setResult existed and nothing called it; the timeline already drew
scene.result. One wire short of a loop -- generate a scene, and what came
back is on the strip and still there after a reload.
The scene id is read when the run STARTS. A run takes minutes and the user
goes on clicking, so the selected scene when it finishes is not the one that
was generated.
A strip cell now falls back to its glyph when a thumbnail does not arrive,
the way the bin's cells already did -- the browser's broken-image icon reads
as a damaged file rather than a missing one.
Width, height and length belong to the PROJECT, not to a scene: a scene
cropped on the timeline and regenerated comes back at the project's size,
because the crop was a timeline decision and a regenerate is a new scene.
They sit in the Constructor beside the prompt -- the two things settled in
one sitting -- and travel to the graph on the path the prompt already uses.
No new mechanism: the pipeline declares roles at `project.video` the same
way it declares the prompt, so core still names no node and no setting. A
role whose value is a number gets a number control instead of a box; that
is the whole of the UI change. What the pipeline does not offer is not
shown, so a graph with no length input simply has no Length.
Bounds are the NODE's and stay there -- it already refuses a value outside
them with a reason, beside Generate. A project file is the one thing here
that outlives the code that wrote it, so everything read back out of one is
checked: a whole positive number, or gone.
Also, four faults the browser suite was already red with before this:
* Six tests still clicked a rail toggle that moved into the timeline head.
* At 420px the panel OVERLAYS the centre -- and the named toggles are in
the centre, so the panel covered the control that closes it. That is the
v4 trap this project forbids. They are not offered at that width now; the
View menu, which nothing can cover, is the way in and out.
* A 768px window laid out 860px wide and scrolled sideways. Root cause, and
it was never about one component: a flex item's automatic minimum is its
CONTENT, so every stack, split, panel body and frame region in the app
handed its content's width upward. A scroller that cannot shrink below
what it holds is not a scroller -- the strip of scenes made the WINDOW as
wide as the strip. Floored at the five places every layout routes through.
* A bin test counted `.cx-cell-glyph` across the whole page, so the timeline
drawing cells with the same glyph in them broke a test about the bin.
The app could run a pipeline and had nowhere to keep what it was making. A
project is an ordered list of scenes, one JSON file each, and it is the only
part of the app that has to survive a reload.
A scene owns its prompt text and the result it produced, and nothing else. v4's
Scene grew three dozen fields; each belongs to the feature that needs it. Per
scene length and fps are deliberately absent — a regenerate uses the project's
values, so a scene cropped on the timeline comes back whole.
The prompt box edits the current scene rather than holding its own copy, so
what is on screen and what a run uses cannot disagree.
app: the region toggles are named buttons in the timeline bar
v4's arrangement. The kit's fallback is one glyph on a rail at each edge of the
window, which is right for a kit that cannot know what its panels are called and
wrong here, where they have names and a bar to put them in.
A pressed toggle was styled only inside a button group, so a standalone one
carried aria-pressed and looked identical either way.
gallery.strip is selectable like its siblings: a filmstrip is the one shape
whose whole job is saying which of these you are on.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Generate was washed out on light and the fade downwards was why. Light's
accent carries WHITE text, which caps how bright the fill can go before
the label starts disappearing into it -- so the button sits AT the accent
now and DARKENS under the pointer, where white only reads better. Dark's
accent carries near-black text and has no such ceiling, so it does the
opposite: rest at the accent, lighter on hover. Measured in both, at rest
and hovered: no gradient stop under 4:1.
* A section label sat directly on the dividing line above it -- the
collapsible's body had no top padding of its own, so "APPEARANCE" read as
belonging to the divider rather than to the card under it.
* A checkbox in this app is a switch with a sentence under it, not a tick in
a list. At 4px of padding the label and its hint sat hard against the top
and bottom of their card, and a hint wrapping to two lines had nowhere to
wrap into.
* Row content is weighted to the left now: hard against the border, a card
read as leaning, because what sits opposite a control is usually the short
half of a wrapped line rather than anything at all.
And one regression caught in the screenshot: the settings row stacks below a
CONTAINER width, and that threshold was written as 300px -- which happened to
be just above what the properties column resolved to. Trimming 4px of padding
off two nested boxes widened it past the line, and every setting in that
column went back to being shredded, in a commit about something else. It is
360 now, which is what the numbers say it should always have been: a 180px
control column needs at least as much again beside it.
v4's arrangement, which is the one that reads as an editor: a zone head
holds what acts on THAT zone, so the app grows a bar of its own for what
acts on the app.
* The menu bar. Brand, View and Settings, and whether ComfyUI is still on
the other end. Only menus with something real in them exist -- an empty
File menu is a promise the app does not keep.
* The prompt is written in the CONSTRUCTOR, a window. v4 called this the
Composer; that name is taken here by the element kit, and two things
called Composer in one repo is one thing nobody can search for. It is
the longest thing a person writes and the least often read, and a
permanent third of the centre column is the room the timeline needs.
The host is built at startup and outlives every opening, so a window
that opens twice does not throw away what was typed.
* The timeline replaces it, and Generate moves into its head with what
the run is doing at the far end of the same head. No bar across the
bottom: one that spans the window belongs to no region.
Four faults found on the way, three of them mine and one much older:
* THE OVERLAY ROOT WAS A STACKING CONTEXT. It documents at length that it
must never be one -- and then set `position: fixed`, which makes one on
its own. So the whole z ladder was measured inside that single element:
an open menu at z 500 painted UNDER a button at z 1. Every overlay in
the app was one positive z-index away from being buried, and the only
reason nobody had seen it is that almost nothing claims a z-index. The
root is static and zero-sized now; every layer in it is fixed already.
