ca5b2db449ebf575ea65edf4ef477b2a7bfdfb11
451
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  adminandClaude Opus 5.5
|
ca5b2db449 |
Keep where each divider was dragged to across a reload
A panel dragged wider was back at its first size after every reload. The sizes are now kept in the browser, next to the folded groups. - editor/panelSizes.ts holds one entry in localStorage, game-maker.panel-sizes: a name per divider and its width, height or share. - A panel reads its size once when it is built, held to the bounds its divider keeps today. Every drag writes the new size. - Kept are the columns left and right of the centre, the console, the split of the left column, the tilemap panel, the preview of the Shader, Particles and Pixel art tabs, each under its own name, and the split of the user management. - The sizes belong to the browser, not to the project. They hold for every project opened there. - A browser that refuses the storage, or holds rubbish under the entry, starts with the sizes the editor comes with. Tests: - Gate on melektaus-atx: 219 files, 3400 tests, CHECK EXIT 0. - Seven new unit tests in tests/editor/panel-sizes.test.ts. All five counter-probes bite. - Smokes on melektaus-atx: 24, 19 and 21, all pass. - One new e2e check in module 48: the left column dragged 40 px wider is as wide after a reload, and dragged back after. - Full suite on melektaus-atx: 1692 checks, all pass. It ran before the fix to the animation players of the commit before last. - Browser counter-probe: without the write to the browser exactly that check fails. - Only the left column is measured across a reload. The other dividers go the same way and have no check of their own. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
8f9d373d7b |
Scroll the + menu to Vector art before picking it in e2e 47
The suite to
|
||
|
adminandClaude Opus 5.5
|
4407c02078 |
Keep who played an animation beside its player, not on it
|
||
|
adminandClaude Opus 5.5
|
d6fd000160 |
Let a behavior hear the end of an animation it played itself
A behavior that played an animation on the parent of its owner never heard it finish. On animation finished listened to the owner and to what stands below it, and the parent stands above. - Play animation notes on the player which behavior asked for it. - On animation finished hears, as before, every animation on the owner or below it, and now also every animation the behavior played itself, on whatever object. Its output object names that object. - Two behaviors that both played the same animation on the same object each hear its end once. - A behavior that is called by Run behavior or heard by an On node counts as the behavior it sits in. - The description of the node and the engine prompt say so. Tests: - Gate on melektaus-atx: 218 files, 3392 tests, CHECK EXIT 0. - Two new unit tests: the behavior hears the animation on the parent of its owner while a bystander does not, and two starters hear it once each. - Both counter-probes bite. - e2e 6,17-26 on melektaus-atx: 480 checks, all pass. - No new e2e check, and nothing was looked at in the browser. The rule for called and heard behaviors has no test of its own. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
7dd3ae1131 |
Tie two objects with a hinge, a spring, a rope or for good (#154)
Two bodies could not be connected: no door on a hinge, no lamp on
a rope, no wheel on an axle. This is the first step of the issue,
the four joints without a motor. The slider and the motors follow
in a second step under the same issue.
- Joint is a new asset type, placed as a child of the object it
holds, like a collider or a rigidbody. "Add joint…" in the
object menu makes one or places one the project has. An object
takes as many as it needs.
- Where the joint sits is the point it holds its object at. A
hinge and a fixed joint have that one point only. A spring and a
rope end at a connected anchor on the other object, counted in
that object's own units, or in the scene's without one.
- The rows follow the type: length, stiffness and damping for a
spring, length for a rope, an angle limit for a hinge. Length 0
takes the distance the two ends were placed at. The limit counts
from how the two were placed.
- The connected object is an object reference in the schema of the
type. Pasting, containers and removing treat it as they treat
every reference since #258.
- shownFor is a new schema option: a row is shown while an enum of
the same bag holds one of the named values.
- The object needs a rigidbody, its own or one further up. Without
one the joint holds nothing and the console says so once.
- A connected object without a rigidbody is a moving anchor: the
joint has a kinematic body of its own that follows it.
- A connected object destroyed while playing makes the joint let go
with a hint. Removed in the editor the reference empties, so the
joint holds on to the scene. The issue asked for "inactive" there.
- "Add joint" ties two objects while the game runs, from a joint
asset as the template. Destroy on the joint lets go, there is no
Break joint node. Destroy with keep children takes the joints
along with the object.
- The scene view draws a line from each joint to what it holds on
to and rings the held point. A selected spring or rope has a
round handle on its far end, and dragging it is one undo step.
- Joints hold in Play only. The editor run simulates nothing
(#148), and the inspector of a placed joint says so.
- Rapier counts a revolute joint's limits absolutely, not from the
start, so the placed angle is added. Measured with a probe.
- The + menu has 25 types now and no longer fits under its button
on a canvas 800 high. It scrolls there, and e2e 04 scrolls to
the last entry instead of expecting it in view.
Tests:
- Gate on melektaus-atx for this state: 218 files, 3390 tests
(3336 + 54), exit 0.
- e2e 1-16,24,39-46,51,52,57,58 on melektaus-atx: 831 checks, all
pass. Module 41 alone: 25 checks (21 + 4).
- 54 new unit tests: 40 in joints.test.ts (rope, spring, pendulum,
limit, fixed, the other end, the node, containers, paste), 6 for
the line the scene view draws, 4 catalog cases of the node. The
remaining 4 come from tests that make a case per asset type.
- Four new e2e checks in module 41: Add joint from the menu, the
rope to the hook with its line and handle, dragging the handle
with one undo, the bob hanging in Play next to one that falls.
- e2e 04, 40 and 46 count the new type in their lists.
- 21 of 22 unit counter-probes fail their tests. The other one is
equivalent: Rapier ignores removing a joint that left with its
body, the guard stays.
- Three browser counter-probes fail their checks (2, 1 and 2).
- One screenshot looked at: icon, line, handle and the rope rows.
- The e2e sum of 1691 is counted (1687 + 4, 1687 measured by CI
on
|
||
|
adminandClaude Opus 5.5
|
e066613583 |
Watch a held body for a time, not for a count of frames
The suite to |
||
|
adminandClaude Opus 5.5
|
e933fad446 |
Give a shader room to draw around the picture it wears (#69)
A shader ran only inside the rectangle of what wears it: the quad has the size of the sprite, and outside of it no fragment runs. An outline or a glow built with Sample texture was cut off at the border. The issue still spoke of Filter.padding, but since #77 the shader is no filter any more, the limit is the geometry. - The Output node has a number Padding next to Edge: so many pixels of the picture per side the wearer draws beyond it. 0 is the default and draws as before, 256 is the most. - The quad grows and its UVs grow with it, so the picture keeps its place and size. In the margin the UVs run on below 0 and above 1. - Transparent is a fourth Edge mode: outside of the picture there is nothing. It is cut in the fragment (texelAt), so it needs no copy of the picture with an address mode of its own. - Edge holds everywhere alike, also for Texture color, Texture alpha and unconnected inputs. - Padding above 0 with Edge on Clamp puts the hint "Padding with Clamp smears the border" on the Output. The shader still compiles. A chosen Clamp and the default cannot be told apart. - All four wearers draw the margin. A sprite counts in pixels of its frame, a text in its units, a particle source grows every particle about its centre. - A nine-slice sprite gets a ring around its nine fields, as wide as so many border pixels. The fields themselves stay. - Without a picture a pixel is one unit of the wearer's own size, or of the particle size. - Particles now take the Edge mode of their shader. It had no effect on them before. - An object wearing a padded shader is never culled, or its margin would vanish at the edge of the screen. - The preview in the shader tab shows the margin and leaves room for it. - Picking, the selection frame, the size and the collider stay with the rectangle without the margin. Tests: - Gate on melektaus-atx right before this commit: 216 files, 3336 tests (3319 + 17), exit 0. - e2e 28,30,31,37,41 on melektaus-atx: 178 checks, all pass. e2e 1-37,47,48,58 ran before the last change (no culling with padding): 1281 checks, all pass. - 17 new unit tests: the compiler (padding, Transparent, the hint) and the geometry (quad, nine-slice ring). - Seven new e2e checks: four at the sprite in module 28, one each for text, particles and nine-slice. - 17 unit counter-probes all fail their tests. Three browser counter-probes fail their checks (5, 3 and 1 of them). - Three screenshots looked at: the sprite with Clamp and margin in the scene, the preview in the shader tab, the text. - The e2e sum of 1687 is counted (1680 + 7), the full suite and the smokes did not run. - Not seen: the hint at the node in a picture, particles and nine-slice in a picture, the game view while playing, sprites without a picture, vector and pixel art pictures. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
46c620ae3b |
Give a behavior more ways in and out, and hear it as an event (#294)
