matthewtrahan
2d82ff6866
ui: More settings -- bed pad, wall lines, sway braces, wide-face coverage, plate only and the display toggle fold into one section grouped by intent (Hold the part / Tall parts / Where supports go / Display), with an "N changed" count and a dot on each changed field
...
Local issue 42, step 3. In view by default: the printer row, Placement, Tines and
Cutouts. Every control keeps its id; the Clearances, Sway, Walls and Display
headers and their recaps go. A field counts as changed only while the build reads
it (a hidden or disabled field doesn't), and Wall lines not when a big nozzle took
its default away (ui/more.js).
2026-10-15 22:09:44 -05:00
matthewtrahan
bd6470b17b
ui: the layer height and support gap are remembered across visits like the nozzle -- they're the printer's, so a saved session neither saves nor applies them; interface material stays with the part
2026-10-15 22:09:44 -05:00
matthewtrahan
3ee3de1fb2
ui: the layer height stays shown while Interface material is on -- the band is one layer of the wall top, but with tines off and a Sure hold pad the field hid and the printer line dropped it (review)
2026-10-15 21:36:44 -05:00
matthewtrahan
ec529e0a5b
ui: a printer row at the top of the settings -- nozzle, layer height, support gap and interface material on one line (0.4 nozzle · 0.2 layers · 0.2 gap) that unfolds in place; Placement stands under it without a header
...
Local issue 42, step 2. The four fields move unchanged (every id kept); the Tines
and Clearances recaps stop repeating the layer height and gap. The line names only
what the build reads: layers drops out while the layer height is hidden.
2026-10-15 21:36:44 -05:00
matthewtrahan
497f857c3d
export: waits while a rebuild is out (review) -- a nozzle or layer change exported mid-build named the old fins with the new settings, the mismatch #279 is about
2026-10-15 21:04:59 -05:00
matthewtrahan
66a5e4c62a
export: the file name carries the nozzle and layer height the fins were built for ( #279 ) -- part-fins-0.6nozzle-0.24layer.3mf, so the slicer gets set to match: a 0.4 build sliced at 0.6 prints its walls as one line and drops tines between 0.24 layers
2026-10-15 21:04:59 -05:00
matthewtrahan
bff031d484
placement: review fixes -- the button group takes its screen-reader name from the translated Placement label, and a focused button's ring draws over its neighbour
2026-10-15 20:49:39 -05:00
matthewtrahan
1ccec211cf
placement: Auto / Full coverage / Draw as three joined buttons with a one-line note on what the chosen mode does, in place of the dropdown (local issue 42) -- the select stays, hidden, as the one place the mode is kept (undo, saved sessions and ui-smoke read and set it); the buttons drive it and follow it through undo/redo
2026-10-15 20:49:39 -05:00
matthewtrahan
e562b67680
rotate: review fixes -- the drag label hides while any toolbar popover is open (it ran under Suggest, and over Rotate opened mid-fade), drops aria-live (a drag queued an announcement per frame), and follows the badge row's sideways scroll on a phone
2026-10-15 18:42:46 -05:00
matthewtrahan
f9351d8239
rotate: a ring drag with the Rotate popover folded shows its angle in a small label above the Rotate badge (the turn so far, then the pose), fading 1.2 s after release -- the readout moved into the popover in #286 , so a drag showed no angle; ui-smoke's ring drag now drives the rotate gizmo (it found the Section plane's)
2026-10-15 18:42:46 -05:00
matthewtrahan
684ab98cfb
toolbar: review fixes -- folding the Rotate popover disarms Lay a face flat (its button was the only sign it was armed, and a stray click re-laid the part), a popover scrolls inside a short window and stays inside the window's left edge, Esc hands focus back to the badge, Undo/Redo take their name from the translated title, the Suggest chevron's CSS goes
2026-10-15 15:51:17 -05:00
matthewtrahan
06a1322466
toolbar: the left pane's pose and view tools move into the bottom view toolbar -- Undo/Redo, then Rotate (90° steps, Reset, Lay a face flat, rotate rings), Suggest, Strength (lit while a load is set), Section and View (Mouse preset, Show layers); each popover opens above its own badge and extends right; the left pane keeps only the stats card (local issue 42)
2026-10-15 15:51:17 -05:00
matthewtrahan
292e8a16b6
plugins menu: review fixes -- the command line's slicer list moves to its plain line (the orange note is for warnings), the LayerLing route is said to be untried, Deno gets its command, Sie in German, Brazilian Portuguese, and Arabic agreement
2026-10-15 15:29:21 -05:00
matthewtrahan
20c2a9e95d
plugins menu: the MCP server, and who the command line is for -- slicers with no plugin system (Bambu Studio, QIDI Studio, Creality Print, Snapmaker Orca, Flashforge) use it in place of a plugin; design tools with an MCP server but no plugins (LayerLing) reach the fins through the MCP server
2026-10-15 15:29:21 -05:00
matthewtrahan
0bea7ad741
pad: review fixes -- a coarse block whose fan folds over (near the oval's tips: reversed facets lying on the top, which closed/oriented checks pass) is meshed fine, its hub is its corners' mean; a Custom pad thinner than half a layer marks the triangles its ramp follows
...
