On the WSL runner every __gm.pixelAt costs about 7 ms, three times as
much as on stein2. Each call renders the whole stage into a fresh
render texture and stalls until the GPU hands one pixel back. pixelRow
and several checks read hundreds of pixels that way. Module 05 spent
36.8 of its 40 s on these reads, module 04 11.3 of 20 s.
- stagePixels in src/ui/render.ts renders once around all points and
reads them in one go. Only whole pixels at resolution 1 take that
path. Anything else stays point by point.
- The hook __gm.pixels(points) exposes it, and pixelRow uses it.
- The loops in e2e 04, 05, 14 and 44 read their grids in one call.
- The old comment in pixelRow said a wider frame comes back resampled.
That is not true. The colours match exactly, so the comment is gone.
Measured on the RTX (d3d12) and on stein2 (Intel, Vulkan): diff 0 at
10 680 points on each GPU. One screen row takes 4 to 18 ms instead of
2.4 to 9.9 s. The six affected modules need 32 s instead of 139 s
under WSL, faster than on stein2.
The rest of the gap is in Chromium's GPU process. perf puts three
quarters of its time in the drivers (NVIDIA D3D12, Mesa d3d12, dxg).
Four browser and Mesa settings were measured, none helped.
Tests:
- ./check.sh on melektaus-atx: 195 files, 2827 tests, CHECK EXIT 0 (+4)
- e2e 1-16,30,37,44,58 under WSL: ALL PASSED, 587 checks
- counter-check: a wrong index in stagePixels fails the unit test
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Until now every account could do the same, and managing accounts
meant running make-user.php on the server. Accounts now carry a role,
and admins get a tab in the editor to create, rename, re-role, reset
and delete accounts. Both issues share the same server files, so they
land in one commit.
#180 Roles:
- Every account is admin or user. A missing or unknown role counts as
user, on the server and in the client.
- auth/session and auth/login return {user, role}.
- Auth::requireAdmin() answers everyone else with 403.
- make-user.php takes the role as a third argument. Without it an
existing account keeps its role, so a password reset from the CLI
never demotes an admin.
- Admins see and save other users' projects like their own. Saving
keeps the original owner.
#179 User management:
- The account menu gets "User management" (admins only) and "Plugins"
(everyone). Both open the same closable tab, accounts on top and
plugins below. The fixed Plugins tab is gone.
- Per account: Rename, Role, Password (no old password needed) and
Delete with a confirmation. "New user..." creates one.
- Each account lists its projects with an Open button.
- Delete is refused while the account still owns projects.
- Renaming moves the projects along and ends every session of the
account. transfer takes the same lock as save.
- On their own row an admin can only set the password. The last admin
cannot be demoted or deleted. The server enforces both.
- The account rules live in AccountRules.php and accountRules.ts as
pure functions, so they are testable without a server.
Findings:
- At 1280 px the Delete button reached past the column edge and was
not clickable. The name column is flexible now, and module 49 checks
the layout.
- The server lets an admin delete other users' projects, but the
editor has no button for it yet. Left open in the issue.
Tests:
- ./check.sh: 148 files, 2214 tests, exit 0 (was 147 / 2204).
- New tests/php/accounts.test.php, 51 checks, plain PHP. CI runs it
as its own step in the gate job, the host has no PHP.
- New e2e module 49 with 18 checks. CI creates the e2e-admin account
in serve.sh. Modules 14, 38, 46 and the scaffold follow the new tab.
- e2e 1-16,37,38,45,46,49: all passed, 581 checks.
- Counter-checks: owner on save, admin view, default role and self
delete in PHP fail 7, 5, 5 and 1 checks. Unchecked requireAdmin,
a wider column and the caller as owner each fail exactly their
e2e check.
- Check sum 1314 -> 1333 (+1 in 38, +18 in 49), computed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The theme asks for Source Sans 3 at weight 500, but fonts.ts only
loaded 400, 600 and 700. The browser synthesised the 500 text. The
issue comment also asked for the symbols the font lacks to become SVG
icons. A scan of src/ against the font's cmap found 20 such characters
in 30 places.
- fonts.ts loads source-sans-3-latin-500-normal.woff2 as well.
- Menus draw their ticks and the "current" dot as icons. The dot is a
new MenuItem.current field and no longer part of the label text.
- Inspector action buttons (reset, apply, fit width/height) are
icons. The menu arrow behind them is the drawn Dropdown chevron.
- Kind marks come from a table in markIcons.ts. Picture kinds use an
icon, # and Aa stay text because the font has them.
- Scene marks in the asset panel, sound play/stop and the transition
arrows are icons too.
- Apply is an icon although the font has the arrow. It sits next to
the reset icon and would look different as a glyph.
- A new Icon widget draws on redraw like Button, so Node tests need no
DOMParser.
- Tests find icons through the SVG id. The new hook
__gm.locateIcons(id) and the harness helpers findIcon/clickIcon
replace locate() on the old characters.
Findings:
- Weight 700 is loaded but the theme never uses it. Left as is.
- The old scene-mark check in module 35 accepted any play glyph on
screen. It now checks the mark on the start scene's row.
- Module 44 needs the textures from module 37 and fails without it.
Tests:
- ./check.sh: 145 files, 2187 tests, exit 0 (before 144 / 2143).
- New tests/ui/fonts.test.ts: every theme weight is loaded, and every
character in src/ exists in every shipped face. The cmap reader
matches fontTools (231 characters).
- Icon tests cover the 19 new files. Each kind mark is either an
icon or text, never both.
- e2e 1-16,22,30,34,35,43,44: 586 green, 0 FAIL, then 44 stopped
for lack of module 37. e2e 1-16,37,44,47,48: 618 green.
e2e 1-14 on the final build: 454 green.
- Two new e2e checks: loaded weights in 14, a reset icon pixel in 04.
- Counter-probes all fail as expected: 500 removed, a play glyph put
back, the struct icon removed (one unit test each). In the browser,
500 removed and reset icon hidden fail exactly the two new checks.
- Editor-suite check sum 1312 (1310 measured by CI at 056a622, +2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The suite job for 0d20776 died in module 06. The new font and theme
moved the layout, and three checks relied on the old one.
- 06: the ranking step right-clicked the spot where step 6d had just
dropped a Log node. Whether that click hits the node depends on how
the layout snaps it. It now asks for a fresh empty spot.
- 14 and 27 expected a shut panel to keep 30 px. The header is
Theme.rowHeight + 4, now 26. A new hook __gm.panelHeader hands the
checks the real height, so the next theme change does not break
them again.
Tests:
- ./check.sh: 141 files, 2080 tests, exit 0
- Local full run before the fix: 06 passed with the new spot, only the
three header checks in 14 and 27 failed.
- GM_STEPS=1-14,27 after the fix: ALL PASSED, 457 checks
- No new checks, the total stays at 1286.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The suite spent 88% of its step time asleep: 564 s of steps, 496 s of
them in sleep(n) - fixed spans estimated by hand and then copied from
step to step. That is why it was slow, and why it flickered.
The signal is afterFrames(count, timeout) in ui/render.ts, exposed as
__gm.afterFrames(). It wakes the loop itself, because an editor that
draws nothing would never redeem a promise of a frame, and it gives up
after a timeout instead of hanging the run. settle() in the harness
waits two drawn frames (~32 ms) instead of 150-300 ms and falls back to
sleep(150) when the hook is missing (an old build): silently not
waiting would make every later check a coin toss.
Every action goes through it now - click, doubleClick, wheel, drag,
dragAsset, press, typeText, select, mutate, undo, redo - and with that
about 550 sleep() calls behind those actions became redundant, helpers
that are nothing but a mutate in a coat included. What a frame cannot
answer got its own condition instead: waitUntil(expr), play()/stop()
waiting on __gm.state.mode rather than on 300 ms, saved() for the
autosave (the two hand-written polling loops sampled at 300 ms for
something that arrives after a second), and showTab() as a real tool -
47 places called evaluate("__gm.showTab(...)") raw, and 24 of them
waited for nothing at all and read the old panel.
564 s -> 243 s of step time, the waiting share 88% -> 76%. Three kinds
of wait were kept, and each one bit when it was taken away: loading
bytes (textures, fonts), game time that a repaint cannot stand in for
(time scale, shader and particle clocks, the live values a running
graph carries), and the inertia the scroll step measures on purpose.
Two things the new speed uncovered, both real: 25-animation read the
frame menu unfiltered and had frozen an incomplete list (the CI failure
on c990dbc), and 33-render-on-change measured idleness while the
autosave of the step before was still painting - it waits for
saveStatus to leave dirty/saving now, so it measures that nothing
happens any more instead of that it waited long enough.
CI run 35 was red: the suite died after step 15 with `widget with text
"Container" not found`. Nothing in the editor broke — the tenth tab (#73) pushed
Container behind the `»` button at the test window's 1280 px, exactly the case
#84 was built for, and three steps still clicked the tab name directly.
openTab(t, tab) in tools/e2e/lib/scaffold.mjs asks tabRow() whether the tab is
hidden and goes through the menu when it is, so the fix lives in one place rather
than in every step that opens a tab. The LABELS record moves there as TAB_LABELS,
since both the helper and 14-overflow need it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The asset type `particles` is placeable and carries a graph at the same time —
`behavior` was the blueprint. Because a continuous emitter gives birth to particle n
at time n/rate, the whole field is a function of the clock rather than of the frames:
no ring-buffer bookkeeping, exact time jumps, reproducibility as a measurable property.
compile.ts follows compileShader's frame but emits JS closures into a Float64Array
instead of GLSL, and without eval; a node with two consumers is evaluated once.
Drawing is one Pixi ParticleContainer per source, so one draw call.
Collision, wind and braking in flight are out for good — that is the price of the
formula model, not an omission.
The tab count goes from nine to ten, which is exactly the case #84 was built for one
day earlier: at 1280 px the row no longer holds every tab, so guard 14b now asks for
hidden.length > 0 instead of === 0.
Nine tabs need 631 px; at 1280x800 the centre column offers 666 px, so the tenth
tab would have spilled under the inspector - the same overflow CI caught in run 5.
TabStrip replaces the bare HStack: it fits the longest prefix that stays inside
the budget, always appends the active tab, and offers the rest behind a fixed
28 px button. The button's width is fixed on purpose - deriving it from the
number of hidden tabs would make width depend on the count and the count on the
width.
The rule itself is a pure function (tabRow.ts), so it is checked without a
browser. setActive() calls relayout() itself because Widget.resize only lays out
on a changed size, and switching tabs does not change any size.
Adding a tab now needs no space arithmetic - only a line in TAB_ORDER,
TAB_LABELS and panes, all three Record<CentreTab, ...>, so forgetting one is a
type error.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The ninth tab (Shader) pushed the row past the centre column: List
ended at 962 while the column ended at 925, so the last buttons sat
under the inspector and clickText('Container') hit the List tab
instead (CI runs 5 and 6, module 16). Tab buttons now use padding 4
via a new Button padding option; e2e step 14b guards the row width
using the widget box rather than getBounds(), which sees only the
label on an untoggled ghost button.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- `.gitignore` said `index.html` unanchored, which also caught the Vite source page;
a fresh clone could neither `dev` nor `build`. Anchored to `/index.html`, page added.
- Removed GM_REMOTE (tools/remote/gm-browser.ps1 and the ssh/scp branch in browser.mjs):
the Windows test machine is gone and the only flaky check that lived on it is closed (#37).
- Removed `npm run preview`, undocumented and unused.
- Comments: wrong number in collider.test.ts (250 → 200), the "declarations used to…"
history in listItems.ts and friends, the 24 px/28 contradiction, section banners in
nodes/index.ts, node-catalog.test.ts and api/index.php, name paraphrases, and the stale
step numbers in tools/e2e/11-…23- from the old global count.
- README: asset types include `animation`, kind order Number/Text/Boolean, tagged colour
and texture items, 6/5 kinds, the `Texture` node, `./build.sh`, PROJECT_FORMAT_VERSION,
and the file tree (kinds.ts, listItems.ts, propertyPorts.ts, Breadcrumb.ts, beads.ts,
tools/assets/).