* Generate's hover ran `filter: brightness()`, which lightens the LABEL
along with the fill under it -- the one thing that had to stay readable
lost contrast in both themes. Only the fill moves now, and which way it
moves depends on the theme: dark's accent carries near-black text so it
goes lighter, light's carries white so it goes darker. Measured: no
gradient stop under 4.1:1, hovered or at rest.
* A fixed-height head could not shrink below the width of everything in
it, and a head now holds a title, three controls and a status. That
min-content became the centre column's, so a 768px window laid out
878px wide and scrolled sideways. It wraps now, as v4's does.
* A CSS patch of mine silently did nothing because its anchor text had
moved and I had left the assertion off that one replacement. Every
patch asserts now.
Ran v4 side by side to see what the difference actually was, and it was
not spacing. v4 fills the window edge to edge and divides its zones with
hairlines; v5 floated rounded, shadowed cards on a visible page
background. One reads as an editing suite, the other as a dashboard.
So panel gains a second variant. "default" is still a card, for something
that floats -- inside a window, on a page. "zone" is a region of the app:
square, no shadow, no border of its own, divided from its neighbours by
the one hairline everything else uses. The workspace keeps no gutter
between regions; the preview zone is flush, because padding around a
picture is a frame nobody asked for, and its surround is letterbox black
rather than a panel colour.
The right column was two stacked panels with two heads for one thing.
It is one zone now.
And the surfaces have depth, derived from each theme's own elevation
scale so neither needs a rule of its own: a vertical fall on every band
and every raised control, the accent gradient and a bloom under a primary
button, a wash off the top-right of the page, tokenised thin scrollbars,
an accent mark on every zone title, and v4's film grain -- overlay at
3.5% on dark, multiply at 1.5% on light, and painted with no z-index so
anything that opens above the frame stays crisp.
The layout was assembled by hand and read like it. Measured at 1440x900:
the three columns started at three different heights, a panel head was 41px
when it held a control and 34 when it did not, the gap left of the centre
was 7px and right of it 19, the properties column sat 11px further in than
the others inside a border that meant nothing, and five controls in one
list had four different widths -- a <select> takes the width of its widest
option, so the row listing checkpoint filenames was 124px wider than its
neighbours.
Now: --gutter is the space between regions, between two panels stacked in
a column, and inside the bands; the head band height is the panel head
height is the collapsible head height; --control-col gives every settings
control the same two edges. The frame's bands run edge to edge and the
content is inset by the gutter, so the app has an outline instead of a
margin that changed per region.
Also here, all of it visible in the same screenshot:
* The status sat against the Generate button with 1300px of empty bar to
its left -- everything was in one right-aligned group. What the app says
is at the start of the row now, what you press is at the end.
* A panel body is a column with a gap: two things put in it by two parts
of the app had whatever space their line boxes happened to leave.
* A container query adds no specificity, so the narrow-panel rule written
above the rules it overrides lost -- and flex-basis in a column is
HEIGHT, which turned four settings into a panel two screens long.
* The theme you picked was thrown away on every load: the appearance
module applied its own declared default over the choice the page had
already applied from storage. It adopts the live value now, and a panel
is built after setup so its control shows what setup settled on.
The prompt was a slot input on a CLIPTextEncode, reachable only by opening
the pipeline window and finding the node. A slot can now say that one of
its inputs belongs on the app's surface -- a place and a label -- and the
shell offers a region with that name or does not, exactly as a module's
panel works. Core validates that a role is names and nothing else.
Values for those inputs travel addressed by slot rather than as a whole
pipeline, so the window that owns the structure and a box that owns one
value cannot disagree about the slots between them. An address that is
not there is refused: a prompt typed into a box for a removed slot has
gone nowhere, and a run that starts anyway is not the run that was asked
for.
Region stand-ins are empty states now, taken down once something mounts --
'modules appear here' left above the modules that appeared explained
nothing to nobody.
Every module panel was decoration. The values lived in the browser and
nothing sent them, so a modifier switched on was one the run never heard
about. They now travel with the pipeline on every check, and core places
them wherever a module says they belong -- announced, not known, so the
node that carries them is not privileged.
Settings with nowhere to go are said beside Generate rather than
swallowed, and the warning survives the run it is about: the status line
is redrawn by every message, so a warning put there was gone the moment
the run said Queued.
Also: a settings row in a docked panel squeezed its label to one word per
line, because the control keeps a 140px floor and took it out of the
label. Below the width where they fit side by side, they stack.
A second run replaced the first on screen with no way back. Results now
accumulate in the Assets panel across runs, and the bin owns what the
viewer shows: the session subscription drove the viewer as well, so every
progress message walked a click on an older result back to the newest.
Grid, list and icons are one collection in three shapes, sharing selection
so switching view is not a way to lose it. No <video> thumbnails, ever --
that is v4's connection-pool stall. A thumbnail that does not arrive falls
back to the glyph rather than the browser's broken-image icon.
Every modal already mounts a .cx-backdrop element over its layer, so the
stacked variant painting another one as `0 0 0 100vmax` of shadow spread was
redundant, and the area beneath it was darkened twice.
It was also the most expensive declaration in the sheet: a spread the size of
the viewport diagonal, on an element inside a fixed layer, with an entry
animation running. Removed in favour of a bounded shadow, and a test reads the
stylesheet for the pattern -- asserting on node.style in jsdom passes against
any rule at all, since no CSS is applied there.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The button opened another window on every press -- reported from a real
session as two of the same modal. Two windows over one pipeline are two drafts
of it: whichever is saved last wins, and the other was edited against a
pipeline that had already moved. One window now, focused rather than stacked,
and a fresh one after it is closed.
commit() took whatever answer arrived. Moving a node carries the whole slot
list and removing one carries an id, so the second answers first and the first
then describes a pipeline that no longer exists -- putting a removed node back
with no refusal and nothing said. The picker guarded this already; the path
that decides what gets queued did not.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ComfyUI refuses the WHOLE graph over one wrong value. v4 hit that with a LoRA
file that had been deleted: a feature that was switched off still stopped every
generation, and the message was "Prompt outputs failed validation" over a dozen
loaders. build() checked wiring and required-ness and never looked at a literal,
so check() called such a pipeline queueable.