A behavior run by another one had a single flow port at its Start and at its End, so it could not branch for its caller. And it could not be used as an event: a call needs a pulse at its flow input, and the events of a called graph never fire. An input like "Confirm" (Space or Enter) had to be built again in every graph. - A declared port of a Start or an End node can be of the kind Flow. It sits above the values and shares their names, exec stays reserved. Only these two nodes declare one, a struct does not. - Run behavior shows a flow port of the Start as a further flow input. A call arriving through it leaves the Start through the output of that name and not through exec. - Run behavior shows a flow port of the End as a further flow output. The call goes on through the one the End was reached by last, and then through exec. exec fires always, as before. - On behavior is a new event node without a flow input. The behavior it names runs at the node for as long as the behavior around it runs, with all of its own events and one set of variables per node. - Its outputs are those of the End. It fires once the inner pulse is through, by the output the End was reached by last, exec only for End.exec. A pulse that never reaches the End is not seen. - The value ports of the Start are its inputs, read anew for each inner pulse. The Start itself never fires there. - It nests to any depth. Under a Run behavior it stays silent. A behavior hearing itself, however far round, is not started and the node says "A behavior cannot listen to itself". - The palette offers every other behavior with an End once more as "On <name>". A node naming one without End says so. - An On node removed while the game runs ends its behavior with On disable and On destroy. The sounds of the entity end only with the behavior placed on it. - Renaming and deleting a declared port follows the edges at On behavior nodes as it does at Run behavior nodes. - The flow outputs of the Start of a behavior placed on an object never fire, like its exec. - The engine prompt of the benchmark names flow ports and On behavior. Tests: - Gate on melektaus-atx right before this commit: 215 files, 3319 tests (3278 + 41), exit 0. - e2e 19 on melektaus-atx: 39 checks (+4), all pass. e2e 1-6,17-24,26,36,51: 628 checks, all pass. - 19 unit tests in flow-ports.test.ts, 19 in on-behavior.test.ts on a real world, the rest in the node catalog. - Four new e2e checks in module 19: a flow port declared by mouse at the End, On Confirm in the graph, the call leaving through the further output while playing, Enter and Space on a real keyboard. - 19 unit counter-probes all fail their tests. One did not at first (a removed node kept its instance), a test was added. Two browser counter-probes fail their checks. - One screenshot looked at: On Confirm with exec and pressed, the second flow pin at the calling node. - The e2e sum of 1680 is counted (1676 + 4), the full suite and the smokes did not run. - Not seen in the browser: flow inputs at the Start, nesting, the ring, values through On behavior, the palette entry. No test holds that the sounds of the entity go on when a heard behavior ends. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
b5339b96c2 |
Tell a graph the frame a key went down or came up
Key is down says whether a key is held. To act once on a press, a graph had only the events On key down and On key up. A behavior that is called by another one and reads its input as data had to keep the state of the last frame in a variable of its own. - Key was pressed is true for the one frame in which the key goes down, the frame On key down fires in. Key was released is the same for the frame the key comes up. - Both sit next to Key is down in the palette and have the key as their only setting. - NodeContext gains wasKeyPressed and wasKeyReleased. The behavior context answers them from justPressed and justReleased of the input, through keyName, so a key saved as "A" meets the key "a". - No repeat. On key down may fire again while a key stays held, a held key is pressed once here. - When the window loses the keyboard, every held key counts as released in that frame, as On key up has it. - The engine prompt of the benchmark names the two nodes. Tests: - Gate on melektaus-atx right before this commit: 213 files, 3278 tests (3264 + 14), exit 0. - e2e 18,20,43 on melektaus-atx: 57 checks, all pass. - Ten unit tests in key-edges.test.ts on a real world, four cases in the node catalog. - Eight counter-probes all fail their tests. One did not at first (the case of the key saved on the node), a test was added. - No e2e check of its own, the sum stays 1676. The nodes were not looked at in the browser, neither in the palette nor with a real keyboard. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
b45bbcc976 |
Hold in tests that a graph edit reaches a running behavior (#216)
The issue said a behavior that runs in the editor keeps the graph it started with, until it is switched off and on. Measured before building anything, that is no longer so. An edit runs from the next frame on. Since when, was not found out. - No code changes. A command raises the project event, EditorRun refreshes, and syncFromModel of the behavior system empties the graph cache. - Decided in the issue: the instance carries on and is not started over. Its variables and timers stand, and On start does not fire again. A restart would write what On start writes or spawns into the saved scene with every edit. - Starting over stays what it was: Run in editor off and on, or Play. - Module 43 of the e2e no longer switches the behavior off and on before it measures the changed layers of a ray cast. Tests: - Gate on melektaus-atx for this change alone: 212 files, 3264 tests (3260 + 4), exit 0. The gate right before this commit ran on the tree of this and the next commit together: 213 files, 3278 tests, exit 0. - e2e 18,20,43 on melektaus-atx: 57 checks, all pass. The check in 43 is the old one, it now holds without the workaround. - Four unit tests in editor-run.test.ts: a changed node setting, a node wired in and one cut off, a flow hung on On start that does not fire until the switch goes off and on, and an undo. - One counter-probe: without the emptied cache the wiring test fails. The setting test does not, settings change in place. - No full run and no smokes. The sum of e2e checks stays 1676. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
c7dc0e4733 |
Leave hiding and deleting a link to its right-click menu
The inspector of a picked link carried the buttons Hide link and Delete link below its facts. Both actions are in the right-click menu of the line since #289, and Delete removes a picked link as well. The user asked for the buttons to go. - buildEdgeSection in editor/inspector/edgeSection.ts only describes now: the two ends, the type, and for a transition its name field. - The button under several picked links and Delete transition in a state machine go with them. They came from the same line of code, and the menu offers both. - The note under several links stays. It speaks of the Delete key. Tests: - Gate on melektaus-atx: 212 files, 3260 tests, exit 0. - e2e 4,6,26 on melektaus-atx: 275 checks, all pass. - No full run and no smokes. The sum of e2e checks stays at 1676, which is reckoned. - Module 06 now checks that the inspector of a picked link shows neither button. Its menu check finds Hide link and Delete link only in the menu, where it had to take the one drawn last before. - Module 26 checks that a picked transition shows no Delete transition. - No unit test names the buttons. No counter-probe was run. The old checks found both texts in the inspector, which is the state the new ones would fail on. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
f02adabcf9 |
Show the type an arithmetic shader node really puts out (#219)
The result of Add, Multiply, Mix and the other arithmetic nodes of a shader graph was drawn as any, whatever went in. The compiler already lifts a colour on any side to a colour result, so the pin said less than the code knew. A colour result could also be wired into a number input, and the shader then failed to compile. - liftedType in core/shader/nodes.ts is the one rule. The compiler reads it when it lifts two operands, and the editor reads it through liftedPortType for the ports it draws. - A SharedPortGroup may bring its own rule as agree. Without one it is agreedType, as before. The thirteen arithmetic nodes share their type by liftedPortType. - Only the any inputs decide. An input nothing feeds counts as the number typed into it. t, edge, exponent, min and max have no say. - The type travels through a chain of such nodes without code of its own, because #210 hands a shared type on. - In a ring the result stays any, unless a colour has settled it. - A number may feed a colour input. Without that a number result no longer fitted Output.color, although the compiler spreads it over three channels. The rule sits in portsCompatible, but only the shader graph has colour ports. - A colour result no longer fits a number input. An edge laid before the colour came in stays and is drawn in the danger colour, and the node at its end names the problem. - The nine shaders in the stored projects (108 edges) all stay valid under the new rule. Tests: - Gate on melektaus-atx: 212 files, 3260 tests (3240 + 20), exit 0. The gate right before this commit ran on the tree of this and the next commit together, with the same numbers. - e2e 28 on melektaus-atx: 81 checks, all pass. - e2e 1-31: 1059 pass, 1 fail in module 18 (the editor clock read 1.97 s after a stop). A rerun of 1-18 passed with 602 checks and 0.18 s. 1-31 was not run in full again. - No full run and no smokes. The sum of 1676 (1673 + 3) is reckoned. - Twenty unit tests in shader-port-types.test.ts, one of them runs all thirteen nodes through every mix of number and colour and compares what is drawn with the code made. - Two tests in shaders.test.ts now lay the edge first and bring the colour in after, the only order that still wires such a graph. - Three e2e checks in module 28 with the real mouse: a colour into Multiply, the red edge with its message, and the colour result refused at alpha and taken at color. - Fifteen unit counter-probes and two browser counter-probes all fail their tests. - Not checked in the browser: the one-input nodes, Mix and the ring, and a number wired straight into a colour input. Nothing was looked at after the build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
9ca220a285 |
Hand a shared port type on through a chain of nodes (#210)
Since #207 the output of First not empty or Select takes the type
its inputs agree on. A second node of that kind behind the first
stayed any, because the first one answered with the any it was
registered with. That was the guard against a circle of such nodes
asking each other forever. The issue asks for the type to travel
through the chain. Answered there: measure with 50 nodes and build
no cache, and the painted input pins follow along.
- shapedDef in core/graph/nested.ts takes a Trail in place of the
flag inherit. Its set open holds the nodes still being asked. A
node reached a second time reports the any it was registered
with, which ends a circle.
- The map done of the Trail keeps each answer of one question. A
node feeding both inputs of the next is asked once, so the work
does not double with every link of a chain.
- ownShape is the part of shapedDef that builds the ports before a
shared group has its say. It is split off, nothing in it changed.
- Inputs stay any and only outputs inherit, as in #207. An edge
that turns invalid further down is left in place and drawn in the
danger colour.
- paintedInputs uses one Trail for all pins of a node, so the pins
show the type handed on through the chain.
- Nothing is cached between two changes. A chain of 50 resolved in
about 10 ms where it took 2 ms before, measured earlier in this
round and not again for this commit.
Tests:
- Gate on melektaus-atx: 211 files, 3240 tests (3227 + 13), exit 0
- e2e 1-6 on melektaus-atx: 298 checks, all pass
- e2e 4,6,18,19,20,23,24,26,28,31,36,43,53: 558 checks, all pass
- No full run and no smokes. The sum of 1673 (1671 + 2) is reckoned,
1671 is what CI counted for
|
||
|
adminandClaude Opus 5.5
|
c88f57cf0a |
Let a project choose how its game link is zoomed (#279)
The page of a released game always drew the camera at Auto, the
largest whole step that fits. A camera of 300x180 in a window of
1280x800 stands at step 4 and leaves a border left and right. The
issue asks for Fit next to it. Answered there: a setting of its own
rather than the scale of the Game tab, and no switch for the player.
- The project has a field gameLinkZoom, 'auto' or 'fit'. A document
without it reads as 'auto', so nothing changes for a project that
is there already. No format bump.
- Project settings has a row "Link zoom" with a menu of Auto and
Fit. It is an undo step of its own, SetGameLinkZoomCommand.
