigation. On Firefox and Safari that is the only way to
write a file too large to hold in memory, so the download cannot happen at all,
for the life of the page.
Being controlled is not being servable, so priming now asks the question
directly instead of inferring it: a four-byte stream and a hidden iframe,
exactly as a real download would, torn down completely so nothing lands in the
download folder. When it goes unanswered the page reloads once, ordinarily,
which puts it back under the worker. The flag lives in sessionStorage rather
than a variable because it has to survive the reload it triggers, and because a
page that is still unservable afterwards must stop rather than loop.
Also stops telling people to change browser. The message said "use the desktop
app, or Chrome or Edge" for a state an ordinary reload undoes, on the one path
Firefox has no alternative to; all ten catalogues now say to reload first.
The hard reloads were on my instruction: the SPA's HTML is served `no-store`,
so a plain reload has always picked up a new build and Ctrl+F5 was never needed.
Hub suite 834 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"The node saw no transfer" was concluded twice from a journal that could not
have shown one: the open was logged at DEBUG, and the daemon runs at INFO. Two
diagnoses were built on that non-observation, and both were wrong.
Turning the root logger up to DEBUG is not the answer either — aiortc logs every
SCTP chunk, which on a 2 GB download is both unreadable and slow. One line per
transfer is not a volume problem, and it is the line that answers "did the
client ever ask for a slot, and what was it told".
Node suite 1172 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Four downloads on Firefox sat at "preparing" indefinitely, with the target queue
already bypassed there, so each opening was hanging on its own. The node journal
showed `d=0/8(q0) u=0/8(q0)` — no transfer had been asked for yet.
`_claimController` had two waits with no deadline at all,
`navigator.serviceWorker.register()` and `navigator.serviceWorker.ready`, while
SW_CONTROL_BUDGET_MS bounded only the wait that comes after them. `_swPromise`
is shared, so one unsettled wait left every download on the page suspended on
the same promise for the life of the tab.
Measured on Firefox 154, against a local 127.0.0.1 site so no hub was involved:
a worker that installs gives register() in 8ms and ready in 0ms; a worker whose
install handler rejects gives register() in 7ms and a `ready` that never settles
— still pending past ten seconds. register() resolves as soon as the
registration object exists, carrying nothing but an *installing* worker; ready
is what waits for an active one.
Every wait is now inside one budget, with two carve-outs so that a deadline
never costs a capability. A `ready` that times out while registration.active is
set is not fatal: ready may be waiting on a newer worker that cannot install
while an older one serves perfectly well. And the mbdl-claim recovery keeps its
own budget outside the deadline, because giving up there would cost Firefox the
only unbounded way it has to write a download to disk.
A deadline alone would have been a better-explained failure rather than a fix: a
registration stuck with nothing but an installing worker does not heal, and
every later visit finds the same one. So when ready times out with no active
worker, the registration is discarded and asked for once more with a fresh
budget, and the page repairs itself instead of needing developer tools.
Four cases pinned, each checked against the unfixed source. Hub suite 830
passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Four downloads on Firefox all sat at "preparing", with the node journal showing
`d=0/8(q0) u=0/8(q0)` — not one transfer opened, so nothing had got past the
client's target opening.
Serialising those openings was new in d6c4808, and on Firefox it regressed what
had always worked: four openings that ran at the same time began waiting on the
slowest. `_targetQueue` is module-level and never reset, so an opening that
never settles leaves the page unable to start any download again until it is
reloaded.
Two bounds, both narrowings of the queue rather than of any capability:
Only an opening that could actually show a dialog joins it. Firefox and Safari
have no `showSaveFilePicker` at all, so nothing there can race anything and the
queue bought nothing while costing everything; they now bypass it entirely,
which restores the previous behaviour by construction rather than by tuning.
And no opening waits behind another for longer than TARGET_QUEUE_BUDGET_MS
(90s) — generous enough never to cut in front of a real dialog, finite because
the alternative is a download panel that only a reload can fix. Releasing early
is safe: whatever is ahead is still the only unbatched opening, so the released
one takes the streamed path and opens no second dialog.
Measured on Firefox 154 against the deployed hub before writing any of this:
`register` and `ready` return instantly, the page is controlled, and four
serialised openings are served in 5-18 ms. The streamed path was never the
delay; the queue was.
Both new cases were checked against the unfixed source: without the bypass the
peak concurrency is 1 instead of 4, and without the budget the stuck-opening
case hangs. Hub suite 826 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Selecting four files on Chrome produced a Save As dialog for the first, then
one for the second only after that file had finished, while the last two timed
out; on a later attempt the three remaining transfers appeared frozen.
Two things were going on. Opening the target inside `prepare` had removed the
accidental serialisation that `for (…) await downloadFile(e)` used to provide,
so `_openTargetInTurn` now queues the openings — but a queue whose head is an
unanswered dialog is a head-of-line block, which is what the "freeze" was.
The code already recovered from a picker with no gesture behind it by streaming
instead, on the `SecurityError` Chrome throws. That branch was never reached:
Chrome does not throw, it shows the dialog anyway and waits for a human. So
anything that has to wait its turn is now marked `batched`, and a batched
opening prefers the streamed path whatever the download mode says. The first
file of a batch — the one actually holding the gesture — still gets its dialog,
so the preference is honoured where it can be. For the rest there is no gesture
left to spend and nothing is lost by streaming: the file still lands on disk,
in the browser's own download folder. Only the choice of folder goes, and it
was not on offer. If the worker does not answer, a batched download falls back
to the dialog rather than failing.
Also logs which path led to a dialog. A dialog is the one outcome nobody can
diagnose after the fact — it looks the same whether it was asked for or fallen
back to — and the report this fixes needed three test cycles to narrow. The two
harnesses that lift `_openDownloadTarget` as text now route console.info to
stderr, since they parse stdout as JSON.