A combo value that is not among the node choices, and a number outside the
bounds it declares, are now refused here -- naming the slot and the input.
No bounds means unknown, and unknown is not a refusal.
And when the queue does refuse one, its per-node reasons were captured by
run.js and never read: the status bar showed only the top-level string, because
describe() read `node` and the refusal path sets `nodes`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A module whose setup() threw was reported as hidden with its panel already on
screen -- appended one line too early. The file states the opposite guarantee.
A redraw destroyed every control and then rebuilt, so a row revealed by another
value -- one the first build never rendered and so never checked -- stripped
the panel and threw on the way to refilling it. Building first makes a redraw
as safe as the first build already was.
A page loaded at a narrow width recorded the panel its own one-at-a-time rule
had just closed as one the user wanted closed, so widening brought back only
the left. What the user chose is now kept apart from what is on screen.
The redraw test counted rows and passed against the bug; a control removed from
a row leaves the row behind. It counts controls.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A draft is keyed by slot id, and a slot keeps its id when it is pointed at a
different node. An unsaved steps stayed attached across a swap -- and since
most samplers declare steps, and the server only refuses an input the new node
does NOT declare, it was written on the next Save with nothing saying it had
never been chosen for that node.
The fake server in the tests did not implement replace, which is why no test
could see it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
One wide modal, a card per group, and a group that opens into a node list
beside that node/'s parameters, saved a group at a time. Groups travel on the
slot, so a pipeline keeps its own arrangement and a new group is a slot
claiming a name nobody used.
Values are drafted until Save; adding, removing, replacing and moving a node
go to the server as they happen, because only it can say whether the result
still wires up.
An unset dropdown reads "not set" rather than showing the first file as though
it had been chosen -- v4 recorded that trap, and picking the entry the box
already shows fires no change, so the one click that looks like the fix does
nothing.
New: region.stack, a frame header, a modal footer that can be refilled.
Not yet: wiring. A socket says so where a control would be.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ComfyUI spells a combo three ways -- a list where the type goes, the string
"COMBO" with the choices under options, and COMFY_DYNAMICCOMBO_V3. Only the
first was read, so every V3 node came back with no choices at all and the
window said "no files of this kind were found" over dropdowns holding seven
entries. A stock install here has 140 of the spelling that was missed.
A union type ("IMAGE,MASK") was compared as a whole string, so a MASK feeding
it was refused -- a legal wire refused, which is as bad as an illegal one
accepted. A MultiType wrapped around a number ("FLOAT,INT" with widgetType)
was read as a socket demanding a source for a field you type into.
graph.py and widgets.py had their own idea of what a type string means; both
now read comfy_types.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
I committed the widget layer without running the full suite and broke
twenty-three tests in files that have nothing to do with it.
The registry caches its scan for the life of the process. A route test
redirects MODULES_DIR at a temp directory holding one fake module, and my new
test called install_into inside that -- which scans, and caches the temp
directory as the installed set. Every later test then ran against an empty
registry: the sampling tests found no modules to install, so a modifier that
should have been applied was not, and the failure surfaced as an assertion about
pre_cfg hooks three directories away.
The test needed a node, not FunPack's nodes, and KSampler is one. The fixture
that redirects the directory now also puts the cache back, because whoever moves
it owns restoring it, and there is a test for that -- the failure it prevents is
unrecognisable from where it lands.
graph.py reads an input's TYPE, which is all a wire needs. Editing a value needs
the rest -- the choices a combo offers, the bounds on a number, whether the box
is multiline -- so that is asked separately rather than widening the question
graph.py asks. Read from ComfyUI's own INPUT_TYPES, so a node written for
ComfyUI is editable here without knowing anything about FunPack.
Caught while looking at the real output rather than after committing: deciding
"a socket is upper case" is wrong, because STRING is upper case -- every prompt
box in the pipeline came back as something nobody could type in. The honest rule
is ComfyUI's own primitive set, plus forceInput, which is a short list core
holds on purpose.
/funpack/api/nodes answers by name. ComfyUI's /object_info returns every
installed node, which on a machine with a few packs is megabytes for a question
about a dozen; a node nobody installed comes back as null rather than missing,
so "this slot points at something absent" reads differently from "I did not ask".
The dev server now registers ComfyUI's nodes and FunPack's own, through the same
function the tests use. Without it the pipeline was fourteen "there is no node
called X installed" and nothing to edit -- so the one thing the dev server exists
for could not reach the half of the app that matters. It now reports the three
problems that are real: no model, no CLIP, no VAE chosen.
The columns were a share of the viewport, which is not a measurement of
anything. A 900x900 window -- an ordinary shape -- gave the properties column
198px, in which "Loosen the starting image" wrapped over three lines and a
section heading broke mid-word; 375px put three columns side by side with
clipped headings and a preview no wider than a thumbnail. A panel is now as wide
as a settings row needs, and below 760px it stops taking space from the centre
and overlays it instead.
The element decides what is OPEN and the stylesheet decides how it is DRAWN, so
entering that mode closes both panels and only one opens at a time -- two
overlays over a narrow centre is the same problem with an extra step. What was
open is restored on the way out, and none of it touches the remembered
preference: a window being small is not the user changing their mind.