- GameScreen hands the field to cameraView where it passed a fixed
'auto'.
- gameScale, the scale of the Game tab, is not touched and does not
reach the link.
- The released copy is frozen, so the change reaches the link only
with "Update release".
Decided here:
- The row is called "Link zoom", not "Game link zoom" as proposed.
The longer label is wider than the label column of the dialog and
pushed its button 12 px out of line with the others.
- The undo step keeps the name "Game link zoom", which is what the
tooltip of the undo button shows.
- A note under the row says "The released game at the largest whole
step that fits, or filling the window."
Tests:
- Gate on melektaus-atx: 211 files, 3227 tests (3219 + 8), exit 0
- e2e 60 on melektaus-atx: 19 checks, all pass
- e2e 35,46,59,60: 74 checks, all pass
- No full run and no smokes. The sum of 1671 (1668 + 3) is reckoned,
1668 is what CI counted for
|
||
|
adminandClaude Opus 5.5
|
9e869df063 |
Free a redrawn picture only once no shader names it (#293)
Editing a vector drawing made Pixi warn "destroyed while still bound
to a shader", twice with every change, once it had begun. It began
where an object wears the drawing through a shader asset. Each
change draws the picture anew and destroyed the one before it, but
the mesh of that object keeps its picture bound until it syncs. The
issue also asks for the button "Add shape" to go: a shape without
points cannot be given any, the Pen draws a new one.
- GeneratedPictures takes two more callbacks. inUse says that a
picture cannot be freed yet, soon runs a retry once whoever asked
is done.
- A replaced picture that is still in use waits in a list of its
own. It is freed right after the redraw that asked, as a rule. On
a later ask where its wearer was not part of that redraw.
- TextureLibrary tells a bound picture by the listeners for change
on its source. Under WebGL the bind groups are the only ones that
listen there. The retry is a microtask.
- The rule is asked for every drawn picture alike, whoever wears it.
- The inspector of a drawing and of a shape has no "Add shape" any
more. "Delete shapes" for several selected shapes stays.
Decided here:
- Under the list of shapes a second note says "The Pen draws a new
shape." while only the asset is selected.
Found on the way:
- Reproduced on a throwaway copy of the project the report came
from: 6 warnings for 3 changes with the shader override on the
object, none without it. After the change none, and no replaced
picture is left waiting.
- The fixture project of the tests never showed the warning. No
object there wore a drawing through a shader.
Tests:
- Gate on melektaus-atx: 211 files, 3219 tests (3209 + 10), exit 0
- e2e 47,48 on melektaus-atx: 113 checks, all pass
- e2e 28,30,31,39,40,47,48: 350 checks, all pass, run before the
wording of the new note changed
- The first gate was red: an existing test read the wording of the
note that had changed. Only three test files had been run after
the change.
- No full run and no smokes. The sum of 1668 (1667 + 1) is reckoned,
1667 is what CI counted for
|
||
|
adminandClaude Opus 5.5
|
ec2fd9084f |
Say what a button without words does when the pointer rests (#209)
The top bar, the magnet, the eye and many marks of the inspector are
pictures without a caption. Whoever did not know one could only try
it. The toolkit had no hint on hover, which also forced the point
menu of #195 to be words. The answers in the issue settle the open
points: a cut tab name shows in full, shortcuts are part of the
text, the point menu stays as it is.
- Button takes a tooltip, a text or a function asked each time. It
shows on a small sign once the pointer has rested for half a
second.
- There is one sign for the page, a TooltipLayer on the overlay
above the menus. A control reports each pointer move and the
pointer leaving, the layer does the waiting.
- A press, a key, the wheel or a blur of the window takes the sign
away. initUI listens for them on the window.
- placeTooltip in ui/tooltip.ts is the rule for the place as a pure
function: under the control and flush left, flush right at the
right border, above the control at the bottom.
- toolbarTip says what the five buttons of the top bar do right now.
Undo and redo name their step from CommandStack.undoLabel and
redoLabel, as in "Undo Paint tiles (Ctrl+Z)".
- A SortableList reports its own marks with a key per mark: the eye,
the duplicate mark, the cross and the plus.
- The magnet, both fullscreen buttons, the eye of the scene tab, the
overflow of the tab strip, the plus of the assets panel, the marks
of the inspector rows and the nine anchors of the resize dialog
carry a text.
Decided here:
- Every button whose text is cut short shows the whole text, not
only a tab.
- A row action of the inspector names the one thing a click does. A
button that opens a menu says only Reset, Apply or Size from the
picture.
- A button just clicked stays quiet until the pointer has left it.
- A greyed out button shows no sign.
- The switches of the eye popup, the cross of a document tab and the
tools of pixel art have none. The tools carry words, other than
the issue assumes.
- The marks of a list say Show or Hide, Duplicate, Remove and Add,
whatever the list holds.
Found on the way:
- Signing out empties the overlay and would have destroyed a sign on
screen with it. The editor takes the tooltip away first. This has
no test of its own.
- The text hooks locate and has skip the sign, or it would answer
for the button under it.
Tests:
- Gate on melektaus-atx: 210 files, 3209 tests (3176 + 33), exit 0
- e2e 46 on melektaus-atx: 20 checks, all pass
- e2e 1-14 on melektaus-atx: 508 checks, all pass
- e2e 1-9,14,39,40,44-48: 685 pass, 2 fail, both in module 14. It
measures tabs that modules 10 to 13 open, and the chain 1-14 is
green. The cut chain was not run on the state before.
- No full run and no smokes. The sum of 1667 (1664 + 3) is reckoned,
1664 is what CI counted for
|
||
|
adminandClaude Opus 5.5
|
345289982b |
Scale, turn and mirror the shapes of a vector drawing (#291)
A shape could only be moved and reshaped point by point. Whoever
wanted it larger or turned pulled every point alone, and for several
shapes or a group there was no way at all. The answers in the issue
settle the rest: the change is baked into the points, nothing of it
is remembered, and the line width stays.
- With Select the selection stands in a frame of eight squares and a
knob, drawn like the one of the scene view. It holds one shape,
several shapes or a group, the same selection that is dragged as
one.
- A corner square scales from the opposite corner, a square on an
edge in its own direction. Shift keeps the proportions, Alt scales
about the middle.
- The knob turns round the middle of the frame, with Shift in steps
of 15 degrees. After the pull the frame stands upright again.
- The inspector has a band TRANSFORM with X, Y, Width, Height and
Rotate by, and the buttons Flip horizontal and Flip vertical.
- Points, handles, gradients and texture tiles are taken through the
same matrix. The rule is a pure function in vectorTransform.ts,
transformShapesCommand makes one undo step of any number of
shapes.
- A press goes to a point or a handle first, then to the frame, then
to a line. shapeFrame.ts holds where the handles are and what a
pull of each means.
Decided here:
- The frame stands 15 pixels off the shapes. The points of a shape
lie on its bounds, so a square on the bounds would cover them.
- The knob only turns. In the scene view it scales as well.
- A side cannot be pulled through the opposite one, the frame keeps
at least 1 pixel.
- Coordinates are rounded to a thousandth of a pixel, so a quarter
turn is exact.
- A texture tile stays upright under a turn and grows to the bounds
of the turned tile. The bands of a gradient stay square to its
axis under an uneven scale.
- The switch Scale stroke of the proposal is not built.
Found on the way:
- The inspector of a shape is now taller than the window. Two checks
of module 47 no longer found their row or button on the screen and
use the new hook inspectorButton, which scrolls there first.
- Not solved: with a shape at the top edge of the view the knob lies
outside the tab. The wiki names it under the limits.
Taken over from #217: the e2e check for two selected shapes without
a group, and unit tests for the snap of the common offset and for
the gradients of the shapes dragged along.