Measured against the deployed hub in Chrome 152: the streamed path serves the
hidden iframe in 2-3 ms on a normal load, after a hard reload (via the
`mbdl-claim` recovery already in `_claimController`), and twice in the same
document. Hub suite 824 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Three defects in the probe, found while extending it to cover the per-member
hot-swap. The third is the one worth keeping.
`transfer_state` is the reply to an open, the acknowledgement of a close, and
the push that carries a grant minutes later. Reading "the next one" therefore
returns somebody else's answer as soon as more than one transfer is in play —
the probe took two stale `closed` acks as the replies to two opens and reported
a working cap as broken. That is exactly the defect `req_id` exists for in this
protocol, committed inside the tool written to check it. Replies are matched on
`tr` now.
The other two: the `transfers show` parser counted the pool summary line as a
lease once that command grew a per-group section (a probe that reads a
human-facing format signs up for this), and the per-member check began with a
member who already held several leases, which measures nothing. It waits for
the operator's own view to go quiet first — waited for, not slept through.
Both probes written today reproduced a bug already recorded in CLAUDE.md: this
one, and yesterday's timer with no strong reference. A tool that verifies the
code is not exempt from the code's rules.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Raising the per-member cap from 2 to 4 left both waiting transfers on
"waiting". The pool granted them correctly and nobody told the peers:
`ops.set_transfer_limits` computed `granted` and never sent a `transfer_state`,
where the node-wide path (`WebRTCTransport.set_capacity`) does.
That is the first row of §5.2 of ~/next/improve-downloads.md — "node granted a
slot, the push was lost" — reached by writing the decision and forgetting the
send. It is the same omission as the missing `touch()` call one layer up, on the
same day: a mechanism that is right everywhere except at the seam where it has
to reach somebody.
The client recovered after its 60-second watchdog re-asked, which is why this
looked like a slow queue rather than a lost message.
`transfer_probe.py --operator` covers it now, separately from the node-wide
hot-swap it already covered — different door, different code path, and only one
of them was tested. Verified failing with the push removed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported as "the slots seem hardcoded to 2": `meshbay-node transfers set 8 8`
and still only two downloads at a time. Not hardcoded — that is the *per-member*
cap, which is a group's setting and is checked before the node's, so raising the
machine's total cannot move it. But the diagnosis was right in the way that
matters: nothing could change it.
`OP_TRANSFER_LIMITS` shipped with exactly one front door, the signed MNP
handler, and nothing anywhere opened it — no client call, no CLI verb, no
loopback route. So the cap sat at its default of 2 for ever, which from outside
is indistinguishable from a constant. CLAUDE.md states the rule this missed:
operator operations are one implementation with several front doors.
- `PUT /api/groups/{id}/transfer-limits`, calling the same
`ops.set_transfer_limits` the signed handler calls;
- `meshbay-node transfers per-member <downloads> <uploads> [--group X]`;
- `transfers show` now separates the node-wide pools from the per-group
per-member caps, and marks each `[set]` or `[default]`. It printed "2 per
member" with no indication of where the 2 came from, which is half of why
this looked like a constant.
Zero is refused here as everywhere else: it is not "unlimited", and a member who
may not transfer at all is a member the operator revokes.
Verified on a live node: the cap changes, survives a daemon restart, and
`transfer_probe.py --want 6` measures 4 granted against a cap of 4 where it
measured 2 before.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clicking Download produced nothing — no row, no icon, no panel — for as long as
it took to open somewhere to write, and then several rows at once. The streamed
path waits for the worker twice; a Save As dialog waits for a person. The row
was created after that, so the slowest part of a download happened with nothing
on screen to say it had begun.
The store gains a `prepare` step, distinct from `run`, and the order is now:
row, then target, then slot.
That last part is why the obvious fix was wrong. Taking the slot first would let
the row appear immediately, and it was tried this morning: a granted slot has to
be taken up within the node's deadline, opening a target can outlast it, and
three downloads became one. (The diagnosis at the time blamed that ordering for
revocations which were in fact a missing `touch()` call — the revert was right
for the wrong reason.) `makeLease` is called after `prepare` succeeds, never
before.
Three behaviours fall out, each with a test:
- a dismissed dialog leaves nothing behind. `prepare` returning false drops
the row: nothing started, so nothing should remain on screen to explain it;
- the row takes the name the file was actually saved under, once known;
- a refusal above the memory ceiling fails the row that is already there,
rather than creating one to kill it.
`preparing` counts as live everywhere — badge, cancel, clearFinished, and
`_busy`, since closing a transport under a preparing transfer strands it exactly
as under a queued one. Six places asked "is this finished?" and were drifting
apart; there is one definition now.
Two mistakes in the tests, worth the note: one counted positions in an output
array by hand and was one out, which reads exactly like a failing assertion
about the code — the values are tagged now, not indexed. And
test_zip_size_limit.py's stub did not run `prepare`, so it no longer reached the
size check the file is about; it now behaves like the real store.
819 hub, 1169 node, 0 failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`--operator` closes the last two items of the live pass, from the side the
client cannot see:
- **a cap raised live starts what was waiting**, with no restart and no
reconnection. Draft-v6 §2.11 promises this and it was false for months:
`ops.set_node_settings` hot-swapped by assigning `webrtc._stream_sem`, an
attribute that has never existed. Now it goes through `set_capacity`, and
this is what says so from outside;
- **a vanished peer's slots are back before anyone asks.** §5 of the plan
makes that a hook on the connection rather than a timeout, and the
difference is two minutes of a node that looks full.
Plus the operator's view of the queue itself, which is the only window into a
transfer stuck at "waiting" — and which reported the module defaults instead of
the operator's values until this afternoon.