The rails paint above the overlay. Without that the panel covered the only
control that closes it, which is the exact fault this arrangement exists to
avoid, arriving by a different route the moment the panel stopped taking its own
space.
Also: the prompt was holding 38% of the height for two lines of text. The result
gets the room; the divider is still draggable.
`node --test` has no timeout by default. Two revert-checks left runs wedged on
session.test.js for forty minutes with no output and no way to tell from the
outside whether they were working or stuck.
/queue answers with queue_running AND queue_pending, and only the first was
looked at. A job sits in the second from the moment /prompt returns until the
worker picks it up -- always briefly, and for as long as it takes whenever
something is running ahead of it, which on a shared box is the ordinary case. A
reload in that window found nothing, handed Generate back, and let the same job
be queued twice: the first copy then ran unnamed, its messages failing the id
check, burning GPU time while the UI narrated the second.
Every fixture in client.test.js was cleaner than the real payload -- none of
them carried a queue_pending key at all, though ComfyUI always sends both. That
is the third time a stand-in tidier than the real thing hid what it stood in
for, so the fixtures now carry both halves, including the malformed entries a
real answer can contain.
Also renamed, because the function no longer answers the question its name
asked: queuedFor returns which half it found the job in, and a job waiting its
turn is adopted as QUEUED. Saying "working" over a run that has not started is
the same small lie as saying "cancelled" over one that has not stopped.
subscribe() delivers the CURRENT state at once, and boot.js registered it right
after wire() had deliberately disabled Generate for the length of the reattach
lookup. That first delivery drew an idle run, and drawing an idle run hands the
button back -- so on every single page load Generate looked available while the
page still had no idea whether it already had a run. The queue stayed safe,
because the generator waits on the same lookup, but a live-looking button that
answers a click with nothing is its own fault.
Fixed twice over, because either alone would rot. The subscription moved into
wire() with the rest of the wiring, and the button being held is now a state the
transport keeps rather than something a caller does once: draw() runs on every
change, and each of those was free to undo it.
The fake run had no subscribe() at all -- which is exactly why the suite could
not see this. That is the second time a fake was tidier than the real object and
hid the fault it stood in for, so it now delivers state on subscribe like the
real one, and announces changes.
boot.js is down to assembling six things and mounting the modules.
reattach() was started and not waited on, and the Generate button is on screen
before it resolves -- so on a reload during a generation the phase still read
"idle" for the length of the queue lookup, and a click in that window queued a
SECOND job. The adopt that would have found the first one then refused, because
the phase was no longer idle, and from that moment the original run's messages
belonged to nobody: not shown, not cancellable, still burning GPU time.
Generate now waits on that lookup, and says so while it waits rather than
sitting dead with no explanation. A queue that never answers still gives the
button back.
The ordering moved into shell/session.js with the rest of it. session.js's own
docstring said the faults it was extracted to catch lived in boot.js "where
nothing could be driven by a test" -- and then the fix was wired up in boot.js,
where nothing could be driven by a test, and reopened the same class of fault.
The wiring is the thing under test, so it lives where tests can reach it.
Also: the fake run's gate now starts OPEN. Held shut by default, any regression
letting an extra start() through hung the suite instead of failing it, and a
hang says far less than an assertion.
/interrupt only consults the RUNNING half of ComfyUI's queue. A job waiting
behind another one is not there, so cancelling it returned 200, logged "not
currently running, skipping interrupt", and let the generation run in full when
its turn came -- the user asked for it to stop, the UI showed it stopping, and
the GPU spent the time anyway. Two tabs against one ComfyUI is all it takes, and
one tab per generation is now the normal arrangement.
ComfyUI has a route for exactly this: /api/jobs/<id>/cancel interrupts a running
job and dequeues a pending one, atomically, and answers {"cancelled": false} for
an id that already finished. An older ComfyUI without it falls back to the
interrupt, which is right for the case people hit most.
Reading that answer matters: a dequeued job never runs, so no
execution_interrupted is ever sent, and the run would have sat at "Queued"
forever after a cancel that worked. A run that IS executing is still left to the
socket to declare stopped -- an interrupt is a request, and claiming it early
shows "cancelled" over a model still finishing its step.
The one test touching cancel drove execution_start first, so it only ever
exercised the running case -- which is the half that already worked.
Two faults from the round, and a third I introduced fixing them.
ComfyUI keys its socket table by client id and pops the old entry when a new
socket arrives with the same one, so two tabs sharing an id meant the second
took the first's updates -- the first froze mid-progress while the GPU carried
on -- and the socket's `finally` pops that id again on close, so closing the
dead tab disconnected the live one. The id lives in sessionStorage now: it
survives a reload of this tab, which is what reattach needs, and does not reach
another tab. Duplicating a tab copies sessionStorage and is still a collision;
said in the comment rather than pretended away.
Generate only went dead when the RUN reached "queued", and reaching it takes a
round trip to ask what to queue. For the whole of that round trip the button was
live, so two clicks each posted their own /prompt: two jobs burning GPU time,
with the UI following one.
Both lived in boot.js, which runs on import and needs a document, so none of it
could be driven by a test. That wiring is now in shell/session.js and tested --
and the first version of those tests used a hand-rolled transport whose draw()
did nothing, which hid the fault the change introduced: the real draw() writes
the status from the run state, and a refused attempt leaves the run idle, so the
reason was overwritten with "Ready" in the same breath as it appeared. The tests
use the real transport, and a refusal now outlives the redraw that follows it.
The fix a commit ago buffered messages only while QUEUED, which is the window a
page creates for ITS OWN run. Reattach opens a second window nobody covered: the
socket is live, the phase is idle, and the queue has not yet said which run
belongs here. Anything arriving in that gap fell straight through and was
dropped -- and a run that ended inside it had already left `queue_running` by
the time it was asked about, so nothing found it either. The GPU produced a
picture, the app said Ready, and the result was gone.