Tests:
- Gate on melektaus-atx: 209 files, 3176 tests (3121 + 55), exit 0
- e2e 47,48 on melektaus-atx: 112 checks, all pass
- No full run and no smokes. The sum of 1664 (1660 + 4) is reckoned,
1660 is what CI counted for
|
||
|
adminandClaude Opus 5.5
|
6d79cf59a3 |
Move a large graph instead of drawing it again (#292)
In a graph of many nodes the frame rate dropped while panning, and more so while one node was dragged. Every pointer move rebuilt every box, every text and every curve, and a drag also announced a changed project, so the panels rebuilt along with it. - Panning and zooming move the picture. GraphCanvas.applyView sets position and scale and draws anew only the grid and the rubber band. It used to call redraw(). - A node draws its box and makes its texts only when what they show has changed. NodeView keeps a signature of what redraw reads and one of what rebuildTexts reads, and returns early while they hold. - A drag does not announce a changed project. GraphDrags writes the model on every pointer move and asks the canvas once a frame to follow (GraphCanvasHost.moved). state.preview() waits for the command of the release. - Nodes outside the picture are not drawn. inView in graphView.ts is the rule, cull() applies it after every redraw and every change of the view. - The graph is a render group, and box and pins go into the batch. The mask sits on a container above the group, on the group itself it would be panned and zoomed along. - The rubber band and the drop mark of a dragged behavior are drawn on their own (drawBand), without a redraw of the graph. Measured in the CI image with 388 nodes and 540 links, fitted to the canvas, 240 pointer moves 16 ms apart: - Pan, script per move: 25.2 ms before, 2.0 ms after - Pan: 4.4 frames per second before, 18.7 after - Drag a node: 2.7 frames per second before, 22.4 after - The still picture: 9 frames per second before, 21 after Not solved: the still picture of such a graph stays at 21 frames per second in the test browser. What is left is the drawing itself, and the frame rates are those of that browser, not of a desktop GPU. Tests: - Gate on melektaus-atx: 207 files, 3121 tests (3114 + 7), exit 0 - e2e 4,6,18,19,20,23,24,26,28,31,36,43,51,53 on melektaus-atx: 542 pass, 1 fail. The fail is module 51 after module 43, an order the suite does not have, and it fails the same way without this change. Module 51 alone: 12 checks, all pass. - No full run and no smokes. The sum of 1660 (1656 + 4) is reckoned, and the CI workflow is switched off. - Seven unit tests: six for inView in tests/editor/graph-view.test.ts and one in frame-preview.test.ts - Four e2e checks in block 6z-xii of module 06: pan, drag, zoom in and fit, a changed value - Seven browser counter-probes fail 2 to 3 checks each Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
32a0af6e2c |
Put the text and boxes of the interface on device pixels
On a screen scaled to 125 % or 150 % the editor looked soft beside a page the browser draws itself. The layout counts in whole CSS pixels, and most of those fall between two device pixels there. The GPU averages both: text loses its edges, and a 1 px line becomes a pixel and a half of mixed colour. Now text and boxes are rounded to device pixels, and a line is one device pixel wide. - ui/pixelGrid.ts holds the rule. crispText() makes a text with Pixi's roundPixels, crisp() does the same for a Graphics of upright boxes, and hairline() is the width of one device pixel. - Every text of the interface is made with crispText() instead of new Text(). - crisp() only rounds where the scale is not a whole number. At 100 % and 200 % nothing changes. - drawControlBox, drawHeaderBar and drawSectionBand draw their lines a hairline wide and flush with the edge. The widgets that draw lines of their own do the same. - Two bands share exactly the line they draw, and a column still ends on a whole pixel. - The stage is nudged by 1/32 px where the scale is not whole. At 150 % every odd pixel is an exact half, and without the nudge one band came out a row shorter than its neighbours. - Curves, icons and everything of the game stay unrounded. Rounding their points would bend them. - The harness takes deviceScaleFactor as an option, and the hooks deviceColumn, edgeContrast and pixelGrid read the frame as the screen gets it. Not in here: the subpixel smoothing a browser gives its own text. Pixi draws a text into a transparent picture first, and there only greyscale smoothing is possible. Measured at 150 %, edge contrast before and after: top bar 31.5 to 47.8, the lists on the left 68.6 to 86.4, inspector 27.6 to 38.4. At 100 % it is 46.5, 90.1 and 37.9, and the picture there is the same as before pixel for pixel, apart from the fps display. Tests: - Gate on melektaus-atx: 206 files, 3114 tests (3104 + 10), exit 0 - Full run on melektaus-atx: 1655 of 1656 checks passed, smokes 24/19/21. The one that failed was a condition of the new step 62 itself: it scrolled an inspector that had nothing to scroll. The step now makes the window lower for that reading. - e2e 61,62 after that fix: 14 checks. The full run was not repeated, and CI is switched off for now. - Ten unit tests, nine in tests/ui/pixel-grid.test.ts - Eight e2e checks in the new step 62, which runs at 150 % - Eleven unit counter-probes fail tests - Four browser counter-probes fail their checks. They ran before the fix to step 62 and were not repeated for it. - Screenshots looked at for 125 % and 150 % Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
275e50af01 |
Let a tile report a touch instead of blocking it (#160)
A tile could be solid, one-way and scaled, but not report without blocking. Water, lava, spikes or a goal zone needed a trigger object of their own laid over the map. Now a tile has a switch Trigger, and the collision events name the tile and the cell they met. - A tile has the switch "Trigger (reports, never blocks)" under Solid. Its cells hold nothing up: no Move, no falling body, no ray. They raise On collision enter, stay and exit with the tilemap as other. - The three events have the outputs tile, cell x and cell y. Without a tile, tile is empty and the cell is -1, -1. - The three events have the input per cell. Off, the default, gives one event per area: cells of the same tile that touch side by side. On gives one event per cell. - The scene view outlines trigger cells in yellow like a trigger collider, merged to rectangles but never with solid cells. - Is touching with the map counts trigger tiles, and Set tile works while the game runs. - Both systems find the touches with the same overlap check in runtime/systems/tileTriggers.ts, not through the physics engine. Decided here: - Trigger wins over Solid. A tile with both does not block, as with a collider. - Water beside lava is two areas, and so are two ponds that touch only at a corner. - An area names the cell nearest to the centre of the collider, and on exit the cell it named last. - The solid cells of a map still do not name their tile. - Normal, impulse and point are 0, as with a trigger collider. - Collision scale and outlines shape a trigger cell like a solid one. One-way, friction and bounciness have no effect. Tests: - Gate on stein2: 205 files, 3104 tests (3067 + 37), exit 0 - e2e 39,40: 106 checks. e2e 1-8,39-43: 490 checks. Both on melektaus-atx. No full run, the sum of 1648 (1645 + 3) is reckoned and CI measures it. - 37 unit tests in tests/core/tile-triggers.test.ts, nearly all against both systems - Three e2e checks with a real mouse in block 40l-iv - 24 unit counter-probes fail tests. Two did not at first, tests were added for them. - Two browser counter-probes fail their checks - Two screenshots looked at (yellow outline, the node with per cell) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
2ff3a78607 |
Put the shapes of a vector drawing on layers and in groups (#190)
A vector drawing was one flat list of shapes. With many shapes the list got hard to read, and several shapes could not be kept or moved together. Now the shapes stand on layers, as in pixel art, and can be put into groups that nest. - A drawing has layers, each with a name and an eye. A hidden layer is neither in the tab nor in the picture a sprite wears, and its shapes cannot be pressed. One layer is the one worked on: the pen and Add shape draw onto it, and SHAPES lists what is on it. - The inspector has a list LAYERS over the shapes. Its plus adds a layer, the cross on a row removes it with all on it, the last one stays. A double-click renames, a drag sorts. - Group (Ctrl+G) puts the selected shapes into a group, Ungroup (Ctrl+Shift+G) takes it away. A group stands over its rows in the list, the rows one step in, and a row is dragged into a group or out of it. - In the tab a press on a shape takes its outermost group, and a drag moves the group as one undo step. - The model has layers with a tree of shapes and groups instead of a flat shapes array. The rules are pure functions in core/model/vectorTree.ts. A drawing saved before opens with all its shapes on one layer. - SortableList takes rows with a depth and drops into a tree. The flat lists of tilemap and pixel art run as before. Decided here: - A group has no transformation of its own and cannot be hidden or locked. Rotating and scaling is the follow-up #291. - A group is never empty: the one that loses its last shape goes with it. - Several selected shapes are dragged together, grouped or not, and a click on a shape of the selection keeps the selection. This is what #217 asked for. - Shift+click selects like Ctrl+click. With Select, Shift pans only beside the shapes, as in the scene view. The middle button pans anywhere, and with the pen and Paint Shift pans as before. - A row of the list picks one shape of a group alone. - Shapes change their layer through a row Layer in the inspector, which is there from two layers on. Rows are not dragged between layers. - A group cannot span two layers, the button is off then. - The pen does not draw onto a hidden layer, the hint says so. - Groups are always shown unfolded. - The four order buttons work inside the own group or layer. Tests: - Gate: 204 files, 3067 tests (3008 + 59), exit 0 - e2e 47,48: 108 checks. e2e 39,40,47,48 before the Shift change: 209 checks. No full run, the sum of 1645 (1639 + 6) is reckoned and CI measures it. - 59 unit tests: 36 in tests/core/vector-tree.test.ts, 23 in tests/editor/vector-layers.test.ts - Six e2e checks with a real mouse in block 47j - 47 unit counter-probes for layers and groups and three for the Shift click all fail tests, run on stein2 - Three browser counter-probes fail their checks - Two screenshots looked at (group in the list, hidden layer) - Only unit tested: the row Layer, sorting and removing layers, copying a group, the Shift pan beside the shapes Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
393caadc9d |
Head the groups of the inspector with bands that fold (#290)