Two mistakes in the check, none in the code, and the second is worth keeping:
the first version set the node cap and the member cap both to 2, so one account
holding two transfers hit both at once. Raising the node-wide cap then correctly
changed nothing — per-member is checked first, by design — and the probe
reported the design working as a failure. It now puts the node cap below the
member cap so the queue is held by the machine, which is the only arrangement
where this can be measured at all.
The cap is restored to whatever the node was running before the probe touched
it, not to a default.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It lived in `QE/`, which is deliberately not versioned — credentials and test
artefacts go there — so a tool that found several defects no test in the suite
could reach existed on exactly one machine.
What it found, none of it reachable from pytest: a cap that was never enforced,
a queue that granted a slot and never told the peer waiting on it, leases that
outlived the session holding them, and `transfers show` reporting the module
defaults instead of the operator's own values.
Not collected: the filename does not match `test_*.py`, and that is the point.
It talks to a real hub with real credentials and takes minutes; what belongs in
the suite is already there.
It still needs two things from `QE/`, which stay out of the repo: `e2e.py`, the
second implementation of the client whose `Client` speaks MNP over a real
WebRTC DataChannel, and `demo.env`. Both are located at run time and their
absence is explained in a sentence rather than raised as an ImportError from
four frames down.
Verified from the new location against the live node: 4/4.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Steps 5 and 6 of ~/next/improve-downloads.md. The node has handed out slots
since step 2 and nothing asked for one; now the client does, and the panel shows
what is happening.
`transport.openTransfer()` returns a Lease: `acquire()` resolves when the node
grants, `release()` gives it back exactly once, and nothing else in the client
speaks to the node about slots. Whether a node hands out slots is read from the
handshake ack rather than guessed from a timeout — "no answer yet" and "this
node will never answer" are indistinguishable in time, and guessing wrong either
stalls every download or defeats the cap.
Two things exist only because a queue can lie: a watchdog re-asks when a pushed
grant does not arrive (the node is idempotent on `tr`, so asking again is free),
and a grant for a transfer the page has forgotten is handed straight back rather
than held until the node's deadline.
The slot is asked for **after** there is somewhere to write, and that ordering
is load-bearing: opening a target takes thirty seconds of streamed-download
timeouts, or as long as somebody leaves a Save As dialog open, and a grant not
taken up in time is revoked. Moving it earlier looked better and broke three
downloads into one. Pinned by a test.
The panel groups by state — running, waiting, finished — rather than re-sorting a
flat list, so a row moves only when its own state does. The ETA is withheld until
the speed window holds real measurement: a figure from the first two chunks
swings between four seconds and an hour, and people plan around the first number
they see. One live region announces state changes and not progress.
Three silent paths closed on the way: a download refused for want of a user
gesture (a browser grants one file picker per gesture, and downloading three
files is one gesture) now falls back to the streamed path, which needs none; a
click with no connection says so instead of doing nothing at all; and a queued
transfer counts as busy, so a transport is never closed under one that is
waiting for a grant that could then never arrive.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Two defects in the lease machinery, both found in the node's own log, neither
reachable from any test on either side.
**`touch()` was never called.** The node ignored `tr` on `file_req` entirely, so
`used` stayed False for every download ever made and the sweeper revoked each
grant thirty seconds in — while the file was transferring at 20 MB/s. The pool
was correct and the handlers were correct; the call between them was missing,
which is why neither side's tests could see it.
**The requeue was a permanent cycle.** A revoked grant went back in the queue,
was granted again a millisecond later because there was room, and was revoked
again thirty seconds on. The node logged the same two reclaims every thirty
seconds for as long as it ran — minutes after the transfers involved had
finished. Three chances now, then the lease is closed and the peer told.
`test_the_counter_never_drifts` could not have caught it: nothing drifted, the
same lease simply never left.
A lease that starts being used forgets its earlier misses: a client that took
two grants to get going is slow, not abandoned.
Also `transfers show` reported the module defaults rather than the operator's
values until something had transferred, so `transfers set 2 2` answered
"applied now" and the next line said 0/8 — indistinguishable, from outside,
from the hot-swap that did nothing for months. The test asserted the defaults
and so agreed with the bug; found by typing the command.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A download froze part-way through, on Firefox, with an empty console and a node
that stayed perfectly healthy. Three separate measurements cleared the node
(615 MB pulled whole over MNP), the transport (three files interleaved on one
connection, 1.5 GB, all whole) and the service worker (three concurrent 150 MB
streams in real Firefox 154) — because none of them was wrong.
The empty console was the evidence. `_sendAndWait` logs every timeout, so no
chunk request had expired: the client was not waiting on the node. Of the three
awaits left on that path only one was unbounded.
**A service worker with no event for about thirty seconds is terminated**, and
`respondWith(new Response(stream))` does not extend its life while the response
is still being written. The reader vanished mid-file and `writable.write()` then
never resolved and never rejected — no error, no log, no failed transfer, just a
progress bar that stops. The first stress probe wrote 450 MB in two seconds and
passed: fast enough to hide it entirely.
Measured in Firefox 154, writing 1 MB every 2 s: without the ping it stalled at
17 MB after 59 s; with it, 40 MB in 80 s, complete.
- the page pings the worker every 10 s while it writes, and the worker
answers. Receiving a message is an event, and an event resets the timer;
- that interval stops itself after two minutes with no write. A target can be
opened and never written to — a transfer cancelled while it waits for a slot
never runs, so nothing calls close() or abort() — and a timer nobody clears
pings for the life of the page. It also kept the Node test process alive for
ever, which is the same defect wearing a louder symptom;
- `writable.write()` is bounded at 60 s and fails with a message naming the
chunk. That does not fix whatever stopped a sink; it turns an unexplainable
freeze into a failed transfer that says so, which is the difference between
a mystery and a bug report.
Also: `Content-Disposition` lost a filename to a single apostrophe.