Reviewer fuzz (210 random boxes, Light/Custom, layers 0.08-0.4, gaps, grips, margins): every slice now equals the full 0.1 mm grid's at 13 heights, no folds, no open edges; 30 cases folded before. Main's band scan was itself coarse in a few places (tips, thin Custom pads), so a few golden slices move to the full grid's.
2026-10-15 14:30:23 -05:00
matthewtrahan
bb649270a7
pad: the Light/Custom pad is 0.1 mm only in the 1.2 mm blocks the part comes near -- every column shared one set of rows, so one dip near the rim made a 128 mm figure's pad 2M cells in Suggest (RangeError at 4x); now 19k, and every slice is the same segments as before
...
Fine blocks keep the full grid's nodes and diagonals, and the rim keeps every node, so the pad's sections at ten heights match main segment for segment (cube X20-45, Light/Custom, 0.12-0.28 layers; man bust, buster, Isaac). A coarse block beside a fine one fans from its centre through the shared nodes, so there are no T-junctions. A coarser grid was the other option and was dropped: at a 0.3 mm cell the gap closes to 0.080 mm at a cube's corners. Goldens re-recorded: only the pad's triangles changed.
2026-10-15 14:30:23 -05:00
matthewtrahan
ab0b25dbe8
mcp: review fixes -- the tools say they may replace a file (output can name one), object must be a number or name, an own-property check on tool keys, ids must be strings or integers, no 2025-03-26 (batches), errors in setting names throughout, and the release note's add command takes a full path
2026-10-15 14:27:57 -05:00
matthewtrahan
25a317722e
mcp: build it in CI, drive it over stdio on Node outside the checkout, and publish support-fins-mcp.mjs with the other plugins
2026-10-15 14:27:57 -05:00
matthewtrahan
9919c32bb8
mcp: Support Fins as an MCP server -- fin_part and fins_only for AI clients (Claude Code, Claude Desktop), each a CLI run underneath, so a tool call writes the CLI's exact bytes; settings schema from the CLI's flags in the site's units; stdio on Deno or Node
2026-10-15 14:27:57 -05:00
matthewtrahan
e0122ae429
cli: export the flag table and let the bundler take another entry -- the MCP server builds its settings schema from the same flags and ships as the same one-file bundle
2026-10-15 14:27:57 -05:00
matthewtrahan
4b46f94fec
ui-smoke: the PETG step sets Support gap 0.3 -- the Material dropdown it set is gone ( b625eed), so every run failed at that step; README counts 87 steps
2026-10-15 11:42:42 -05:00
matthewtrahan
b59de82e3a
cutout: review fix -- a piece that no line separates from a nearby hole (it would already overlap it) stays ungrown instead of growing further in
2026-10-15 11:42:12 -05:00
matthewtrahan
877ffc9100
calibration: rebuild the cutout coupon's print files with the review fixes -- still no edge on more than two faces, PrusaSlicer time and filament unchanged
2026-10-15 11:42:12 -05:00
matthewtrahan
fe90b6c857
cutout: review fixes -- each strip piece grows by moving its edges out (0.4-0.6 eps, its own distance) and clipping to a line separating it from each nearby hole, in place of slab overlaps, corner slides and corner pulls: those left pieces non-convex (a folded cap, 68 faces turned inward on a lattice wall) and opened sub-micron gaps; now every piece is convex, none reaches into a hole or leaves a gap, and the test checks for inward faces too
2026-10-15 11:42:12 -05:00
matthewtrahan
c228633023
calibration: rebuild the pad and cutout coupons' print files for #223 -- the pad's predated the pad top-winding fix (1833 reversed faces, same triangles), the cutout's had flush pieces (951 shared edges); PrusaSlicer time and filament unchanged
2026-10-15 11:42:12 -05:00
matthewtrahan
cc2a75dd17
cutout: a cut wall's pieces overlap instead of sitting flush -- slabs stacked at a cut, bands ending on a post's cap, bands meeting where a short wall's zone closes, corners touching across a strip side all shared welded edges (4-6 faces to an edge), which slicers that weld by position report as non-manifold ( #223 : 951 on the cutout coupon); now none, same printed shape
2026-10-15 11:42:12 -05:00
matthewtrahan
c7d20908d7
surface: review fixes -- a NaN vertex (corrupt binary STL) gets a -Infinity floor, where a NaN floor unsorted the lists and the early exit returned 2 over a true 1; the test now warms each grid past the sort gate, so it checks the lowest-first path and not only the plain scan
2026-10-15 07:39:01 -05:00