The groups of the inspector had a small grey label as their title, in the same grid as a property row, so with many rows one group hardly stood apart from the next. Now every group has a band as its head, from edge to edge of the panel, and a click on it folds the group. - A band runs across the padding of its column in the header colour, with the title and a mark that says whether the group is open. The titles stay as they are, as answered in the issue. - A click on a band folds its group: the rows go, the band stays and the next band moves up against it. A control in the band (Open, + Variable, + Property) does its own job and folds nothing. - A placed object has a group SWITCHES under its name, asset and parent: Enabled, Keep on scene load and the switches of a behavior. An asset has SWITCHES and PROPERTIES, and a group without rows is left out. - The same bands head the tilemap panel, the Variables tab, the side of the Pixel art tab and the shape list of vector art. A list with a title (Layers, Frames, Shapes) folds with a click on its header, clear of its plus. - ui/sections.ts holds the rules as pure functions, SectionStack is the column and SectionHeader the band. Every title of the inspector goes through addTitle or sectionHeader in inspector/host.ts. Decided here: - A fold is kept for good, in localStorage of this browser. The issue asked whether for the session or for good and the answer left it open. - A fold is remembered by title, per panel. PROPERTIES folded once is folded on every object, asset and node. A number in a title does not count. - A behavior asset has all its rows under SWITCHES. A Variables asset and a state machine get a group SCOPE. - A greyed out control in a band takes the click too. Without that the click fell through to the band and folded the group. - The Curve tab has no column of this kind and is unchanged. Found on the way: - The inspector of a sprite is taller than the 800 pixel window of the checks now. The hooks that hand out a row scroll it into view first, and module 28 scrolls before it looks for a row by its text. Tests: - Gate: 202 files, 3008 tests (2982 + 26), exit 0 - e2e: no complete run for this state. Modules 28 to 61 passed alone (749 checks). A full run was cut off in the accounts module at 1492 passes without a failure, the modules after it and the smokes are missing. The sum of 1639 (1635 + 4) is reckoned, CI measures it. - 26 unit tests in tests/editor/sections.test.ts - Four e2e checks with a real mouse, three in module 02 and one in module 48 - 25 unit counter-probes all fail tests, run on stein2 - Screenshots of the bands looked at before the last build, none after it Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
c5ba21533d |
Pick a link on its line and delete it from there (#289)
A link could only be removed through the menu of its port, and the line itself took no input. A transition of a state machine could only be deleted in the list of the inspector. Now the line is the handle, in a behavior's graph and between states alike. - A click on a line selects that link and draws a halo behind it. A press counts within 6 screen pixels of the line, and where two cross the one drawn last wins. - Ctrl-click or Shift-click adds a link or takes it away. The rubber band takes every line it crosses, into the same selection as the nodes and boxes it caught. - Delete or Backspace removes what is picked as one undo step. A right-click on a line offers Hide link and Delete link. - The inspector shows a picked link: the node and port at each end and the type it carries, with Hide link and Delete link below. For several it says how many. - The arrow of a transition is picked the same way, on its name too. Delete or a right-click with Delete transition removes it. - The menu of a port is unchanged, Disconnect included, as answered in the issue. - editor/graph/edgeHit.ts holds the hit test as pure functions, and editor/inspector/edgeSection.ts builds the section. The selection has a new kind edge, and node and annotation carry edgeIds. Decided here: - A hidden link has no line and is never picked this way, also while it is shown on loan. Its mark at the port stays its handle. - The menu of a line speaks about the whole selection, like the menu of a box. With nodes or boxes held alongside it reads Delete N items. - Hiding picked links lets go of them, since no line is left. - A link wins over the box it crosses, and a mark wins over a link. - A transition has no Hide entry. - With nodes or boxes in the selection the inspector keeps showing the node or box, not the link. Tests: - Gate: 201 files, 2982 tests (2967 + 15), exit 0 - e2e, full run: 1635 checks passed (1629 + 6), exit 0 - Smokes: 24 / 19 / 21, exit 0 - The full run was done for exactly this state. - 15 unit tests in tests/editor/edge-selection.test.ts - Six e2e checks with a real mouse, block 6z-xi in module 06 and 26e7 in module 26 - 12 unit counter-probes all fail tests, run on stein2 - Three browser counter-probes fail, the build went through each time: no hit on the line, no halo, a band that skips lines - Screenshots of a picked link and a picked transition looked at before the last build, none after it Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
0f85476b93 |
Add points by double-click and delete the picked one (#288)
Four editors edit points, and all of them put a point down on a single press on the line and took one away on a double-click. A stray click added a point, and there was no way to see which point a key would hit. All four now work the same way: the Curve tab, the Vector art tab, a polygon collider in the scene view and a frame's collision outline in the Texture tab. - A double-click on a line, edge or segment puts a point there, in the Curve tab on free space too. The point is picked and stays in place. It can be dragged with the next press, as answered in the issue. - A click on a point picks it, and it is drawn filled. A double-click on a point takes nothing away any more. - Delete or Backspace takes the picked point away. Without a picked point the key removes the shape, the object or the outline as before. - A right-click on a point picks it and opens a menu with Delete point. In the Curve and Vector art tab the four point types stand below it. Alt-click stays. - A polygon keeps three corners and a curve keeps its ends. The entry is greyed out then. - The pick ends with a click beside the points, with another object, shape or frame, and after the point is gone. - editor/pointGestures.ts holds the rule as a pure function, and editor/pointMenu.ts builds the menu for all four. Decided here: - A single click on a line no longer adds anything. In Vector art it drags the shape, on a polygon it drops the pick, and in the Texture tab it makes that outline the active one and drops the pick. - In the Texture tab a click inside the frame still appends a corner, and that corner is picked. Delete takes the last corner first. To remove the whole outline, click one of its edges before. - The point the Pen put down last is picked, so Delete takes it and not the shape. - Delete hits the picked point also when the tab has lost the keyboard focus, after a click in the shape list or after the menu. - A vector shape that loses its last point is removed whole. Before, an empty shape stayed in the list. Findings: - Fixed: the new hint line in the Texture tab was cut off at the edge. The shorter text was not looked at in a picture again. - Not fixed: in the scene view the collider's name label covers a corner that sits near the centre. That was so before. Tests: - Gate: 200 files, 2967 tests (2944 + 23), exit 0 - e2e, full run: 1629 checks passed (1625 + 4), exit 0 - Smokes: 24 / 19 / 21, exit 0 - After the full run an unused parameter and a comment were tidied. For that final state the gate and e2e 42,47,53 (105 checks) ran again, the full run did not. - 23 unit tests in tests/editor/point-gestures.test.ts, for the rule and for each of the four editors - Four e2e checks with a real mouse, two in module 42 and one each in 47 and 53. The existing checks for adding and removing a point use the new gestures. - 20 unit counter-probes all fail tests, run on stein2 - Two browser counter-probes fail, the build went through both times. Without the Delete wiring the key removes the collider. Without the right-click hook the menu is missing, and that run stops in module 42, so 47 and 53 were not reached with it. - Three screenshots looked at, the picked point is filled. None of the Vector art tab. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
1aa824ecb4 |
Keep objects through a Load scene with a switch (#165)
A Load scene destroyed every object of the old scene. A player, a HUD or a music object had to be built again in each scene, with its state pushed through shared variables. A top-level object with the new switch Keep on scene load now travels into the next scene as the same entity, with everything below it. - The switch is a common property right under Enabled, off by default, on the asset and on a placed object. The inspector shows the row on top-level objects of a scene only, as answered in the issue. - runtime/keptObjects.ts holds the rules as pure functions. keptEntities picks what travels, objectsToBring copies the stored objects as the run parents them, and bringObjects puts them into the next scene. - World.instantiate takes a handed entity as it is instead of making one. The next syncToModel then writes the run's values into the scene. - BehaviorSystem.release takes the running instances out before the old world is destroyed, and the next system adopts them. They hear no On destroy and no second On start, and their sounds play on. - PhysicsSystem.resume gives a dynamic body the velocity and spin it had. - Duplicates: the object brought along wins. The next scene does not build its own object with the same id, which is the way back into the scene the object came from. - Decided here: the next scene does not build a top-level object of the same asset that is kept itself either. The issue only names the id. With the id alone a level that places its own player would hold two after a swap. - Decided here: kept objects land behind the scene's own, the switch is no animation row, and collisions are found anew. A finding, fixed along with it: Load scene did not build a scene as it is stored, but as the run had left it. A moved object stayed moved and a destroyed one stayed gone. The way back into a scene needs a clean build. - GameRun copies a scene's objects and groups before it first builds it and puts them back before it builds it again. - GameRun.end puts every such scene back. The player calls it on stop, so spawned objects no longer stay for "Click to play again". - The consequence: an edit made in a scene during Play is gone the next time that scene is loaded. Tests: - Gate: 199 files, 2944 tests (2929 + 15), exit 0 - e2e, full run: 1625 checks passed (1619 + 6), exit 0 - Smokes: 24 / 19 / 21, exit 0 - 15 unit tests in tests/core/keep-on-scene-load.test.ts - Six e2e checks in module 35: the switch is set with a real mouse, the child has no row, the object is in the new scene, its counter still reads 1, its sound plays on, and the stored scenes are unchanged after Stop - Eleven tests with fixed row lists only count the new schema row - 16 unit counter-probes all fail tests, run on stein2. One did not fail at first (same id not left out), it has its own test now. - The browser counter-probe fails 4 of the 6 new checks - Not looked at in the browser: the switch as a port of the Properties node, and a released game with a kept object Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
3e8c88c66d |
Add a node that names the collision layer of an object (#211)
The issue asked for three nodes: Set layer, Set collides with and Layer of. A probe showed that the finding behind the first two is outdated. The Properties node of type Collider already writes both settings while the game runs, and both collision systems see the change at once. As agreed, only Layer of is built, and writing stays with Properties. - Layer of is a data node with the input object and the outputs name and number. The name is what the project calls the layer. - NodeContext.collisionLayerOf answers the number, and the new field collisionLayers on GraphLookup brings the names to the node. - layerHolderOf in solids.ts says whose layer counts. An object stands for its first collider child, a collider or tilemap for itself. - Decided here: a collider that is switched off counts too, because its layer is set either way. - Decided here: without a collider the name is empty and the number is -1. - The node is a read, not a run, so it answers at the moment it is asked. Right after a Set in the same flow it names the new layer. - The output number was not asked for. It fits the layer port of Properties as it is. Changing the layer during a run had no test at all so far. Five tests now hold it in both systems. Tests: - Gate: 198 files, 2929 tests (2914 + 15), exit 0 - e2e 1-6: 285 checks passed (06: +2 for block 6z-x) - Five unit tests for the change during a run: a mover is blocked, passes and is blocked again, a ray with a layer filter follows the change, and a dynamic body falls through the ground in Play - Six unit tests for Layer of, three against each system, and four cases in the node catalog - 6z-x finds the node in the palette, checks its ports and logs "Ghost" and "1" for the owner's collider in a Play run - Two counter-probes for the change during a run fail 1 and 4 tests, and only new ones, so those places were unguarded before - Six counter-probes for Layer of fail 2 to 3 tests each. All eight ran on stein2, not in the container. - The browser counter-probe fails 6z-x with "Layer 1" and "1" - The e2e sum of 1619 is calculated (1617 measured in CI run 276, plus 2). Neither the full suite nor the smokes ran for this state. - Not looked at in the browser: that the layer port of Properties shows the same picker as the inspector. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