Held whenever the id is unknown, not only while queueing. Then two questions on
load rather than one: what is still running, and -- if nothing is -- which of
the runs this page actually SAW on its own socket was ours, answered from
ComfyUI's history by the client id it recorded. Only ids seen during this page
load are asked about, so a result from an hour ago cannot be resurrected as
though it had just arrived.
Every reattach test called adopt() before sending anything, so none of them was
ever in the window that adopt() exists to close.
Three faults, all living in the length of the /prompt round trip -- a real
network delay during which the socket is already live and everything ComfyUI
broadcasts arrives.
* Cancel in that window POSTed `{"prompt_id": null}`, and ComfyUI reads a
missing prompt_id as a GLOBAL interrupt: it stops whatever any client is
running. On a shared box that is someone else's generation, and the clicking
user's own request may not even be queued yet. The cancel is remembered and
sent once the run has a name.
* `mine()` treated "no id yet" as "everything is mine", so a stranger's
execution_success and images flipped this run to Done -- showing another
client's picture, for a run that had not started. Messages arriving before the
id are now held and sorted out once it is known; dropping them instead would
lose a run short enough to finish inside the round trip. The buffer is bounded,
because on a shared server it fills with other people's traffic.
* The socket only opened at Generate, so the reattach the client id exists for
never happened: after a reload the app sat at Ready while its own generation
ran on, and pressing Generate queued a SECOND job beside it. It listens at
load and asks ComfyUI's own queue which run belongs to this client -- asked
rather than remembered, because a remembered id reattaches to a run that may
have finished while the page was gone.
Every existing test awaited start() before sending a message, so none of them
ever entered this window. The new ones hold the /prompt response open, which is
what a round trip actually is.
ComfyUI's own queue and its own socket, not a second one alongside them. A
FunPack queue would be a second thing tracking what is running, and the one the
user believes is whichever spoke last -- so this speaks to /prompt, /interrupt
and /ws as any ComfyUI client does and holds only a view of the server's state.
The graph is asked for, never assembled here, for the same reason.
The client id is remembered: ComfyUI addresses progress to the client that
queued the run, so a reload with a fresh id hears nothing about the generation
already in flight -- the run continues, the GPU time is spent, and the app says
Ready.
Faults found on the way, each in code that already had passing tests:
* `[hidden]` was inert across the whole kit. The UA rule is element-level, so
`.cx-btn { display: inline-flex }` beats it -- Cancel was marked hidden, sat
there, and nothing reported anything. jsdom cannot see it: the attribute was
always correct. One reset rule, and a browser test over four element kinds.
* A panel passed no height to its body and a split pane passed none to its
panes, so the viewer asked for 100% of `auto` and collapsed to a single line
inside a panel that was the right size around it.
* subscribe() delivered the first state outside the guard every later delivery
gets, so one bad listener took down whoever was wiring the UI up -- at
startup, before anything existed to report it.
* A whole tests directory was never run: the test script globbed
composer/tests only. `node --test` discovers them itself now, so there is no
list to forget.
* `empty` was not in the content rule's key list, so three elements were never
asked to prove their empty-state text comes back as text rather than markup.
The dev server now mounts the REAL routes -- register() already took a route
table -- instead of reimplementing them over http.server. It answered POST
/api/pipeline with 501, so the edit-and-queue path could not be looked at, which
is the one thing the dev server exists for.
v4's Simple mode is the Editor with about ten `display: none !important` rules
over it, which is not a smaller app but the same one wearing a mask. It broke in
the way that arrangement always breaks: a collapsed column was marked [hidden],
nothing can slide a display:none element into view, and the panel button did
nothing at all and said nothing. So the same shape is built the other way round
-- a region that does not exist yet is ABSENT, and the toggles live in rails
that are siblings of the panels, never inside what they hide.
Two kit elements, because both are layout primitives rather than app code:
frame.app establishes the one definite height everything else fills, and
workspace.docked is the centre-shrinks-between-two-panels arrangement.
Three faults found by looking at it rather than by reasoning about it:
* The prompt was in the DOM with no pixels. A vertical split inside a parent
whose height came from its content resolves its second pane to zero, and
nothing anywhere reports that.
* The whole app rendered as a strip at the top of an empty window: `100%` of a
`min-height` is `auto`.
* Two fixed-width panels plus their rails are wider than a 480px window, and the
page scrolled sideways -- taking the transport bar with it.
The collapse is deliberately not animated: a remembered-closed panel would
animate shut on every load, which reads as the app doing something rather than
as a preference being restored.
Also: jsdom gives a document with no url an OPAQUE ORIGIN, where reading
window.localStorage throws on the property access itself. Everything that
remembers a preference was silently taking its fallback path in every jsdom
test, so "it is remembered" could not be tested at the cheap layer and looked
like it worked.
Reverting chain start-up, node schema reading and decode to log.failed left all
420 tests green -- so the fix was pinned in three of six places and free to rot
in the other three. Testing log.broke on its own never covers this: what recurs
is a call SITE reaching for the wrong function, and a unit test of the right one
passes either way. Each site is now exercised through the code that logs.
One finding, confirmed, and proven by reverting each site in turn.
The wording fix a commit ago was written by hand in two files and the same
mislabel was still sitting in four others -- traits, chain, modifier install and
node schema all reported a thing that loaded, recognised its work and then
raised as one that "did not load", sending a reader after an import error that
never happened. Three hand-written strings would have drifted, so log.broke()
now says it once and every site calls it.
Two of the new tests were vacuous when first written -- "nothing said did not
load" is also true where nothing ran at all, and picking the first provider
found lands on one whose traits exclude the test's model. Both assert the line
is present before asserting the wrong line is not.