matthewtrahan
cf8eb08268
surface: sort a grid's lists lowest-first only once it has been asked enough to repay the sort -- sorting every short-lived region array slowed a dense mini's Suggest 5.3 -> 10 s
2026-10-15 07:39:01 -05:00
matthewtrahan
cf0c950d1b
surface: the lowest-height lookup walks each cell lowest-first and stops once nothing left can reach below the best -- a tall column scanned every surface stacked above it; same minimum, bit for bit
2026-10-15 07:39:01 -05:00
matthewtrahan
98c0d5a94d
suggest: review fixes -- a ranking in flight is dropped (worker terminated) by whatever clears the list (new part, undo, reset), a message error falls back inline, and the test ranks a torus whose answer the gap really changes
2026-10-14 21:32:24 -05:00
matthewtrahan
0f9cd68cf5
suggest: rank in a Worker with the page's tunables -- a 250k-face figure froze the page for 14.5 s; now the longest frame is 17 ms and the ranking is the same, default or changed gap; inline when no Worker can run
2026-10-14 21:32:24 -05:00
matthewtrahan
662b6b3a8f
import note: a file with negative parts says the walls inside one show in the slicer once it slices -- the export carries the negative part, and a slicer draws it over the walls in its cut until then ( #104 )
2026-10-14 21:25:19 -05:00
matthewtrahan
6d4fbb26b7
3mf: review fixes -- a coloured part's material table and the first slicer part no longer share resource id 5 (the slicer parts start past it), and the bed shift counts the slicer parts so a modifier reaching past the part stays on the bed; pinned with the interface crest on, which all three slicers still read as one two-part supports object
2026-10-14 21:25:19 -05:00
matthewtrahan
07bd5ca224
cli: a 3MF output carries a slicer's own parts back out the way the site does, in the separate form so a negative part can't cut the walls in its hole ( #104 ); an STL output says it left the modifiers out
2026-10-14 21:25:19 -05:00
matthewtrahan
2fac7171f7
export: a 3MF with a slicer's own parts goes out as the file had it -- the part uncut beside its negative parts, modifiers and blockers, posed with it, for the slicer to cut again ( #104 ); the fins and the saved session stay the cut part's, and opening the export here again cuts to the same part
2026-10-14 21:25:19 -05:00
matthewtrahan
2029c4b63a
3mf: a slicer's negative parts, modifiers and support blockers are written back beside the part, each its own kind with its settings verbatim ( #104 ) -- Bambu/Orca and PrusaSlicer 3 as one assembly named in both configs (each reads its own, checked with all three slicers' command lines), PrusaSlicer 2 as triangle ranges of one mesh; never into the object holding the fins, since a negative part cuts every part of its object
2026-10-14 21:25:19 -05:00
matthewtrahan
1bcda45da7
render: review fixes -- keep drawing while the camera moves at all (the orbit's easing tail moves under OrbitControls' 1 mm 'change' threshold), and the fps count restarts after a pause
2026-10-14 21:14:50 -05:00
matthewtrahan
75c31cd2e4
render: draw while something may be changing (input, orbit easing, a mesh added or removed, a build landing) and every 500 ms otherwise -- idle went from 60 frames a second to 2
2026-10-14 21:14:50 -05:00
matthewtrahan
511d81fc91
drag: review fixes -- the paint record keys on the colour attribute itself, and the comment says which parts still shade every frame
2026-10-14 21:10:10 -05:00
matthewtrahan
35035da5df
drag: a heavy part's live shade leaves frames between runs, and a repaint writes and uploads only the faces whose colour changed -- a ring drag on a 1.56M-face mini went 17 -> 48 fps; release still shades in full
2026-10-14 21:10:10 -05:00
matthewtrahan
b044d5846e
picks: reflow pick.js's header comment
2026-10-14 21:05:50 -05:00
matthewtrahan
bc7e9ed52d
picks: review fixes -- the vendored BVH is cached a year like occt and manifold, and pick.js's load-cost note stands on its own
2026-10-14 21:05:50 -05:00
matthewtrahan
601c18c665
picks: every Mesh raycast through three-mesh-bvh (vendored 0.9.15, MIT) -- a Draw hover on a 1.56M-face mini took 75 ms a pointer move, now under 1 ms with the same hits; the BVH builds on a big mesh's first pick
2026-10-14 21:05:50 -05:00
matthewtrahan
0a01443560
grids: review fixes -- a pierce found on the fine grid reports the one the old 64x64 walk met first (9% of pierces moved up to 1.5 mm, and callers trim the station nearest it), buildFins drops its seated part when done, gridsize's comment counts cells right