d39bf83ea9 |
Pick the frame's outline when adding a collider to a sprite (#218)
"Add collider" on a sprite always made a box of the sprite's size, even when the sprite's picture is a frame that has a collision outline. The outline then had to be picked by hand as "Frame collision". - newColliderProps in colliderCommands.ts is the rule as a pure function. A sprite whose picture is a frame with an outline gets a collider with shape "frame" and that frame. Anything else keeps the box of the sprite's size. - The frame is the one the sprite shows when the collider is added, also when only this placement overrides it. - Decided here, the issue says nothing about size: a frame collider is as big as its frame, so a sprite drawn bigger or smaller with the same proportions gets the collider's scale set to match. - Decided here as well: a sprite stretched out of the frame's proportions keeps the box, because an outline cannot be stretched in one direction. Tests: - Gate: 198 files, 2914 tests (2911 + 3), exit 0 - e2e 8,42: 46 checks passed (42: +1 for 42e-ii) - Three new unit tests in tests/core/collider-parts.test.ts: the frame with undo, the scale with an overridden frame, and the cases that keep the box - 42e-ii goes through the context menu: "Frame collision" with scale 2 on the sprite wearing the wedge frame, a box on a sprite without a picture - Five unit counter-probes fail, 1 to 2 tests each. They ran on stein2, not in the container. - The browser counter-probe fails 42e-ii with shape "box" - The e2e sum of 1617 is calculated (1616 + 1). Neither the full suite nor the smokes ran for this state. - Vector art has no test of its own, it takes the same branch as a whole texture. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
defc34bd09 |
Add a single prompt that builds the engine from scratch
The prompt is meant as a benchmark: one agent, one prompt, and the goal is a finished engine rather than a measure of how far it gets. - The scope follows the decisions made on 2026-10-01. In: projects, scenes with a hierarchy, camera, containers with nesting and instance edits, behaviors, shaders and particle effects as graphs, colliders, animations, text with custom fonts, textures and sprites, state machines, variables assets, structs, typed lists, sound, the console and custom curves. - Out: accounts, plugins, the performance panel, export and player, tilemaps, the pixel art and vector art editors, run in editor, snapshots and hierarchy groups. - The stack is fixed except for the UI framework, which the agent picks itself. The interface is DOM, Pixi only draws the views. - "Done" is checkable: three sample games with smoke tests, numbered acceptance criteria and an ACCEPTANCE.md the agent has to fill in. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
d6171fd386 |
Preview pixel art and curve fields once per frame too (#274)
The previous commit left two findings open: in the pixel art tab the palette colour and the opacity of a layer still told the editor about every input, and so did the place fields of the curve tab. They now follow the same rule as the inspector rows. - PixelArtSide writes colour and opacity into the model on every input and requests the preview through a FramePreview. Both commits cancel a waiting preview first. - CurveCanvas.movePicked does the same for the t and v fields of the picked point. Showing another curve cancels a waiting preview. - Also fixed: when the colour picker ended on the colour it started from, the model was put back but nobody was told, so the previewed colour stayed on screen. commitColor now says so, as the opacity already did. The gain was not measured for either tab, only the behaviour is tested. Tests: - Gate: 198 files, 2911 tests (2907 + 4), exit 0 - e2e 47,48,53: 117 checks passed - Four new unit tests in tests/editor/frame-preview.test.ts: colour, opacity, curve fields and the return to the starting colour - Seven counter-probes fail, one test each. They ran on stein2, not in the container. - No new e2e check, the sum stays at 1616. Neither the full suite nor the smokes ran for this state, and there was no counter-probe in the browser. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
c4a9280140 |
Preview inspector edits once per frame, not per input (#274)
Scrubbing a number in the inspector and dragging a colour stop of a gradient told the editor about every input. Pixi repeats each pointer move shortly after, so both gestures paid for two previews per move. The issue asked to measure first and to build only from 10 % per move. The measurement showed about 50 %. - previewingHandlers still writes the model on every input but requests the preview through a FramePreview of its own. The commit cancels a waiting preview first. - The preview handler of the GradientBar in shapeSection.ts does the same. - Wider than the issue describes: the row handlers also serve typing, the arrow keys and the colour picker. All of them preview once per frame now. Measured on stein2, fixture project with 120 extra objects, 120 moves 8 ms apart, three runs each: - Scrub: 236 project events before, 51 to 61 after. 45 to 50 ms per move before, 22 to 27 after. - Colour stop: 238 project events before, 50 to 51 after. 44 to 49 ms per move before, 22 after. - One preview still costs 13 to 17 ms. That price per frame is what remains. Finding, not fixed here: in the pixel art tab the palette colour and the opacity of a layer still preview on every input (PixelArtSide.ts), and so do the head fields of the curve tab. Neither was measured. Tests: - Gate: 198 files, 2907 tests (2903 + 4), exit 0 - e2e, full run: 1616 checks passed (1615 + 1), exit 0 - Smokes: 24 / 19 / 21, exit 0 - Four new unit tests in tests/editor/frame-preview.test.ts, two for a row and two for a colour stop - One new e2e check in module 02: a held scrub gave 10 project events in 12 frames, and the value is there before the release - Four unit counter-probes fail, 1 to 2 tests each. They ran on stein2, not in the container. - The browser counter-probe fails the new check with 50 events in 16 frames Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
c1cc80bb45 |
Add nodes that ask whether an animation still runs (#233)
A graph could start, pause and stop an animation and hear it finish, but it could not ask in between whether one runs or how far it is. Two data nodes answer that now. - Animation playing takes an object and an animation and gives playing, progress and time left (s). It answers at the moment it is read: right after Play animation in the same pass it says on, and in the frame On animation finished fires it says off. - With Animation at None it answers for the running animation started last on the object. - Animation length gives the seconds of one pass, from the timeline and the frame rate alone. - The rule is a pure function in src/core/model/animation.ts (playerStatus, leadingPlayer). The nodes reach an animation's length through a new field animation on GraphLookup. - animationPort is exported so both nodes show the same picker as Play animation. Decided in the issue: a Once (reset) counts as running while it blends out, and progress and time left also hold for Loop and Ping-pong, at the place of the playhead. On Ping-pong progress falls again on the way back and time left counts to the next turn. Decided here, the issue says nothing on it: - A paused animation does not count as running. Progress and time left stand still. - One that Stop animation fades out does not count as running, and its time left is 0. - Time left of a Once (reset) includes its blend out. While it blends out progress is 1 and time left is the rest of the fade. Finding, not fixed: on the node the name of the chosen animation overlaps the output label progress in the same row. That comes from the fixed node width in NodeView and was seen in the Fit view only. Tests: - Gate: 198 files, 2903 tests (2891 + 12), exit 0 - e2e 1-6: 282 checks passed, 06 has +2 for block 6z-ix - Nine unit counter-probes fail, 1 to 4 tests each - The browser counter-probe fails 6z-ix with ["false","false"] - e2e total 1615 is computed, not measured in a full run. The smokes were not run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
26cc6324af |
Stop resolving every object in every frame of a run (#287)
World.syncToModel runs after every frame of a run and resolved each object of the scene anew to compare it with its entity. With 222 entities that was most of the Editor line in the performance panel, also for entities that are never drawn. - The world keeps per entity what the model said when it last read the object (syncFromModel, spawn) or wrote it. A frame compares the entity with that and resolves nothing while both agree. - A place that differs is written as three numbers. - A value that differs is written as an override, and that one object is resolved again. An override can bring rows of its own, such as a sprite's shader. - The objects of entities that left the run are dropped only in a frame in which one left. World.remove notes that. - Editing during a run works as before: the editor has the model read again on every project event, which renews what the world keeps. One difference: a write to the model that goes past the editor and does not have it read again now stays. The next frame used to overwrite it with the entity's state. No code does that, during a run only the world writes. Measured on stein2 with a copy of the reference project of #284, ms per frame: syncToModel alone 0.92 to 0.12 (Node, 600 frames), the Editor line in the browser 1.14 / 1.20 / 1.29 to 0.25 / 0.25 / 0.27. Physics and Drawing are unchanged. Tests: - Gate: 198 files, 2891 tests (2887 + 4), exit 0 - e2e, full run: 1613 checks passed, exit 0. That measures the total that was only computed since #273. - Smokes: 24 / 19 / 21, exit 0 - Four new unit tests, no new e2e check: 60 unchanged frames resolve nothing, an edit during the run arrives and stays, an override with rows of its own is seen with them, container content in an editor run is written into the instance edits once - Seven counter-probes fail, 1 to 4 tests each Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
cfd288c4b7 |
Halve the frame time of a playing scene with a tilemap (#284)