- tests/helpers.ts takes the duplicated `findObject` and `sceneWithCamera`.
Proof: ./check.sh 797 tests, full e2e ALL PASSED 633 checks against this bundle.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Split GraphCanvas (1076 -> 831 lines) into EdgeLayer and graphMenus;
the port and marker menus were the same list and are now one function.
Value specs of the inspector and the list tab share valueSpecs.ts, and
the remaining copies (compound commands, type glyphs, typed asset
lookup, name rule, play/editor world setup, marquee) are merged.
Four copies had already drifted apart; merging them fixes:
- "Disconnect N links" in the port menu is one undo step, not N
- "+ Add collider" is hidden on a placed container instance, as in
the hierarchy menu, instead of always failing
- a bool property coerced from the text "false" becomes false
- a port command on an unknown node type reports instead of crashing
Comment-only edits leave the bundle byte-identical. README corrected in
35 places (step limit, jump-run clock, suite duration, 409 body, ...).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A list property in the inspector was a single text field holding a comma
separated line -- unreadable past a handful of items, and it could not
carry anything but text. It becomes a button ("3 items") that opens a
sixth centre tab, where every item is a row built by the very widget the
inspector would build for that item kind.
The renderer was free: a PropertyDef of kind 'list' already carries
`items`, so item i becomes a synthetic def of that kind -- number, text,
boolean and colour rows appear without a new branch.
The way into the tab goes through the state (EditorState.listSlot,
openList(), event 'list'), like every other tab switch here; a tab may
not call showTab itself. ListSlot lives next to Selection rather than in
the tab folder, or state.ts would have to import upwards.
A row's handlers must not close over a snapshot of the list: they outlive
a commit, since the tab only rebuilds when the length changes, and the
second edited row would otherwise revive the value the first replaced.
showTab switched five `visible` lines by hand, where a sixth tab is
silently forgotten; it is a Record<CentreTab, Widget> now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A scene now stores only the container object; its content is resolved from
the template on every read, with deviations kept as sparse instanceEdits on
the container object. This drops the propagation bookkeeping entirely --
templateSync.ts is gone and the three structural gaps (add, remove, reparent)
cannot diverge any more.
Values stay editable on an instance, structure does not; Unpack writes the
content into the scene, keeping ids. Play mode bakes containers into the
throwaway copy before entities are built. No migration: formatVersion is 2
and the seeders write the new format.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Container tab reuses SceneViewport against any object holder, so a
container template is edited exactly like a scene. Right-click no longer
changes the selection, which previously made nesting impossible: opening
the menu on a container loaded it, and "Place in container" then targeted
itself.
Template edits (move, rename, override) now reach already-placed instances
per slot, through a shared templateSync used by both the apply path and the
plain object commands. Copies the user has touched keep their own value.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Remove comments that only restate the code, keeping the section banners
and the notes that carry a reason, a measured fact or an external
contract.
ButtonOptions.width claimed to be a fixed width but only set minWidth,
so the Graph and Texture pickers grew with the name they showed. It is
now what it promises, with a separate minWidth for the buttons that have
two forms. Narrow buttons shrink their padding, otherwise a 24px icon
button had 8px left for its glyph and truncate dropped it entirely.
__gm.labelsDropped() guards that class: the overflow guard only saw text
that was too wide, never a label truncated to nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- NumberField takes a decimals option (default 3); frame rects and the
slice dialog use 0, matching their integer model.
- Button truncates its label to the drawn width; Dropdown reserves the
chevron slot. NodeView reuses the shared truncate().
- Setting scaleMode leaves the style's shared resource id in place, so
the old GPU sampler stayed in use until a reload. Call style.update().
- Graph gets a header bar like the Texture tab: GraphEditor is now the
frame, the former editor is GraphCanvas. A selection the tab does not
concern no longer unloads the behavior.
- New e2e step 14 guards text overflow across all widgets.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
adminandClaude Opus 5.5