2026-10-14 20:30:08 -05:00
matthewtrahan
d353ca302e
grids: solidClearance walks the old 64x64 grid for a big wall triangle's box (thousands of fine cells were half a 512 mm build), seatedPartTris hands back one array per pose so its surface grid is built once, and one face per cell
2026-10-14 20:30:08 -05:00
matthewtrahan
6bd5cdc143
grids: size the surface and inside grids to the mesh, not a fixed 64x64 -- a cell's share of a dense part grew with its face count and every station's query scanned it (~80% of Auto on a 768 mm figure)
2026-10-14 20:30:08 -05:00
matthewtrahan
4dbb6fc094
cut: review fixes -- a cutter that will not load leaves the file open uncut with a note (and is retried next file) instead of refusing it, an inside-out negative is turned outward so it cuts rather than adds plastic, the shell/debris repair is for negatives only (the part is never changed without a word), and negatives that remove the whole part leave it uncut with a note
2026-10-14 20:21:26 -05:00
matthewtrahan
31228abb1d
cut: a slicer's negative parts are cut out of the part on import, so the fins are built for what prints ( #104 ) -- Manifold (manifold-3d 3.5.4, Apache-2.0, vendored, loaded only for a file with one); a scanned negative of closed shells plus sliver debris still cuts, a part or negative that is not a closed solid is said and left uncut; paint stays on the faces the cut leaves whole; the CLI bundle carries the WASM; the import note says how many were cut
2026-10-14 20:21:26 -05:00
matthewtrahan
bb6d714d57
cut: WIP -- a slicer's negative parts cut out of the part with Manifold (manifold-3d 3.5.4 vendored, loaded on first use); not wired into the import yet
2026-10-14 20:21:26 -05:00
matthewtrahan
f4166105e6
3mf negative parts: review fixes -- a PrusaSlicer 2 second instance (a wrapper over the first one's mesh) keeps its negative out (it read solid), a part placed twice is two extras rather than one merged shell, the CLI warns when it leaves one out, and a Bambu plate whose part ids collide is pinned
2026-10-14 20:21:26 -05:00
matthewtrahan
fd6c962820
3mf: PrusaSlicer 3's part types (PrusaSlicer3_project.json) are read too, and the three slicers' own files are pinned -- a cube with a negative cube saved by Bambu Studio, OrcaSlicer and PrusaSlicer 3, stripped to the model and the part-type config
2026-10-14 20:21:26 -05:00
matthewtrahan
b2bd24ef16
3mf: a slicer's negative parts, modifiers and support blockers/enforcers stay out of the part ( #104 ) -- Bambu/Orca model_settings.config subtypes and PrusaSlicer Slic3r_PE_model.config volume ranges are read, and each comes back by kind beside the part: a PrusaSlicer negative volume was read solid into the part, a Bambu one (type other) was dropped as unknown; the import note says they are not in the export yet
2026-10-14 20:21:26 -05:00
matthewtrahan
e2b8e846ba
weld: review fixes -- fold the hash's high bits into the slot (a power-of-two micron grid piled into a few slots, 45x slower), NaN_KEY before its use, the edge-case test covers +-Infinity and float32-huge
2026-10-14 15:45:24 -05:00
matthewtrahan
33cbb2e695
tests: README lists weld.test.js
2026-10-14 15:45:24 -05:00
matthewtrahan
32141cf3aa
weld: number-keyed hash tables, not string-keyed Maps -- a 1.5M-face part welded in 3.5 s on the page thread at load, now 0.5 s; same keys, same first-seen ids, identical topology
2026-10-14 15:45:24 -05:00
matthewtrahan
2667473892
nearface: review fixes -- the first pass walks only each ring's shell (a probe in a hollow sphere's middle paid the whole cube per ring), and the never-reached forced bucketing goes
2026-10-14 14:50:40 -05:00
matthewtrahan
c8b4f98afe
tests: README lists nearface.test.js
2026-10-14 14:50:40 -05:00
matthewtrahan
c272fc2a34
tines: aim each tine through a grid, not a scan of every face -- biteDirsAt measured all faces for every tine (tines x faces), 60% of Auto on a 512 mm figure; nearestFaces gives the same faces, ties and order bit for bit
2026-10-14 14:50:40 -05:00
matthewtrahan
2ac92aa5ed
3mf colours: review fixes -- a paint string is copied on read (a scanned slice kept the whole decoded model file alive for the session), and a test that paint inside a Bambu object file (p:path) comes through
2026-10-14 14:31:54 -05:00
matthewtrahan
e18334bca1
3mf: a painted part keeps its colours through fins + export ( #236 ) -- Bambu/Orca paint_color and PrusaSlicer mmu_segmentation ride verbatim on the same triangles, core basematerials/colorgroup colours come back as one material table; the supports stay uncoloured
2026-10-14 14:31:54 -05:00
matthewtrahan
2cfd7c607d
tines: review fix -- the spec says how much solid a moved end ran through apart from how far it came in
2026-10-14 13:42:33 -05:00
matthewtrahan
f4e286aae5
tines: an end stops at the first solid -- the far probe (2x the reach) passed a part thinner than it and ran the tine through, and bisected back from a piece beyond a thin near one (cone/needle/tube X40: tine corners 0.3-0.5 mm into a thin edge at 0.2 layers); where the walk finds nothing the end is unchanged to the bit
2026-10-14 13:42:33 -05:00
matthewtrahan
6c3860ad79
wedges: review fixes -- the test walks each wedge's tines back from its end (36-vert chunks ran misaligned past the blade and missed a quarter of portal's), and the spec's count is portal's real 41 -> 42
2026-10-14 13:15:08 -05:00
matthewtrahan
97c6c65b30
wedges: tines follow the layer height -- a wedge's tines were always 0.2 tall with the 0.2 reach, a third of a 0.6 layer, which the slicer drops or merges (portal X45 at 0.6: 32 tines, now 0.6 tall)
2026-10-14 13:15:08 -05:00
matthewtrahan
8c8b50b137
section: review fixes for the sheet grab -- mouse and pen only (touch keeps one-finger orbit), the press still reaches the page so open menus close, Ctrl+click cancels rather than grabs, hover raycasts once a frame, the grab cursor survives Draw's hover, the drag stops at the part's reach; the toolbar clears whichever left-pane card is beside it
2026-10-14 13:07:13 -05:00
matthewtrahan
7e0415ed19
section: drag the sheet itself to push the cut (outside the part's outline; it brightens and shows a grab cursor), and the view toolbar moves bottom-left beside the rotate panel, smaller, frosted glass
2026-10-14 13:07:13 -05:00
matthewtrahan
af89d31070
view toolbar: review nits -- a badge names its popover (aria-controls), and Esc folds the open popover away
2026-10-14 13:07:13 -05:00
matthewtrahan
ab7996a94d
section: the Section badge moves to a new view toolbar along the bottom of the 3D view -- one popover of key settings at a time, a click on the open badge turns the tool off; built so more view tools can join it
2026-10-14 13:07:13 -05:00
matthewtrahan
d91f47c559
section: the handle's rings stay hidden until the section is on -- attached to the hidden handle, they drew at the origin over the empty start page
2026-10-14 13:07:13 -05:00
matthewtrahan
b07b12ddd0
section view: review fixes -- on touch the arrow and a ring could both grab one press (the rings now stand down on the arrow's own mouseDown, and a cancelled touch restores them); Shift + scroll pushes the cut 1 mm a notch, since the arrow hides when you look square at the cut; the selection, hover and draw-preview overlays stay out of the glass pass
2026-10-14 13:07:13 -05:00
matthewtrahan
7b7a3e5217
section view: one handle, a ghost, and a home in Display -- an arrow pushes the cut and two rings tilt it (no Move/Turn modes); Ghost (default) draws the cut-away half as glass, Hide leaves it out; it starts square to whichever of the part's own axes faces the camera, so a bore seen from the side opens lengthwise; a Section button in the Display header with Ghost/Hide, Flip and Re-aim under it replaces the rotate-panel checkbox
2026-10-14 13:07:13 -05:00
matthewtrahan
1a2b3effbf
section view: review fixes -- a tap on the plane's handle no longer falls through to a pick (the handle clears its axis on release before app.js reads it, so the press is remembered); Move shows one arrow along the cut's normal; the cut turns and reseats with the part
2026-10-14 13:07:13 -05:00
matthewtrahan
736cf9e5d1
section view ( #250 ): a plane you move and turn hides the part and its supports on one side, so the inside of a bore can be seen and drawn on -- clicks reach only what is shown, the cut faces paint flat slate, Flip / X / Y / Z snap it through the part; view only, never exported or saved
2026-10-14 13:07:13 -05:00
matthewtrahan and Per Hjort
813a01f537
nozzle: a free edge's outermost row steps in as the tip widens -- at 2.0 the 3 mm tip and its 2.5 mm tines hung 1.2 mm past a cube's side (X30), tabs out of the wall instead of under the part
...
Co-authored-by: Per Hjort <35040883+HjortCreations@users.noreply.github.com >
2026-10-14 12:30:30 -05:00
matthewtrahan and Per Hjort
877111da59
nozzle: review fixes -- the tine reach grows past a wider contact tip (sphere at 45deg: 6 tines at 2.0, now 26; 30 at 0.4), walls stop at 6 mm so a row never fuses, nozzles to 2.0
...
The reach is measured from the wall's centre, so a tip wider than 0.4's 0.6
swallowed the 0.5 reach and a curved underside lost its tines; the extra
half-tip is added to it (0.15 at 0.6, 0.3 at 0.8, 1.2 at 2.0; 0.4 and finer
unchanged). A wall over 6 mm -- half the 12 mm row spacing -- fused rows into
one block (3.0 x 8 lines: 6.7 kg on a 252 g plate), so walls and wedges cap
there, the page offers only the Wall lines that fit the nozzle, and the
nozzle stops at 2.0 (2.5's two-line wall is already 6.25). tests: every
offered nozzle x lines stays under half the row spacing; the sphere keeps 70%
of its 0.4 tines at 2.0 with nothing inside the part.
Co-authored-by: Per Hjort <35040883+HjortCreations@users.noreply.github.com >
2026-10-14 12:30:30 -05:00
matthewtrahan and Per Hjort
e44cb1d846
nozzle: large-format sizes 1.0-2.0 mm, and a nozzle guide -- the same scaling rule (a 2.0 nozzle builds a 5 mm wall); layer height stays capped at 0.6
...
From Per Hjort's #210 , which took nozzles up to 3 mm (2.0 here, see the review fix). Layer height stays at
0.6: the gap, flanges and pad are fixed heights tuned for thin layers. The
guide (docs/NOZZLE.md, adapted from #210 's) lists the widths per size, the
limits (thicker walls fit fewer places) and what to check in the slicer. 0.4 is
unchanged (sweep vs main: no mesh changed).
Co-authored-by: Per Hjort <35040883+HjortCreations@users.noreply.github.com >
2026-10-14 12:30:30 -05:00
matthewtrahan
e15f31548e
tines: review fixes -- the reach multiplier stops at 3x (a slicer plugin's own layer height can pass the site's 0.6), and a test that a 0.6-layer tine's snapped top is clear of the part at its mid-layer
2026-10-14 09:32:07 -05:00
matthewtrahan
9d4a0eeddc
tines: the reach follows the layer height -- at a fixed 0.5 mm a thick layer found no solid under a shallow underside and built every wall there with no tines
...
The reach probe sits at the tine's mid-height, half a layer under the
underside, so it only finds solid when reach * tan(slope) > layer / 2: the
cutoff was ~17deg at 0.3 mm layers, ~22 at 0.4, ~31 at 0.6. The M4 front
upright had 223 tines at 0.2, 117 at 0.4, 53 at 0.6; buster upright none at
0.4. The reach now keeps the 0.2 layer's ratio (tineReach * max(1, layer/0.2)):
228 / 220 and 15 / 10. 0.2 and thinner are unchanged (sweep vs main: no mesh
changed). Found checking #210 's large nozzles: not the nozzle, the layer.
tests: a tilted block at 20 and 30deg gets as many tines at 0.3/0.4/0.6 as at 0.2.
2026-10-14 09:32:07 -05:00
matthewtrahan and Per Hjort
103f61f337
wall lines: review fixes -- a thick wedge reads its edges' underside, not just the centreline (0.45 mm into a tilted cylinder at 8 lines); a squat brim stays under half the row spacing; a session saved before Wall lines loads at 2
...
The wedge fix only runs for a blade thicker than the tuned 1.2 mm, so 2 lines is
unchanged (sweep vs main: no mesh changed). tests: cylinder and tube at X45 with
8-line wedges have nothing inside the part; the brim cap at 0.8 / 8 lines.
Co-authored-by: Per Hjort <35040883+HjortCreations@users.noreply.github.com >
2026-10-14 08:41:44 -05:00
matthewtrahan and Per Hjort
e40a63341f
nozzle: a Wall lines setting (2/4/6/8) -- thicker walls, wedges and braces for a tall part or a big printer; the tip and tines stay one line
...
From Per Hjort's nozzle profiles (#210 ), ported onto web/nozzle.js. Lines
multiply the structural widths (wall, wedge; a brace rib is never thinner than
the wall). The feet move out with the wall's face, and a neck into the tip that
would lean flatter than 55deg gets the height to stand at 55. 2 lines is the
tuned wall: sweep vs main unchanged (no mesh changed). A thicker wall fits fewer
places, so a tight part can lose a wall or two at 8 lines; the readout reports
them. A setting of the part (undoable, saved with a session), sent to the Worker
in opts.tunables.wallLines. The plugins build 2 lines.
Co-authored-by: Per Hjort <35040883+HjortCreations@users.noreply.github.com >
2026-10-14 08:41:44 -05:00
matthewtrahan
93c36f9561
plugins: gap review fixes -- Blender's float32 gap at 0.4 is in range, Onshape's gap bounds are the site's 0.1-0.4, Cura README
2026-10-14 07:54:03 -05:00
matthewtrahan
93bb5565f4
plugins: the Material option becomes the site's Support gap -- PLA's pad numbers, a 0.1-0.4 gap field in every dialog, Cura's Match Cura and Onshape's PLA/PETG/Custom go
2026-10-14 07:54:03 -05:00
matthewtrahan
b625eedd30
web: drop the PLA/PETG Material dropdown -- PLA's numbers stay the defaults, PETG's wider gap is the Support gap field's
...
The dropdown only moved the Support gap (0.2 -> 0.3, already its own field), the
Sure hold pad's thickness and tack, and the grams density (2%). Default output is
unchanged (sweep vs main: no differences). Startup now reads the Support gap field
into PROP.gap, which the material's reset used to cover. The plugins keep their
material option until their own PR; options.test.js lists it as plugin-only.
2026-10-13 19:14:53 -05:00
matthewtrahan
1f605c5a5c
downhill: find the top edge by following the climb, not one straight step
...
On a curved slope (a spiral ramp's underside) a straight 1 mm step uphill cuts
across the curve and leaves through the side edges, which then read as the top:
the ramp got one line from a misplaced seed. The climb is now followed in
quarter steps, re-aimed each time, and seeds are ordered across the top edge's
own downhill direction. The ramp gets an arc down each edge.
Sweep vs main (NOCHECK): 805/864 byte-identical, 0 blocking.
2026-10-13 16:24:35 -05:00
matthewtrahan
0339d9b596
downhill: review fixes -- re-pointed tine captures, try/finally, hashed join check
...
- raceAgainst re-points each kept wall's `caps` into the rebuilt capture: after
a raster swap the downhill race sliced the wrong tines (and the judge read
the wrong lowest tine)
- judgeDownhill restores globalThis.__TINECAP in a finally, and classifies once
- downhillLines finds an earlier line through a plan hash, not a scan of every
station
- the downhill race no longer carries rasterRegions over (it overcounted)
- tines.js says the capture seam is no longer test-only; a test renamed to what
it checks
Sweep vs main (NOCHECK): 803/864 byte-identical, 0 blocking.
2026-10-13 16:24:35 -05:00
matthewtrahan
cb460b96ce
spec: downhill walls -- seeds, stops, the race and the whole-build check ( #25 )
2026-10-13 16:24:35 -05:00
matthewtrahan
f93f29ed51
engine: downhill walls -- walls that run down an overhang and follow it to the plate ( #25 )
...
prop/downhill.js traces the overhang's steepest descent from each end of its
top edge and every 12 mm between. On a flat slope that is a straight line; on a
curved one the wall bends with it. Across a wall that runs downhill the surface
is level, so its top settles and clears the part on both flanks -- walls laid
along a slope's edges (#125 's curve paths) failed the clearance probe at every
station on the #25 part.
Auto races them per region against the other placements (raceAgainst, 3%
margin, at most 1.5x the weight). A downhill wall with any vertex inside the
part is dropped first: a flank flush with a part face at the region's top edge
read as clear (lbracket Y20 on the plugin path). Then fins/downhill.js builds
the part without them too and keeps the downhill build only if it holds at
least as much overhang, grips no higher and keeps every wedge -- wedges, short
walls and the fill pass build around whatever the race kept, and the race
can't see them.
buildProps leaves them off unless asked (opts.downhill: true); buildFins in
Auto/Full turns them on and judges; downhill: false builds without them.
covered. Sweep vs main (NOCHECK): 802/864 byte-identical, 36 cases gain >2 pts
coverage, 0 blocking; arch/cone/needle X40 goldens move (75/92/92 -> 100%).
The torus X45 squat-pass test builds with downhill: false -- downhill walls
take that ledge over and hold more of it (70 -> 75%).
2026-10-13 16:24:35 -05:00
matthewtrahan
8f0784b81b
prop: the raster race takes any second placement (raceAgainst)
...
withRaster's per-region race, keepers and tine-capture bookkeeping now run any
second pass: `tag` names it on the kept walls and in the swap count, `maxGrow`
caps how much heavier its walls may be, `rel` sets the share of held area it
must add. withRaster calls it with the old values, so the raster race builds
byte-identical output.
2026-10-13 16:24:35 -05:00
matthewtrahan
fd7952b009
probe: --opts merges extra buildFins options into each build
...
e.g. --opts '{"downhill":false}' scores a build with an engine switch flipped,
against the same ruler as the default build.
2026-10-13 16:24:35 -05:00
matthewtrahan
7aaf57796e
onshape: the user guide lists the Nozzle setting (steps + Print Settings table), PDF rebuilt
2026-10-13 14:11:45 -05:00
matthewtrahan
86b9963aaa
onshape: nozzle comment names the site's field, not a file main doesn't have yet
2026-10-13 14:11:45 -05:00
matthewtrahan
a5f5b07322
onshape: a Nozzle print setting -- tines and contact tips scale with it, walls and fins only thicken, as on the site; a feature saved before it builds at 0.4
2026-10-13 14:11:45 -05:00
matthewtrahan
736471b7fd
prusa: nozzle review fixes -- no reference to site files main doesn't have yet, README re-wrapped, plugin version 0.3.0 for the new dialog field
2026-10-13 14:11:13 -05:00
matthewtrahan
ddaca3e461
prusa: a Nozzle field on Add a Fin -- the tine scales with it both ways and the fin only thickens, as on the site; a 0.6 or 0.8 nozzle dropped the 0.5 tine
2026-10-13 14:11:13 -05:00
matthewtrahan
50e21cb03e
layer height: max 0.6 mm on the site and in the plugin dialogs -- 0.6 and 0.8 nozzle prints slice thicker than the old 0.4 cap, and the tine must be one of their layers
2026-10-13 14:09:43 -05:00
matthewtrahan
854da684f8
nozzle: a printer setting -- remembered across visits and re-applied on reload, so a soft reload no longer built 0.4 under a 0.8 dropdown; brace rib growth scales too
...
A saved session leaves the nozzle alone, like the build volume. SWAY.thPerMm
grows with the rib widths so a brace reaches thMax at the same height. FIN-SPEC
notes what stays unscaled (tine reach) and the 0.2 comb losing a tine or two.
2026-10-13 13:21:17 -05:00
matthewtrahan
b99d60ce2b
nozzle: a Nozzle size field that scales the support widths -- every width was a 0.4 nozzle's, so a 0.6 or 0.8 dropped the tines or laid the wall as one line
...
Tines and the contact tip scale by nozzle/0.4 both ways; the wall, wedge and
sway brace rib only grow (a finer nozzle doesn't stiffen a free-standing wall).
Heights stay on the layer height. 0.4 is unchanged (sweep vs main: identical).
The Worker gets it in opts.tunables.nozzle; the plugins stay at 0.4 for now.
2026-10-13 13:21:17 -05:00
matthewtrahan
faed619228
ties: a strut's end runs TIE.into (0.35) past each wall's centreline, not th/2 -- square on it ended flush on the wall's far face, at a slant a corner showed 0.2 mm through it unchecked; the part clearance now runs out to the ends. The last strut's top at its far end stays under the limit (it rose 0.71 mm past it when the limit fell on a whole climb; review). tests: the top under three limits, ends inside the wall faces; fill: the hexprism at X30 gets ties, none inside the part or into a tip taper
2026-10-08 18:24:49 -05:00
matthewtrahan
435a768c03
FIN-SPEC: joined walls and ties -- 55deg struts 1.2 mm deep, walls >= 8 mm tall 2-12 mm apart, two ties each, up to 70% of the lower wall
2026-10-08 18:24:49 -05:00
matthewtrahan
2d1224548a
fill: walls may join, and the fill walls are tied into a frame -- (1) pass 2's last try lets a wall cross or meet another FILL wall (never Auto's or the part): a cleat sole at 45deg needs walls down its slope, across the one along it, and each was refused for touching it. (2) after the ribs, each fill wall over TIE.minHeight standing on the plate is tied to its nearest neighbours, closest pairs first, two each (prop/ties.js), clear of every support but its own two walls. The collider tags fill walls per wall so each check skips only what it may meet. Full vs Full must 78.2 -> 78.4% (ssd mounts +23 / +13, standing figure X30 +14, 32 mm mumble +19; one loss, 32 mm gnome up -4.2 on 120 mm2); walls 2456 -> 2559, grams 1400.7 -> 1421.6, build s 235 -> 343 (joined walls: more tries; ties ~0.8 s on the figure). Auto sweep byte-identical. tests: a joined wall is checked against Auto's walls only
2026-10-08 18:24:49 -05:00
matthewtrahan