The performance panel showed Physics and Editor as the large parts of a frame in a scene of 222 entities. Nearly all of it came from a tilemap that only stands there. - Whoever writes cells in place now says so with cellsChanged. Only then does cellsVersion compare the content. The comparison ran about 60 times per frame and was the largest single item at 14 percent. - The four writers call it: the painter during and at the end of a stroke, PaintTilesCommand on execute and on undo. A new writer that forgets it stays invisible. - The game view is drawn once per frame in play. With an animated tile in view it was drawn twice, once by the run and once by animate(). animate() now gets the redraw count from the start of the tick. - TilemapField builds the geometry of a mesh only when its source changes: the cells, the tile size, the origin or the image. For that groupByTile and layerRows return the same objects for the same list. - PhysicsSystem skips a tilemap whose source and switch are unchanged. The switch is kept because a map that starts switched off would otherwise have dead colliders once it is switched on. - World.syncToModel looks entities and assets up in a map per pass instead of a linear search per object. Measured on stein2 with a copy of the reference project, ms per frame: Physics 2.64 to 0.97, Drawing 6.89 to 2.56, Editor 1.95 to 1.31. Left open: syncToModel still mirrors every entity every frame. That is its own issue. Tests: - Gate: 198 files, 2887 tests (2883 + 4), exit 0 - e2e 1-17,28,33,39-43,59: 810 checks passed, 40 has +3 - Smokes: 24 / 19 / 21, exit 0 - Three unit tests wrote cells directly and now call cellsChanged - Eleven counter-probes fail. Two did not at first: the stroke needed a check of its own, the physics test has to start with the map switched off. That last one was repeated on stein2 only. - e2e total 1613 is computed, not measured in a full run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
6491888f56 |
Stop offering private variables as new animation rows (#270)
A behavior variable switched private is hidden from the placed object and from the Properties node, but the row dialog of an animation kept offering it. Rows that exist already stay as they are, as decided in the issue. - offeredDefs in src/core/model/animation.ts is what the dialog offers now: what the target can animate, less private variables and less what another row animates already. The filter moved there from RowDialog.ts, so it can be tested without the editor. - animatableDefsOf is unchanged. The label and the runtime read it, so a row on a variable that went private keeps playing and keeps its label. - Deviation from the proposal: under Change target the row's own variable stays on offer, private or not. The dialog would open with no property and a disabled button otherwise. Other private variables are not offered there either. - The command does not refuse a private variable. Only the dialog does not offer it. Finding, not fixed: e2e 56 fails when it runs alone, in 56a-ii from #257 (typing "9" did not arrive: focus is BODY). With 55 before it the module passes, so a full run does not show it. It was not measured on HEAD without this change. Tests: - Gate: 198 files, 2883 tests (2882 + 1), exit 0 - e2e 55,56: 27 checks passed, 56 has +56e - e2e 25: 31 checks passed, it uses the same dialog - Three unit counter-probes each fail the new test - The browser counter-probe fails 56e, mood shows up in the dialog - e2e total 1610 is computed, not measured in a full run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
5e21d473a4 |
Fix references into nested placements and on asset delete (#285, #286)
Two findings from #269. A reference from a container into a placement nested in it was dead from the start, and deleting an asset left references to the objects it took along showing "(missing)". - A reference is stored the way its holder shows the object. In a container tab that is q~x for content of a nested placement q. - referenceCandidates translates it when the container is placed: in placement r, q~x becomes r~q/x. pathOfId and idOfPath in instance.ts do the mapping, objectChoices offers the same spelling the other way. The rule holds at any depth. - Storing q/x as proposed in #285 was dropped. The container tab would then have had to translate its own ids. - No migration. All projects and snapshots hold three references, none of them with ~ or /. - Create container lets a reference to content of a member placement follow it. It was left dead before. - Paste lets such a reference point into the copy. It kept pointing at the original before. Both go through the new renamedReference. - Delete asset records what every holder shows before the cascade and clears every reference to an object that is gone afterwards, with dropReferencesToGone from #269. The row shows None, undo restores the references (#286). Tests: - Gate: 198 files, 2882 tests (2873 + 9), exit 0 - Unit object-references: 26 passed (17 + 9) - e2e 1-17,57: 566 checks passed, 57 has +57i and +57j - Six unit counter-probes fail between 1 and 3 tests each - The browser counter-probe fails 57i and 57j as described in the issues - e2e total 1609 is computed, not measured in a full run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
214218d1b2 |
Clear references to a template object in every scene on delete (#269)
Deleting an object in a container tab only cleared references inside the template. Placements in scenes and other containers kept their instance edit and showed "(missing)" in the inspector. - shownObjectIds records which objects every holder shows before the delete. dropReferencesToGone compares that with the state after it and clears each reference that no longer resolves. - A reference is resolved from its own place with referenceCandidates, the same way the runtime does. One rule covers instance edits, nested placements and full ids like p~crate. - A cleared instance edit is dropped when it now says the same as the template's own value. Without such a value a marked None stays, as if picked by hand, and Reset hands the row back to the template. - All cleared references come back with a single undo. - References that were already dead in saved projects stay untouched. Findings, not fixed here: - A reference from container D into a nested placement q is stored as q~<id>, but the copy is named r~q/<id> in every placement of D. The runtime finds nothing, so that reference is dead from the start. - Delete asset removes placed objects by cascade but does not clear references to them. Tests: - Unit object-references: 17 passed (13 + 4) - Gate: 198 files, 2873 tests, exit 0 - e2e 1-17,57: 564 checks passed, 57 has 13 (+1 for 57h) - Five unit counter-probes and one browser counter-probe fail as expected - e2e total 1607 is computed, not measured in a full run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
0f6e628616 |
Check the declared type when a graph writes a typed variable (#271)
A list variable with an item type only kept its shape in the editor. A graph could write any list into it at runtime, and the List node had no type at all. Now the declaration decides what a write may store. - Writing a list into a typed list variable leaves out every item that does not have exactly the fields of the type. The rest is written and the console names each item that was left out, like Add item does. - The same rule holds for typed list fields inside a struct variable and for lists inside typed items, at any depth. - A typed struct variable refuses a struct with a field the type does not have, and a value that is no struct. A struct with missing fields still merges in as before, so a partial Pack keeps working. - The check runs in the Variables node (own variables and Variables assets) and in the Properties node when it sets a behavior variable of another object. - Ports learn their declaration through GraphLookup: what a variable declares about its type, and which behavior owns a graph. The rule itself lives in the new src/core/graph/typedValues.ts, fitsType moved there from SetItem. - The List node gets a Type setting. With a type its items are shaped and marked, follow the struct asset, count when the type is deleted and offer Split into fields in For each. The mark never reaches the running game. - A type on a List node holding something other than structs is refused with a console line. - Setting the List node back to None removes the marks, unlike a list variable. A node list has no per-item type row, so marked items could never be freed there. Undo brings everything back. - tests/core/kinds.test.ts looked for the value as the first parameter of a literal. Type now comes first on the List node, so the test finds the value by its key. Tests: - Gate: 198 files, 2869 tests (2856 + 13), CHECK EXIT 0 - e2e 55: 19 checks (+55j write at runtime, +2 for 55k List node) - e2e 54: 16 checks, smokes 24 / 19 / 21 - Five mutations, each fails between 1 and 4 tests - e2e total 1606 is computed (1603 + 3), no full run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
f2282a81fb |
Stop loading the bold face the editor never draws in (#273)
The editor waits for every face in FACES before it starts. Source Sans 3 in 700 was one of them, but nothing in src/ sets a weight above 600, so it cost a download and a wait for nothing. - fonts.ts loads Source Sans 3 in 400, 500 and 600 plus Source Code Pro 400. The 700 file (15 596 bytes) is gone from src/ui/fonts. - A search for bold, font-weight and HTML headings in src/, web/*.html and the player found nothing that would fall back to 700. - fonts.test.ts now checks both ways. A loaded Sans face must be set somewhere in src/, and every .woff2 in the font folder must be in FACES, so a stray file the glyph test would still read shows up too. - The weight scan moved into usedWeights(), since two tests need it. Tests: - Gate on melektaus-atx: 197 files, 2856 tests (2855 + 1), exit 0. - e2e 1-14: all passed, 492 checks, loaded faces ["400","500","600"]. - e2e 14 alone fails its two tab width checks, as known: it needs the documents 7-13 open. The font check was green there as well. - dist/assets holds no 700 file after the build. - Counter check: loading 700 again with its file fails the new "loads no face the editor never sets". - Counter check: the file back without a FACES entry fails "... and no other" and names the file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
cbda6d64a7 |
Let only the owner delete a project, not an administrator (#272)
#179 decided that nobody deletes a project on someone else's behalf. The API still let an administrator do it, because the admin store treated every project as its own. Option A from the issue is now in place: opening, editing, snapshots and release stay open to the administrator, deleting stays with the owner. - Store::delete checks the owner under the save lock and throws NotOwnerException for anyone else. It follows the pattern of StaleRevisionException. - The route answers that with 403 "Only its owner may delete a project". A file without an owner is covered the same way. Non-admins still get 404 for a project that is not theirs. - The editor has no way to delete a foreign project, the project picker lists only one's own. Nothing changes there. - An account that owns projects can no longer be deleted by an administrator at all. Its owner has to delete them first. - e2e 49 deleted the throwaway projects as administrator, both in its sweep and in 49i. It now signs in as the owner for that. 49i checks both answers. Tests: - tests/php/accounts.test.php: 56 checks passed (was 52, one check turned around, five new) - e2e 49: 22 checks passed (was 21), deleteByAdmin 403, deleteByOwner 204 - Gate: 197 files, 2855 tests passed - Counter-checks: without the owner check 4 PHP checks and 49i fail - The e2e total of 1603 is computed (1602 + 1), not measured Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
f14c4ecb77 |
Show the frame rate the measured times allow in the overlay (#275)
The fps line of the performance overlay stops at the display's rate, 60 on most screens. It does not tell how much room a game has left. The new top line answers that from the times the overlay measures. - possibleFps in src/editor/frameMeter.ts is 1000 ms over the sum of the five areas, rounded down. It has no number while nothing took any time, and the overlay shows a dash then. - The editor's share is in the sum, because it costs time while playing in the editor. A released game can do a little better. - It is an upper bound. The GPU's time is not measured and not in it. - e2e 59e reads the lines one further down and checks the new line against the times it measured itself. Tests: - Unit frame-meter: 11 passed (+2 for possibleFps) - Gate: 197 files, 2855 tests, CHECK EXIT 0 (2853 + 2) - e2e 59: ALL PASSED, 19 checks (+1), on melektaus-atx - Counter-check: possibleFps without the editor's share fails the unit test - The e2e total of 1602 is counted (1601 + 1), not measured Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