`encodeURIComponent` leaves `'` alone and `'` is the delimiter in RFC 5987's
`filename*=<charset>'<lang>'<value>`, so the header became unparseable and
Firefox named the file after the URL — 449 MB of film arrived complete as
"mtsshk9w-ohqty535". `(`, `)` and `*` get the same treatment, and a plain ASCII
`filename=` rides alongside.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Three headers govern whether a page may be framed, and all three had to be
wrong for the streamed download to work — so fixing them one at a time cost an
afternoon of redeploys and retests. They were visible together in a single
`curl -I` against the deployed hub, which is where this should have started.
The streamed-download path navigates a hidden iframe to `/_mbdl/<id>` so the
service worker is asked for the response it is holding. On Firefox and Safari
that is the only way to write a large file to disk: neither has the File System
Access API, and OPFS is capped at 10% of the volume's size (measured on Firefox
154: 389,233,459 bytes of a 3,892,334,592-byte volume, refused to the byte),
which a film exceeds.
- `frame-src` was reCAPTCHA's two origins with no `'self'`, so the frame
could not be loaded at all. Added when the captcha needed a frame; nobody
connected the two.
- `frame-ancestors 'none'` forbids all framing, this origin included.
- `X-Frame-Options: DENY` says the same in an older dialect. The spec says a
browser must ignore it when frame-ancestors is present — relying on that
while shipping a header that contradicts our own policy is asking to be
surprised, and we were: the CSP was fixed and the download stayed broken.
`'self'` and `SAMEORIGIN` refuse every foreign origin exactly as `'none'` and
`DENY` do. The clickjacking property is untouched; what they additionally allow
is this origin framing itself, which is the only thing the download needed.
Pinned three ways: `frame-src` must carry `'self'`, `frame-ancestors` must be
`'none'` or `'self'` and never name an origin, and the two framing headers must
agree — the defect was the disagreement, and either one read as correct alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Step 3 of ~/next/improve-downloads.md. Step 2 built the pools with constants;
this gives them to the operator, in the two scopes they belong to.
**The pools are the machine's.** `[node] max_concurrent_downloads` and
`max_concurrent_uploads`, default 8, on the §2.11 pattern: node.toml for a
fresh install, a roster.db override for immediate effect, editable from the
Node page and from `meshbay-node transfers show|set`, applied live through the
one `set_capacity` step 1 fixed.
**The per-member cap is a group's.** How many transfers one member may run at
once here — on the node like every other group setting (not the hub, which
would have authority over someone else's disk; not node.toml, which is
hand-written and needs a restart), changed by a signed operator instruction
(`OP_TRANSFER_LIMITS`, subject "d=2,u=2" so what is signed names the outcome),
broadcast to the group, and read live by the pools.
That was the one thing step 2's shape could not express: `per_member` was a
single node-wide number. `group_limits` and `member_cap(kind, member)` make it
a lookup — the group's own value if it has one, the node's default otherwise —
and it is deliberately the only dimension that is not node-wide.
Three refusals, each with a test:
- **absent means the default (2), never "unlimited".** A group that predates
the setting coming back unlimited would leave the node-wide pool as the only
control, which is the situation slots exist to end;
- **zero is not "unlimited"**, and is not "this member may not transfer"
either: the floor is one everywhere, and the CLI says to revoke the member
instead;
- **an unreadable row reads as unset**, not as zero — the same discipline the
sealed messages follow, where a payload that does not open must never become
a default state on its own.
`handshake_ack` carries this member's own caps for this group, so the interface
can say "2 of your 2 slots are busy" instead of drawing a bare spinner. Absent
reads as "no limit known" and the hint is not drawn — never as "unlimited",
which would have the interface contradicting the node.
1164 node, 793 hub, 0 failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Step 2 of ~/next/improve-downloads.md. A download is invisible to the node: it
is a series of independent `file_req` messages, with nothing saying one started
or ended, so there is nothing to count and nothing to cap. The lease is that
missing object.
`meshbay_node/transfers.py` holds the decisions and has no asyncio and no
transport in it, on purpose. The failure modes this has to survive — a slot the
node never gets back, a client waiting on a grant the node has forgotten — are
races through a DataChannel and unprovable there; here the clock is a parameter
and every method returns what changed, so the caller does the I/O and the tests
drive the worst case directly.
What it decides:
- two pools, downloads and uploads, separate from the stream pool: different
resources with different costs, and merging them makes both caps meaningless;
- per-member cap checked *before* the node-wide one, so a member at their own
limit queues behind their own transfers rather than holding a slot a second
member has none of. Per account across their devices, or the cap becomes a
function of how many tabs somebody opens;
- a queue that skips a member at their cap instead of waiting for them —
granting strictly in arrival order lets one member's limit stall everyone;
- `tr` drawn by the client and idempotent, which is what makes a reconnect safe;
- bounded per member, because unbounded queues are how a node runs out of
memory politely.
Every way a slot comes back, with the session teardown as the one that matters
(a closed tab, a quit browser and a dead network all arrive at
`shutdown_tasks`, and none of them needs a timer): explicit close, session
gone, a grant nobody took up in 30 s passed to the next in line, and a granted
transfer silent for 120 s reclaimed with its peer told, so a widget can offer a
resume rather than sit on a lie.
`GET /api/transfers` is the operator's window: when somebody reports a transfer
stuck at waiting, it is the only thing that says whether the node ever had them
in a queue — a log cannot, when the symptom is that nothing is happening. It
carries no filename and no path, which a test pins, because this is exactly
where one would be tempting.
Three things found while writing it, two of them mine:
- the randomised property test rejected `in_use <= cap` at once, and it was
right to: lowering a cap never interrupts a running transfer, so the count
legitimately sits above the new value. The invariant is that a *new* grant
never happens past the cap;
- the sweeper was started with `self._spawn`, which ties a task to one
session's set. It died with whichever peer opened the first transfer, and
every other peer's abandoned lease then stopped being reclaimed — a node that
fills up over days with nothing in the log. It belongs to the node now, with
its strong reference on the transport context;
- the pools are node-wide while `_peer_registry` is per group (finding H1), so
a slot freed in one group can grant one in another and the peer to notify is
not in the notifier's registry. Silently wrong in the first version.
Nothing enforces a lease yet: `file_req` is untouched, no client asks, and the
node grants everything. That is step 4's flag day, and this lands alone.
1148 node, 793 hub, 0 failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`ops.set_node_settings` hot-swapped the stream pool by assigning
`webrtc._stream_sem`. That attribute has never existed on WebRTCTransport — the
pool is `ctx["_transcode_sem"]` — so `hasattr(webrtc, '_stream_sem')` was always
False and the branch never ran. The setting was accepted, written to roster.db
and node.toml, and applied only on the next restart, which is exactly what
draft-v6 §2.11 says it does not need. An operator lowering the cap on a
struggling machine, or raising it after "Server busy", saw nothing happen and
had no way to find out why.
`WebRTCTransport.set_capacity()` is the one implementation, on the object that
owns the state, so the download and upload caps the transfer-slots plan adds
next do not each grow their own copy of the mistake.
Resizing has semantics worth stating: the new cap governs new streams and never
interrupts one that is running, because a slot is held for the length of a film
and lowering a number must not take somebody's film away. The replacement pool
is built with the permits that remain (`new - in_flight`, floored at zero) — a
full set would briefly allow more concurrent viewers than either the old cap or
the new one.
That needs a count of slots in use, so `_stream_video` now maintains one instead
of the code reading the semaphore's private `_value`: a number this code keeps
itself survives the semaphore object being replaced underneath it, and the same
counter makes the "N of M in use" log lines mean something.
test_stream_capacity.py drives the real transport and the real `_stream_video`;
`test_ops_calls_the_real_mechanism` fails if the dead attribute comes back.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Nine of the eleven were defects in the suite, two were assertions describing
behaviour the code had deliberately changed. None was a bug in the node.
Seven had one cause. `check_media_tools()` writes two module globals;
`monkeypatch` restores what a test patched and knows nothing about what the
call under test then wrote, so a test that pointed `shutil.which` at
"/opt/bin/{n}.exe" left `_ffprobe_path` there — a Windows path, on Linux — for
the rest of the session. Every later test that actually runs ffprobe died on
FileNotFoundError, in two files about video transcoding, for a reason nowhere
near themselves. Run those files alone and they passed; that is what made it
look like an environment problem for so long.
The autouse `_restore_media_tool_paths` fixture in conftest.py puts both back
after every test. That closes the class, not just this instance: any future test
that resolves media tools is undone whether it remembers to or not, which is the
only way an order-dependent suite stops being one. Verified by removing the
call-site guard entirely and running the whole suite — green, so the fixture is
carrying it, and the call site keeps a pointer rather than a second copy of the
explanation.
The other four:
- two service tests were the only ones in test_platform.py that never set
`sys.platform` to "win32", so they hit "service mode is Windows-only";
- test_apps_enabled_policy expected `["chat"]` where `roster.enabled_apps`
inserts "files" at the front on read (and `ops.set_enabled_apps` on write),
because Settings is the one way back if every app were turned off. The code
is right; the assertion predates the guard, and is now ["files", "chat"];
- test_invite_then_join_delivers_the_gek passed a bare Path as a group's
`roots` two lines below building a RootSet for the transport. The handshake
died on `'PosixPath' object has no attribute 'describe'` and answered `error`
— scaffolding that never followed the move to several named roots (draft v6,
change 1).
1081 passed, 4 skipped, 0 failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`thumbs` holds every generated thumbnail, every TMDB poster and backdrop, every
Cover Art Archive image and every cached audio transcode. Rows were removed only
when their source file left every group's index (`prune_file`), so a library
that merely changes over years grew this database with nothing to bound it.
Nothing in it is precious — every row is keyed off a value the node can
re-derive — which is what makes eviction the right answer rather than a bigger
disk.
512 MB, evicted on write (a cache only grows when written to; a timer is one
more thing to own and get wrong). `used_at` is marked on every read, including
the lookup by synthetic id that `_fetch_and_cache_poster` makes on every visit
to a poster grid — without that, the images shown most often would be the
coldest rows in the table. A single blob larger than the cap does not empty the
table for nothing.
The migration is the part that touches deployed nodes. `CREATE TABLE IF NOT
EXISTS` adds missing tables and never missing columns, so `used_at` would have
reached a fresh test database and never a real one. `_migrate()` does the
ALTER TABLE and seeds existing rows with "now" rather than 0 — otherwise the
first write after an upgrade evicts the whole cache, a correct-but-hostile
reading of "least recently used" for rows whose age nothing recorded.
The index on that column lives in `_migrate()`, not in `_SCHEMA`: run from the
schema script it executes before the ALTER on an existing database and fails,
which would have been every deployed node refusing to open its cache on the
first start after upgrading. Found by the migration test.
Verified against a real node's database, rebuilt into its pre-migration shape:
rows preserved, column present, seeded, index created, reopening harmless.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`save:abort` deleted a cancelled download, but nothing covered the application
being quit, killed or crashing mid-transfer: the write stream was abandoned and
a truncated file kept the final name — the exact thing save:abort's own comment
calls worse than no file at all, because it looks complete to whoever opens it
next.
Downloads go to `<target>.part` and are renamed after the stream has flushed,
which is the convention the node already uses for uploads (`_do_file_upload`).
A crash now leaves a self-evidently unfinished file. `before-quit` also clears
any `.part` still open, synchronously — it does not wait for promises — so a
deliberate quit leaves nothing at all.
Verified by hand: .part during the transfer, survives SIGKILL, renamed on
completion, and gone after a cancel or a clean quit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`terminate()` signals the parent only. Chrome's zygote, renderer and gpu
children outlive it by a moment and go on writing into the profile, so rmtree
walked a directory that gained a file between its readdir and its rmdir and
raised "Directory not empty". The probe exited non-zero, and every test in the
file errored at setup — intermittently, roughly one run in three, for a reason
nowhere near the chat code they were testing.
TemporaryDirectory(ignore_cleanup_errors=True) in all four probes that own a
profile: a few bytes left in a throwaway directory are harmless, failing the run
is not. `proc.wait()` after `kill()` was also missing — a killed process still
has to be reaped.
layout_probe.py never cleaned up at all (mkdtemp, no removal) and never waited
for Chrome; it leaked a profile into /tmp on every run.
Ten consecutive runs of test_chat_send.py are clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`pipelinedDownload` with no writable allocates `new Array(totalChunks)` and
keeps every decrypted chunk, so whatever `_openDownloadTarget` returns null for
is held whole in RAM. That floor had no upper bound: the
`!window.showSaveFilePicker` branch returned null at any size, so on a browser
without the File System Access API a 20 GB film went to memory whenever the
streamed path did not answer. Nothing logged, nothing refused; the symptom was
the tab dying, with no error attributable to this code.
MEMORY_CEILING is 100 MB and every `return null` in that chain now goes through
a guard that throws above it. The refusal names the size, the limit and why the
streamed path declined, and lands in the transfers panel as a failed transfer
rather than in a console nobody opens.
This is a guard, not a limit on what can be downloaded: with the streamed path
primed and retried (previous commit), a file of any size still goes to disk
progressively on every browser. Two things had to change for that to be true:
- the streamed path is now tried in "ask" mode too, for a file over the ceiling
on a browser with no Save As of its own. The mode decides whether to show a
dialog; it was silently deciding whether a film could be downloaded at all;
- FilePreview had no size check whatsoever — a multi-gigabyte PDF or .csv was
fetched whole, and the text branch decoded all of it to keep 500 000
characters. It refuses above the same ceiling and offers the download.
ZIP_MAX_BYTES (512 MB) and the ceiling do not contradict: the archive limit
bounds the archive, the ceiling bounds what may be built in the page, so a
400 MB zip is allowed when there is somewhere to stream it and refused when the
only route left is memory. The build-in-memory confirmation only appears below
the ceiling now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On Firefox and Safari the service worker is the only unbounded way to write a
download to disk: neither has the File System Access API, and OPFS is not a
substitute — measured on Firefox 154, its quota is exactly 10% of the volume's
size (389,233,459 bytes of a 3,892,334,592-byte volume, refused to the byte),
which a film exceeds. So when this path declines, a large download has nowhere
left to go, which makes its reliability a correctness property.
Four ways it declined, all of them avoidable:
- it was registered inside the first click on Download, so that click paid
install, activate and claim while somebody watched a button do nothing;
- `_swReady` cached a null for the life of the page. One slow first click left
the tab unable to stream anything again, curable only by a reload nobody knew
to do. Only a successful controller is remembered now;
- control was waited for with a 3 s cap. It is 15 s, and a page that is active
but not controlled asks the worker to claim again (`mbdl-claim`) instead of
declaring the path unavailable;
- a missed navigation gave up at once. It gets a second attempt with a fresh id
and iframe, the failed one torn down completely first.
Also closes a MessagePort leaked per download, and gives the reason a name
(`lastStreamFailure`) so a refusal can say what happened. The timeouts became
parameters: the defaults are the production values, no caller passes any, and
the tests do not spend a minute waiting.
`openTarget` gets an unrelated but adjacent fix, in the same file: it creates
the destination with `getFileHandle({create: true})`, so an empty file exists
before the first byte, and `abort()` leaves the target untouched — every
cancelled download left a 0-byte file behind, and since `freeName` avoids
collisions, three cancels left film.mkv, film (2).mkv and film (3).mkv, all
empty. Its `abort()` now removes the entry. Safe here and only here, because
`freeName` guarantees the name was not taken: the `showSaveFilePicker` path
must not do the same, where the person may have picked an existing file whose
contents `abort()` correctly preserves. Verified by hand in Chrome.
test_streamed_download_reliability.py runs the real module under Node against a
stubbed browser — it fails if the null is cached again.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HCGdheDLxGReuKHga3BtST
|
| |
|
|
|
| |
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
ensureTray() was reachable only from the window:minimize-to-tray handler, so
the indicator did not exist until you had already hidden the window into it.
That is backwards on both desktops — most of what a tray is for is finding an
application that is not in front of you — and on Windows it read as the app
having no tray presence at all.
Created during app.whenReady(), after registerBridge() and before
createWindow(). The order matters: buildTrayMenu reads the nodeService that
registerBridge assigns, so the other way round puts the Start/Stop entry on
the menu one five-second poll late.
The menu's labels were the one thing that came *from* the minimise call, since
the main process has no i18n. A new tray:labels IPC (platform.setTrayLabels)
carries them instead, sent from the renderer's boot once initLocale() has a
catalogue; a language change reloads the page, so the same call covers it. The
window between launch and that first message shows TRAY_FALLBACK, in English.
§5.10's two platform gates become one — trayOS() in main.js, which every tray
path calls. test_desktop_shell.py's existing test is rewritten against it and
two are added: the launch ordering, and that no tray path tests
process.platform inline instead of calling the gate.
Windows still files a new tray icon under hidden icons until the person drags
it onto the taskbar. No API promotes it; documented in WINDOWS-PORT.md §5.11
rather than worked around.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The flags were regional-indicator pairs, which Windows has never shipped a
glyph for: Segoe UI Emoji renders the pair as its two letters, so the menu
read "GB", "FR", "NL" down the left-hand side instead of a flag. It only ever
looked right on a system with an emoji font that draws them, such as GNOME's.
Names only, rather than vendoring ten flag images. A flag is a country and
not a language anyway — none of them is the right answer for "Português
(Brasil)" — and each name is already written in its own language, which is
the self-identifying signal the flag was standing in for.
The `flag` field is removed from LOCALES rather than left unread; app.js's
menu was its only consumer. The submenu keeps its indent from
.user-menu-sub's padding, not from the icon box that is now gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"private" and "invite" were on every card because they are the default for
every group on the hub — two badges that never varied and so never told
anyone anything. Who hosts a group does vary, and it is the thing that tells
two groups of the same name apart: the hub enforces name uniqueness per owner
account, not globally.
- The "My groups" cards (app.js) drop both badges and render the name through
GroupName, which puts the owner beside it as a smaller "@handle" — the same
way the group header and Explore already write it. "admin" stays; it varies.
/v1/groups/mine already returns owner_username, so no API change.
- Explore drops its join_policy badge for the same reason. An open group still
announces itself through the Join button, which is the useful half, and its
@host was already there.
- .group-card h3 .gn-owner is set to 0.8em: a card's h3 is reset to 1em, so
the 0.55em meant for a page heading rendered the handle at about half the
body size — smaller than the name, as it should be, but past readable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It is the one section in group settings that is open on arrival while having
nothing to say in the ordinary case: the device you are reading it on is
already linked, and the panel exists for two rare errands — approving another
device, or removing one.
Uses the CollapsibleSection the sections around it already use, with
defaultOpen={false}; the component renders the same .settings-section
wrapper, so nothing about the layout moves.
The prompt that *is* actionable — "this browser is not linked to this node
yet" (device.add_title) — lives in group-page.js and is untouched, so a
browser that genuinely needs linking still says so on arrival.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
An arbitrary ceiling, not a technical one: the zip writer streams and holds
one chunk plus a record per file, so it would happily produce a hundred
gigabytes. Past half a gigabyte the honest answer is a subfolder at a time,
or the files individually.
Enforced in file-utils.js's downloadDirectory, which is the one
implementation behind every zip button — Files' single folder, Files'
multi-folder selection, and the Photos album button (docs/photos.md §3).
- Per directory, not per selection: Files zips a whole multi-directory
selection in one click, so an oversized folder is refused and its siblings
still download.
- Before _openDownloadTarget, so no save dialog opens for an archive that is
never going to be written.
- The bound is strict, so a folder of exactly 512 MB still goes through.
- Counted in the 1024-based units formatSize already prints, so the number in
the refusal is the number in the constant.
group.zip_too_large in all ten catalogues. test_zip_size_limit.py runs the
module under Node and pins the refusal, the inclusive bound, and that nothing
is asked or started when a folder is over.
The user guide's "a 40 GB folder costs 40 GB of disk" is no longer true and
now documents the cap instead.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The dashed warning outline and the "?" flag say a TMDB lookup ran and came
back with nothing — that is how an operator finds the titles worth a "Fix
match". With TMDB switched off for the group no lookup is made at all:
media_meta_req answers confidence 0 for the whole library by design, so the
poster grid drew every card as a failure of the very thing the operator
chose, with no poster ever coming to clear it.
Poster mode now shows the file's own thumbnail plainly in that case, the
same as Flat does, and keeps the badge for what it was built for.
- PosterCard takes tmdbEnabled and gates both the outline and the "?" on
`tmdbEnabled && !confident`. The title still falls back to the parsed
filename exactly as before, and the ready/imageReady anti-flash gating is
untouched.
- VideoDetailModal gates its "no confident match" paragraph the same way. A
show still opens that modal with TMDB off — it is where the season list
lives — so the card alone would have left the claim one click deeper.
- Its operator-only "Fix match" / "Rematch" buttons are gated too: both act
on a match that does not exist, one opening a search the node answers
empty, the other dropping a cached match never made. Those buttons are the
only entry points to TmdbSearchOverlay and doRematch.
- Both props default to true, so a call site that forgets one keeps the badge
rather than silently losing it.
search-page.js is deliberately left passing `enabled: true`: a cross-group
view has no single group's switch to read, and it never showed the operator
buttons anyway (isNodeAdmin={false}).
Recorded as §10.7 in docs/mediacenter.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V8EDjk6pkYZrCbo63m2x87
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found on two live machines within minutes of deploying: every message from the
person running the node arrived at every other member under "this account is
using a key you have not seen before".
An operator's authority is node-wide and is recorded in `members` with an
**empty** group_id — `is_authorized` has always said so, in a clause written for
exactly that. `group_devices` spelled the rule out a second time as
`WHERE m.group_id = ?`, which excludes them. So the operator was absent from the
roster relayed to members, no chain could reach their device key, and Tier 2
reported the most ordinary event there is — the operator talking in their own
group — as a key substitution.
A notice that fires on normal use is worse than no notice: it is the one people
learn to dismiss, and §4.8 budgets exactly one for the whole feature. That makes
this a defect in the property, not only in a query.
The clause now lives once, as `_MEMBER_OF_GROUP`, shared by both callers so they
cannot drift again. `DISTINCT` because an operator who is also an explicit member
of the group matches both halves of it. `get_member` is untouched: it is a raw
lookup and its callers already fall back to `get_member("", user_id)` themselves.
Three tests, and the first fails against the old query with the reported
symptom: the operator appears in the roster of a group they host and
`is_authorized` agrees; an operator who is also a member is listed once; a
revoked one comes back through neither.
No stored state to clean up — nothing was written to a client's pins when the
account was missing, so the notice stops as soon as the node serves the roster
correctly.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TZZxYjz8YeWRz13xDi8LJr
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Chat messages have been signed by the sending device since MNP 2.0, but a
reader had no way to know that the device belonged to the account the node
named: the signature proved *a device*, and `sender_id` was still the node's
word. This closes that for any account a client has already seen.
**What was blocking it was not effort — the evidence was not being kept.**
`_do_device_add` verified the countersignature that admits a second device and
stored only `added_by_pk`: *which* key approved, never the proof. And
`device_add_transcript` binds `nonce_node`, the approving connection's handshake
nonce, so even a stored signature was unverifiable by anyone who had not been on
that connection. `identities` gains `add_sig`, `add_nonce` and `add_ts`, added
before the migration's early return — which fires on every roster widened since
2026-08-18, i.e. all of them, so putting them inside it would have meant they
never arrived.
`group_roster_req`/`resp` relays, sealed under a new groupbox purpose and
answered to **any member of the group**, every live device of every active
member with the evidence that admitted it. The node decides nothing: it hands
over evidence and the client walks the chain from each account's root outwards
(`_verifyRoster`). That is deliberate — the node is the party the property holds
against, so it is not asked to assert trust.
Two holes the tests caught while this was being built:
- "no signature" was being treated as a trust root, so a node that writes the
roster could put any key in an account's row and have it laundered straight
into the verified set. A root is a device that names **no** countersigner.
- pinning only the verified subset at first sight raised "key changed" on
legitimate second devices whose countersignature predates this change. First
sight pins everything the node says, because that is what trust-on-first-use
means and an alarm that fires on normal events stops being read.
The property, and it must not be rounded up: **once a client has seen an
account, a node that later substitutes a key for it is detected. Nothing is
gained at first sight**, where there is nothing to compare against — the same
boundary `per-node-identity-v1.md` draws, unmoved.
The cost, stated because it is real: the roster is member-visible, so every
member learns how many devices the others hold and their public keys. It stays
inside the group, the hub is not involved, and it is scoped per group. A member
who cannot see the keys cannot check them.
User-visible surface: one notice, "this account is using a key you have not seen
before", in ten languages. Nothing else.
16 tests — 7 on the node (the evidence is stored, it verifies from the roster
alone, a fabricated device carries none, another group's members are not
disclosed), 9 running the shipped `_verifyRoster` under node against rosters
built by the shipped Python: a chain of three in any order, a signature by the
wrong key, one for another node, one for another account, and two fabricated
devices signing each other admitting nothing.
Tier 3 (operator-signed roster attestation) stays deferred, with nothing
depending on it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TZZxYjz8YeWRz13xDi8LJr
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`chatbox.py` said per-device subkey derivation means "two devices never share
an AES key". That is false in the deployment that exists, and stating it hid
the reason the design is actually safe.
Two clients of one account on one node normally hold the **same** identity key:
a second browser fetches the keypair bundle from the node and recovers the
existing key rather than minting a new one, and so does a fresh Electron
install. Device linking — a distinct key, countersigned — is the exception, not
the rule, which is why nobody is ever asked to pin anything when they open a
second browser. So two clients routinely share a device key and therefore its
chat subkey.
What makes that safe is the nonce, not the derivation: 96 random bits, never a
counter. Two independent senders under one key collide only on the birthday
bound, unreachable at chat volume; two independent senders advancing one
*counter* collide immediately, which is precisely what C1 and §15.0b are about.
The true property is "no mutable sending state at all" — the hazard removed
rather than partitioned — and the design therefore degrades correctly into the
deployment as it is, where a chain-based one would have failed silently on the
day someone opened a second tab.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TZZxYjz8YeWRz13xDi8LJr
|
| |\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Both sides landed a breaking MNP change and both called it 2.0, which is right:
the sealed upload, the removal of `stream_seg` and mandatory chat encryption
share one flag day. They are recorded as one version in `__init__.py` rather
than as a race between two.
The resolutions that were decisions rather than mechanics:
* **`MNP_MIN_SUPPORTED` moves to "2.0".** The sealed upload alone was a
*confined* break — a 1.x peer could still connect, browse, download, stream
and chat, with only its uploads refused by `upload_not_sealed` — so the floor
deliberately stayed at "1.0". Mandatory chat encryption ends that
confinement: a 1.x peer can neither produce a sealed chat message nor read
one, so it would connect, look fine, and be unable to say anything. Refusing
it at the handshake is the honest form. The per-message `upload_not_sealed`
path is untouched and still right if the floor is ever lowered.
* **`sendChat` throws on an `error` reply**, from origin, applied to the sealed
send. It matters more after this change, not less: the node now refuses a
stale epoch, a malformed envelope and a device claim that is not the
connection's own, so there are three new ways for a message to be rejected
and none of them may look like a message that was sent.
* **`req_id` supersedes the per-type routing** this branch added for
`chat_keys_resp` and `device_hello_ack`. Both blocks are kept beside the
existing `chat_hist_resp` one, for the same stated reason — a node too old to
stamp — and their comments no longer claim to be the mechanism that closes
the class. `req_id` is.
* **`chat_send_probe.py` is rebuilt on origin's structure**, not beside it: two
scenarios, a stub that stamps `req_id`, `music_meta_req` as the older pending
request. The encrypted path is layered on — a real Ed25519 device key
generated in the page, and a `chat_keys_resp` sealed by the shipped Python,
because a payload the page built itself would prove only that the page agrees
with the page.
* **`test_reply_correlation.py` now sends a sealed message.** Its subject is
which of the two messages leaving that handler carries the id; plaintext chat
was only the fixture, and the node refuses one now.
* `groupbox` keeps both new purposes (`upload`, `chat_keys`); `protocol.py`
keeps origin's removal of `STREAM_SEGMENT` and this branch's correction of
the "Double Ratchet message" comment on `CHAT_MESSAGE`, which was wrong when
it was written and is wrong differently now.
Full suite on the merged tree: 1993 passed, 11 failed — the same 11 that fail
on a pristine checkout (2 Windows service tests, 1 apps-enabled policy, 7
transcode tests that pass in isolation, and the WebRTC invite test that hangs
on its own).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TZZxYjz8YeWRz13xDi8LJr
|