One finding, confirmed, plus four sites it did not name.
`log.failed` means "did not load", and both the latent and the decode paths used
it for a provider that loaded, claimed the model and then broke while working --
"did not load -- RuntimeError: ..." sends whoever reads the log looking for an
import error that never happened. The levels were separated for exactly this.
One finding, confirmed.
Two corrections to the commit before this one.
A rank-1 latent's length is in the NODE that builds it and never in the format:
core turns seconds into samples with a 44100 rate and a 2048 hop that
latent_format does not carry, and the same rank means tokens of geometry for
Hunyuan3D. Deriving `length` samples gave StableAudio a latent of 1 where core's
own default is 1024 -- three orders of magnitude short, reported as success.
Refused now, with the reason and the way out: a module for that model.
The sweep claimed to skip H3 and did not. H3 has a rank-3 latent, so it was
derived and passed at 32 channels for a video branch that has 24 -- a test
passing on a model it is not correct for. So the previous message's "produces
exactly what core's nodes compute for every single-tensor model" was wrong on
both counts, and the sweep now says what it actually proves: the derivation
agrees with what a model PUBLISHES, which is a floor and not a certificate.
Rejected from the same report: an absent model module letting H3 fall through to
the generic shape. Core cannot detect that without naming models, which is the
one thing it must not do; a module that fails to import is already reported as a
Warning, which is the level that can see it.
LTX and WAN needed no module: both publish everything the derivation reads, and
it produces exactly what EmptyLTXVLatentVideo and Wan22ImageToVideoLatent
compute, by a different route. Pinned, along with a sweep over all 101 configs
ComfyUI ships -- rank, channel count and no empty axis -- so a family that
arrives after this was written is covered or fails here rather than at the first
sampling step.
H3 remains the only model needing a module of its own, because its latent is not
a single tensor.
Detail declared no traits, which read as "works on everything" and meant "is
offered on everything and drops itself on half of them". Core's unsharp mask
pads in two dimensions and raises on anything with more, so on a model with a
time axis the toggle was present, switchable, and inert -- a knob that does
nothing, which is the fault this project has shipped once already. It now
declares spatial_latent and is absent where it cannot work.
The guard's regression case survives, because the guard is not about a module
being wrong about itself: a hook meets the tensor it was handed. The stub model
now announces a latent rank, without which every trait-declaring modifier was
filtered out and those tests passed by installing nothing at all.
Round seven, one finding, confirmed in part -- the maths is deliberately not
being fixed, the advertisement was the defect.
shape_problems() reads the slots array. Everything else in the body -- the body
being an object at all, which slot an edit targets, which node it replaces it
with -- was read straight off whatever JSON arrived. `[]` reached `.get` as an
AttributeError; `{"slot": ["a"]}` reached an `in` against a dict as an
unhashable key. Both answered 500 in plain text, which is the refusal nobody
can read that this route exists to avoid.
Round six, two findings, both confirmed.
The source of every link was looked up in a dict rebuilt from the slot list on
each lookup, so validating a chain of N slots did N rebuilds: 20000 slots took
21 seconds. build() runs on the event loop, so that cost was paid by every
other request the server was holding, not only by the caller. Built once.
Round five, one finding, confirmed.
Round four, both findings confirmed against the code.
A cycle is the one wrong graph that passes a per-link check: every type
matches, every id exists, every required input is filled. build() returned it
as queueable, and ComfyUI caught it at execution instead -- a run accepted and
then killed from inside the executor, with the reason in a place the app has no
reason to be reading. Walked with an explicit stack: slots come from a request,
so a long chain must produce a refusal, not a RecursionError.
The window clamp measured size against the viewport and position against the
viewport minus a grabbable strip. Those are two budgets that do not add up, so
a 900px window remembered at x=1000 opened at x=720 of an 800px screen with its
close button and grip past the right edge -- while every "a strip is reachable"
assertion still passed. Dragging keeps the loose rule, because putting a window
half off the screen on purpose is ordinary; opening, restoring and a shrinking
viewport are not placements anybody made, so the window is placed whole when it
fits.
A malformed slots payload crashed the route with a TypeError and answered
500 in plain text. The round-two fix handled the ABSENCE of slots and not
their SHAPE, so reading slot["id"] off a string went straight through -- and
a refusal the app cannot parse is the failure this endpoint exists to avoid.
The shape check is its own function now, because a malformed REQUEST and an
unfinished PIPELINE are different things and get reported separately.
And a window could still be made unreachable, by resizing rather than
moving. Round two clamped position and said "a window can never be put
somewhere it cannot be recovered from"; size was never bounded, so a corner
drag grew it to 4937px in a 900px viewport and carried its own close and
maximise buttons off the screen, leaving Escape as the only way out.
Capping size to the viewport was not enough either: with x above zero the
right edge still left. Resizing is now capped to the room available from
where the window sits, which is a deliberately different rule from dragging
-- a window may be DRAGGED partly off, because a grabbable strip is
guaranteed and that is ordinary behaviour, but it may not be GROWN past the
edge it is anchored to.
Two rounds of fixes, two rounds of holes in those fixes. Both found only by
attacking the code as it stood after the previous fix rather than stopping
at the report that prompted it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A floating window remembers its position per id, and only the DRAG handler
clamped. So a window moved to the right of a wide screen and reopened on a
narrow one opened entirely off the viewport: no titlebar on screen to grab,
nothing to drag back, and the position remembered -- unrecoverable without
clearing localStorage by hand. Reproduced at x=1400 in an 800px viewport.
The clamp lives in apply() now, so every path goes through it: opening,
restoring a remembered box, coming back from maximised, and the viewport
shrinking under an open window. Also guarded for a viewport narrower than
the grabbable strip, where the old bound went negative and pushed the window
further off rather than back on.
jsdom cannot see any of this -- getBoundingClientRect is zeros there, so a
window off the right edge measures identically to one in the middle. The
three tests are Playwright, and removing the clamp fails two of them.
And POST /api/pipeline could not accept an empty pipeline: `slots or
default()` resurrects the built-ins, because an empty list is falsy. A client
that removed every slot got them handed back. Absent means "use the default";
empty means "there is nothing", and only one of those is a fallback.
One of the tests I first wrote for this asserted len(x) < len(y) + 1, which
is true of almost anything. It counts the pipeline at every step now and
fails if it ever grows.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The plan asked for jsdom AND Playwright from the start and only jsdom was
ever built. jsdom has no layout engine: getBoundingClientRect answers zeros,
container queries never fire, a theme is an attribute rather than a computed
cascade. So every claim about POSITION or SIZE in this kit was unchecked, and
the unit suite could not have noticed either of these.
A popover bounded its height and not its width. `max-height: 60vh` was there
from the start; there was no max-inline-size at all, so a long unbroken
string -- a file path, a model filename -- made the popover wider than the
screen and anchor.js could not save it: it can move a box, not shrink one.
Measured at 1436px inside a 1280px viewport, off the right edge in all four
corners.
A horizontal split could not fit a narrow screen. The first pane carries an
inline size and `flex: 0 0 auto`, so it would not shrink, and the page
scrolled sideways at 360px -- 407px of content in a 360px viewport. It wraps
now, which stacks the panes, rather than a media query: the kit sizes itself
from its container and not from the window.
Nine layout tests: every overlay corner, an overlay taller than the viewport,
both themes resolving every token they use, light actually being lighter than
dark, and no horizontal scrollbar at five widths. `npm run test:all` runs both
suites.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The sandbox was a claim the code could not make. `panels.js` said "no DOM
node, no document... a module cannot style anything because it is never
handed anything that could" -- while handing over `composer`, whose factories
return handles wrapping a real element. A module could take one, style it,
and reach the whole document through node.ownerDocument. No lint to evade,
just the intended API.
No module ever used it. Panels are rendered from the DECLARATION, and a
module that wants to draw something should be declaring it. So it is gone,
and the guarantee is now true as stated: the kit hands a module nothing that
touches the DOM.
What is NOT claimed any more: that a module CANNOT reach the DOM. A ui.js is
an ES module in the page realm and `document` is a global. kit_lint is a
discipline check, not a boundary -- globalThis["docu"+"ment"] defeats every
regex and always will. It now catches the bracket-access spellings, which
were the first thing tried and went straight through, and says plainly what
it is and is not.
The test that was supposed to protect this asserted "no value IS a node"
while a value that MAKES nodes sat in the same object. It checks what it
names now.
And the overlay compaction test computed a before-snapshot and discarded it
with `void before`, checking only that survivors do not collide -- true even
if nothing ever moves. Two attempts to fix it failed against CORRECT code
(earlier tests leave live claims, so the rung has a legitimate gap; and some
demos close themselves, so closing them again is a no-op). It now opens two
overlays of its own so the invariant runs every time, and removing notify()
from zlayer fails three tests instead of none.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Full control did half of what it said and never said which half. It disabled
the mid-run crash guard and left every value refusal in place, so the stated
purpose -- a one-step schedule, a width the model cannot use, let them and
warn -- did not work. Worse, a rejected value was quietly replaced by the
default, so asking for 999 gave 0.4 with a warning about 999.
Now one switch turns off both: the value is passed through, every problem is
said as a warning naming that the consequences are the user's, and a run made
without the guards is identifiable as one afterwards. A key NO module
declares is still never kept, whatever the setting -- nothing reads it, so
keeping it cannot honour the request, only disguise the typo.
core/graph.py was thoroughly tested and called from nowhere: the claim that
a built-in node can be replaced or removed was a property of a library
nothing invoked. It has routes now, and tests that drive them over a real
socket -- the modifiers node removed and the sampler rewired, the VAE loader
swapped for core's, and both refusals coming back as reasons.
Those routes separate two things the old shape conflated. "refused" means
the edit did not happen. "incomplete" means it did and the pipeline still is
not ready -- an unset file picker on a fresh install is the normal case, not
a failed edit, and showing them together would say the wrong thing about both.
And build() could not see a missing REQUIRED input at all, because the
schema lookup flattened required and optional into one dict. A slot with no
model file built clean and failed only once queued.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Stopping a generation is a normal thing to do -- a bad seed, a wrong prompt,
a rental minute -- so it is not an error path that can be left approximate.
A leak here is invisible: the run AFTER a cancel just behaves slightly wrong.
The sampler says a run was cancelled and re-raises. It is not ours to
absorb; ComfyUI ends the queue item. But a run that simply stops with
nothing said is indistinguishable from one that broke.
Verified against the real engine: a 40-step generation interrupted mid-way
reports execution_interrupted, and the next one completes as run 2. Tests
cover what must be left behind, which is nothing -- no wrappers on the shared
model after five cancelled runs, the sharp anchor put back on a denoiser that
outlives the run, and no modifier marked as dropped, because a cancelled
modifier did not fail and disabling it next time would punish the user for
stopping.
Temp previews needed no change: core's PreviewImage already writes to the
temp directory under a unique prefix, so where the app SHOWS them is a UI
decision and is recorded as one rather than built here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A module's OWN default could be NaN or infinite. The override path refused
one and the declaration path did not -- and a default is the value most runs
use, because a setting nobody names never reaches the override check at all.
So the fault this project has already been bitten by, `sigma > nan` false at
every step and a modifier silently doing nothing, was reachable through the
one door that was never checked. Bounds too: an infinite min or max makes the
range check meaningless in one direction.
And two slots could share an id. The check for what exists was a set, which
cannot see a duplicate, so building the prompt by id overwrote one slot with
the other: a node silently missing from the queued graph, or one left reading
its own output. Refused now, before anything else is reported, because ids
are how a link names what feeds it and every later check is meaningless while
two slots answer to one name.
Both were found by attacking the whole tree rather than the diff, and both
are in code that already had tests -- the tests exercised the neighbouring
door in each case.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The run id froze after the first generation. Runs were marked by the
modifiers node, and ComfyUI CACHES a node whose inputs have not changed --
only the seed changes between most generations, so that node executed once
and every later run reported under the first one's id. Per-run log dedup
degraded to per-process with it, which is precisely the v4 fault log.py's
own comment describes, reintroduced through a caching interaction the fix
did not account for.
The sampler marks the run now. It is what generates, and it re-executes
whenever anything about the generation differs. Verified through the real
PromptExecutor with caching on: three seeds, three run ids. The
dropped-modifier record is also emptied there, because it travels on the
model and ComfyUI may hand the same model back -- a modifier that failed
last time deserves this time.
And a bad settings payload was only caught during execution, after the
checkpoint had already loaded. On a rented GPU that is the cost paid before
the guard fires, one layer up from the sampling case this project already
guards. ComfyUI calls validate_inputs inside validate_prompt, so the payload
is now understood before the queue accepts it: a typo'd module id is refused
with nothing loaded. Both doors check the same things, and a test pins that
they agree.
There was no test file for runs at all, which is why the first fault could
be introduced by a fix and survive a green suite.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
v4 could not remove a built-in node from the pipeline and the reason was
structural: the wiring lived in code as a CORE dict nobody could reach, so
"remove this" had nothing to delete and "use a different one" had nothing to
change. Code is not a thing a user can edit.
The pipeline is now data. Three operations over it, each refusing with a
reason rather than half-doing the job:
build turns slots into the prompt ComfyUI queues, checking every link's type
and refusing a graph that cannot work. replace points a slot at any node
producing what its consumers read -- the built-in ones have no special
standing, which is what makes them replaceable rather than merely present.
remove takes a slot out and rewires what it fed to what fed it, when the
node has exactly one input matching each CONSUMED output; v4 required one
for every output a class could produce, so a spare FLOAT nobody wired made a
node unremovable.
Demonstrated against real schemas: FunPack's VAE loader swapped for core's
VAELoader, the modifiers node removed with the sampler rewired to read the
model directly, and removing the empty latent refused because nothing
upstream could feed what it fed.
Then run for real. The default pipeline, with three loader slots replaced by
one checkpoint slot, built from data, accepted by ComfyUI's own
validate_prompt, and executed by its own engine to execution_success and a
picture. Every wiring and type validated on the first attempt; the only
things ComfyUI objected to in the untouched default were the three unset
file pickers, which is correct -- a fresh install cannot know which model
you own.
Added a checkpoint loader, because the split loaders are right for LTX and
H3 and cannot read an SD-family single file at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
v4's Chain Sampler became a hub: 8183 lines, roughly seventy optional
inputs, a _sample_chunk that grew six ALG arguments, and a hardcoded list of
knobs silently inert on one model family. Every technique that needed the
loop was added to it, so it collected all of them.
This one picks an algorithm and a schedule from ComfyUI's own registries --
so a sampler it gains appears here without an edit -- and runs them. Three
things stop it becoming a hub again. It never learns what a model-side
modifier is. Sampler-side ones reach it through ONE call site,
chain.process(ctx, latent), which cannot grow because the step travels in a
context object rather than in the signature. And it declares the hook points
it offers, so core hands it only what fits and it never names a modifier.
Some modifiers genuinely cannot ride on a model: they need the step index,
the schedule, or a second pass. Those attach here instead, and one asking
for a hook this sampler does not offer is absent and said so, not
half-running.
Verified against ComfyUI's own guider path with real weights: bit-identical
output. That took correcting the method, not the code -- see below.
MEASUREMENT, corrected: sampling on this Mac's MPS backend is
NONDETERMINISTIC. The same node with the same seed and the same inputs
differs run to run: 2.42 by max latent difference, 0.0008 by mean pixel gap.
Comparing by MAX said this sampler diverged from core by 14.2 and it was
about to be "fixed"; by MEAN against a measured noise floor it is 0.00000
while core against itself is 0.00183. Every comparison here now runs a
control twice first. Re-checked the Detail modifier the same way: 145x the
floor, so that claim stands.
Also matched what core's sampler actually does and this did not:
fix_empty_latent_channels using the ratios our own empty-latent node reports,
the preview and progress callback, the move back to the intermediate device,
and dropping the ratio keys so nothing rescales twice.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Colour is a preference, so it is a setting: hex in, hex out, one spelling,
refused if it is not one. A module handed "red" in one place and "#f00" in
another guesses wrong somewhere.
Image and mask are NOT settings and deliberately have no type. They are
data -- tensors the app uploads and the graph wires -- and ComfyUI already
has socket types for both. Putting a drawn mask in a JSON settings payload
would be inventing a second way to move something that already moves.
Runs now have an id. Previews, progress, results and log lines all have to
be attributable to one generation, and ComfyUI gives a node no way to ask
which run it is in. So FunPack marks its own, every log record carries it,
and the history can be filtered to a single run. Readable rather than a
uuid, because it is going to be read: "run 7" is findable in a wall of
output and a hex string is not.
The colour element was caught by the registry-driven rule test before it was
finished -- an element with no catalogue demo is neither shown nor attacked,
so coverage keeps up with the registry instead of lagging behind it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
digital-garbage