77f4c2c1a3 |
Show entities and time per area while playing in the Game View (#275)
A game that stutters gave no hint where the frame goes. The Game tab now has a Performance switch. It shows a small overlay in the top left corner of the Game View while a game runs: fps, the number of entities and the mean milliseconds per frame for each area, averaged over half a second like the fps counter. - src/editor/frameMeter.ts holds the pure rule. FrameMeter sums the areas per frame, closes a window every 0.5 s and formats the lines. - World.timeSystem times every system the world ticks. PlaySession maps the systems to Behaviors, Physics and Animation and sets the hook again after Load scene. - Animation has its own line. Counting the AnimationSystem under Behaviors would have skewed that number. - Drawing is the build of the game picture plus app.render(). It is main thread time only. The GPU work after it is not in the number, and app.render() draws the whole canvas including the panels. - Editor is the rest of the editor tick after the systems and the game picture: mirroring into the model, panels and console. - The overlay lets clicks through to the game. It shows dashes until the first window closes. The switch is off by default and is not saved. The released game (#266) runs without a meter. Tests: - Unit frame-meter: 9 passed. - Gate on melektaus-atx: 197 files, 2853 tests, CHECK EXIT 0. - e2e 59 on melektaus-atx: ALL PASSED, 18 checks (13 before, +5 in 59e with a real mouse on the switch). Run twice. - Three mutations were caught: the editor share not subtracting the systems (2 unit tests), systems running in pause not reported (1 unit test), the overlay getting no times (59e). - The e2e total of 1601 is computed (1596 + 5), not measured. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
d471150efb |
Keep e2e 58 in the scene view after the eyedropper checks (#276)
The CI run for
|
||
|
adminandClaude Opus 5.5
|
ded71a92c9 |
Pick the target of an object variable in the Scene View (#276)
Choosing the target of an object variable meant finding its name in a dropdown. With many objects, or several with the same sprite, the name alone does not say which one it is. The eyedropper lets you click the object where you see it instead. - Every object dropdown gets an eyedropper button next to it. This covers variables, struct fields and list entries, in the Inspector and in the Variables tab. - The button opens the Scene View or the container tab the object lives in. A strip above the view names the field and the current candidate and offers Take and Cancel. The cursor is a crosshair. - A click picks a candidate and writes nothing yet. The candidate gets its own cyan frame, the selection stays on the object with the variable. - Clicking follows the selection rule. `pickOn` in `src/viewport/pick.ts` uses `pressOn` and only counts objects the dropdown offers. A click on empty space keeps the candidate. - Enter or Take writes through the same handlers as the dropdown. That is one undo step, or none if nothing changes. - Esc, Cancel, a right click in the view or a second click on the button cancel without writing. So do a tab switch, another selection, undo, redo, play and loading a project. - After Take or Cancel the editor goes back to the tab the pick came from. - While the eyedropper runs, the view does not drag, marquee, paint, show handles or open its context menu. Panning and zooming stay. Found on the way: - Started from the Variables tab, the click missed the object. The Scene View is hidden while that tab is open, and the layout skips hidden children. The strip appeared then, the view moved down by 32 px but kept its old height. `Editor.display()` now lays the row out again after showing it. Tests (on melektaus-atx): - unit pick and object-pick: 27 passed, 13 new - gate: 196 files, 2844 tests, CHECK EXIT 0. Two more come from svg-icons, which checks every icon twice. - e2e 57: ALL PASSED, 12 checks, 5 new (57e to 57g) - Counter-checks work. Without the layout fix 57g fails. Picking objects the dropdown does not offer fails 2 tests. Writing on cancel fails 2 more. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
42647d309b |
Run the gate on stein2 and the suite on the WSL runner (#283)
The gate needs no GPU, so it goes back to the runner on stein2 (`game-maker-ci`, same image). The suite stays on `game-maker-wsl` on melektaus-atx. The two jobs no longer share the four cores of one runner while they overlap. - The change of `runs-on` for the gate is the user's own edit. - The header comment now says which job runs where. GM_TEST_WORKERS stays at two, for the day both jobs share one runner again. Tests: - Both runners were online in Gitea before the push (`stein2-docker`, `melektaus-atx-wsl`). - Prettier passes on the file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
543b052136 |
Stamp the release in the page for the drag check in e2e 41 (#282)
The first CI run after
|
||
|
adminandClaude Opus 5.5
|
6d4247ebf8 |
Paste where the view shows it, editor offsets included (#263)
Paste mixed two frames. The pointer comes from the Scene View, which draws every object moved by the editor offsets of it and its parents. Paste turned that point into a local position through the parents' plain world transform, so the copy landed beside the pointer by the offset of the parent it went into. Without a pointer, an original with an offset of its own had its copy placed beside its game position, which the view does not show. - A copied root now remembers where the view showed it (`ClipRoot.shown`, from `editorTransform`) instead of its world position. - New `editorLocalPositionFor` in `properties/transform.ts` turns a shown point into a local position through the parent chain with the editor offsets, as the view draws it. Paste uses it. - The copy itself still gets no editor offset of its own, as since #240. In the game it stands where it was drawn, minus the offsets of its parents. - Re-parenting under a parent with an offset still keeps the game position, so the object jumps in the view by that offset. That is a different rule from paste and is left as it is. Tests: - Unit (clipboard, groups, references, transform): 50 passed, 2 new. - e2e 51 on melektaus-atx: ALL PASSED, 12 checks, 1 new (51f). - Gate on melektaus-atx: 195 files, 2829 tests, CHECK EXIT 0. - Counter-checks: without the paste fix both new unit tests fail (copy at 540/470 instead of 500/500) and 51f fails by exactly the 60/-40 offset. With the root remembering its world position again, the own-offset unit test fails. - No full run. The e2e total of 1591 is 1590 plus 1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
9cff5282b6 |
Wait for the dragged body before checking the hold in e2e 41 (#282)
The drag check in e2e 41 was red about every other run under WSL and twice on stein2. The diagnosis in the issue was wrong: "held" were not two samples but the first and the last one, and -3475 is the height of the ninth of ten moves. The last move reaches the body a few frames later (preview, physics step, drawn one step behind). Reading right after the move sometimes beat it, and slower frames under WSL made that likely. - The check first waits until the crate is at the pointer, for at most 2 s. If it never gets there, the check fails with "arrived: false" instead of stopping the run. - The page then records the crate's height in every frame and the check fetches the samples once. The count follows the frames, not the CDP round trips, and the check asks for at least 30 held frames. - Held and released samples are split by page timestamps taken right before and after the mouse release. "Fell" is the last sample within 200 ms of letting go. Tests: - e2e 41 on melektaus-atx (WSL): 6 of 6 runs ALL PASSED, 21 checks, 59 to 64 held frames. - Gate on melektaus-atx: 195 files, 2827 tests, CHECK EXIT 0. - Counter-check: without passing the hold to the world the crate stays on the ground and the check fails. - Not caught by this check: a body let go without zeroing its speed. After a second of holding the speed is 0 anyway. The unit test "grab too short for a physics step" covers it, as before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
ffedd81803 |
Read many pixels from one render in the e2e checks (#283)
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> |
||
|
adminandClaude Opus 5.5
|
857e23c69b |
Run CI on a WSL2 runner with the RTX 4070 (#283)
stein2 should stop carrying the full e2e run. The new runner lives in the WSL2 distro game-maker-ci on melektaus-atx and uses the same image and fonts. Its GPU sits behind /dev/dxg and Mesa's d3d12 driver, which headless chromium cannot reach, so the harness gets a second way to start the browser. - GAME_MAKER_BROWSER=wsl starts an invisible Xvfb and an X11 chromium on ANGLE/GL. Without the variable, stein2 keeps the headless Vulkan launch exactly as before. - The harness logs the renderer and stops on software rendering. Under WSL a missing d3d12 driver falls back to llvmpipe, which still offers WebGL 2 and would pass unnoticed. - The CI image gains xvfb. The font set is unchanged: fc-list gives the same 149 fonts with the same hash. - ci.yml runs both jobs on the new label game-maker-wsl. The stein2 runner keeps game-maker-ci, so moving back is one line per job. Findings: - Wayland on weston reached the GPU too, but drew at about 40 fps and let the editor clock stall in long runs (GM_STEPS=1-17: 2 failed). - Vulkan through a self-built dzn caps ANGLE at ES 2.0 because VK_EXT_provoking_vertex is missing, so chromium drops the GPU. - The flaky check in e2e 41 (#282) fails more often under WSL: two of three solo runs were red. Tests: - ./check.sh: 195 files, 2823 tests passed - WSL container, GM_STEPS=1-17: ALL PASSED, 553 checks - WSL container, full run: 1589 of 1590, the one FAIL is #282 - WSL container, smokes: 24 / 19 / 21 checks - stein2 headless, e2e 39+40: ALL PASSED, 100 checks Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
adminandClaude Opus 5.5
|
9b5eeaab57 |
Empty a deleted tile's cells on every tilemap (#268)
Deleting a tile left the cells painted with it in the tilemaps. They drew nothing, blocked nothing and did not show in the delete dialog, but still read as a tile for "Tile at" and stopped a fill. - DeleteAssetCommand empties the cells of a deleted tile on every layer of every tilemap in the same step. Undo puts the tile and its cells back together. - The delete dialog counts them, for example "3 cell(s) on 1 tilemap(s) are painted with it and become empty." (cellsPainted). - presentTile treats a cell naming a missing tile as empty. "Tile at" (World.tileAt) and floodFill use it, so cells already in existing files read and fill as empty. There is no migration. - The brush, the rectangle and the fill check the tile in hand. A tile deleted after it was picked no longer paints new invisible cells, the editor says "Choose a tile to paint with." instead. - Unchanged on purpose: the picker still reports such a cell as naming a tile that no longer exists, and erasing removes it. Tests: - ./check.sh: 195 files, 2823 tests, CHECK EXIT 0 (+6). - GM_STEPS=39,40: ALL PASSED, 100 checks (4 new in 40t). - Full e2e sum is now 1590 (1586 confirmed by CI on round 8, plus 4). - Counter-checks, all fail as expected: 5 in unit tests (delete without emptying, undo without cells, fill and "Tile at" without presentTile, wrong map count) and 3 in the browser (dialog without count, delete without emptying, brush without the check). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |