Commit Graph
3250 Commits
Author SHA1 Message Date
CamWightandClaude Opus 5.5 669aaa7744 Drawings: a dimension finds its circle again after the edge is split (#4892) (#11869)
* Labelled measurements on a view's snapshot (#5866)

measure: linear, aligned, angle from model points, radius from an edge, with
units and decimals for the labels. No new maths: drawing_measure's rules, and
a dimension's radial rule through one Lookup::whole_arc. Labels go through one
function, reading_label, which dimensions' text now uses too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Pieces on one circle are that circle, and one piece is that edge (#4892)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-04 06:40:10 -05:00
CamWightandClaude Opus 5.5 a3375c58e3 Render: a triangle tests only the samples of each row it can touch (#5624, W17b) (#11042)
The raster walked a triangle's whole box, testing every sample, and the pixel budget counted that box.
A plain plate's few long triangles have boxes that are mostly misses, so a view at 2:1 counted 18 tests
a sample. Now each row of the box tests only the span the triangle can touch there, and the budget
counts those spans (an empty row costs one unit, as the walk does).

row_span works the span out from each edge's weight along the row, allowing for the test's own
tolerance and the rounding of its arithmetic, so it is never short; sample_weights is the raster's test
pulled out unchanged. Pictures are the same to the byte. A product image shares the raster and the
count, so its 128 M admits more; the tests built on its old count hold more triangles.

With the misses gone nearly every counted test paints, and a painted surface costs about 240 ns a
test on one thread, so a picture's limit falls from 400 M to 250 M to stay near a minute at worst.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-04 06:12:20 -05:00
CamWightandClaude Opus 5.5 a301359652 Dimensions on the sheet: lines and arrowheads in paper millimetres (#4896) (#11867)
Each dimension of a view on the sheet is one group, ObjectKind::Dimension,
after the views: extension and dimension lines and filled arrowheads
(Terminator::place, #9466), laid out by dim_layout_sized in paper millimetres, so it is the same size at any
view scale. All seven kinds draw; a chain is one line of frames at one offset
with each extension line once, an angle sits where its lines meet on the view,
and a cylinder's line runs across its silhouettes at the end on its offset's
side.

drawing.scene and drawing.export list each dimension: its status, whether it
is drawn, and why not. A lost one, or one on a view that isn't drawn, has no
lines and leaves the sheet and the file written. The provider keeps each
dimension's group by its record, world state, view drawing and units: a moved
view moves them unread, and an edit reads that dimension alone, in one short
step after the views. Text joins with B1.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-04 04:53:24 -05:00
CamWightandClaude Opus 5.5 fd4b7a7bc0 Seg2: one rule for an arc under a shape-keeping map (#4816) (#7316)
`Seg2::mapped(f, arc)` is a segment under a map of the plane that keeps shapes. Points go
through `f` itself (an arc's centre, every control point). `ArcMap { angle, mirrored, scale }`
says how the map moves an arc: it turns it by `angle`, after a mirror across the x axis when
`mirrored` (start `angle - start`, sweep reversed), and scales its radius by `scale`.
`ArcMap::turn(angle, mirrored)` keeps lengths; `ArcMap::rigid(f)` measures a map of points that
keeps lengths, from where it puts the origin and the unit axes, with a scale of exactly 1.

`Seg2::similar` and the three arc maps written out by hand use it:
- a sheet-metal region onto a panel's flat (`region_to_flat`, which measured its map by hand and
  now asks `ArcMap::rigid`);
- the mirrored flat (`flip_x_loop`);
- a sketch region mirrored in its x axis (`mirror_y`).

The points go through the same functions as before, so every result is the same, bit for bit.
This comes from the architecture review of the drawing work (2026-10-25): four places wrote the
arc rule out, each its own way, and one measured its map by hand.

`map_controls` is private to geom now: on its own it hands an arc back unmapped, which is what the
three hand-written arms were there to avoid.

Tests: every kind of segment under a mirrored map with a turn, a scale and a shift, point by point,
its length scaled and an arc's sweep reversed; the three old formulas, bit for bit (the sketch
mirror keeps a +0.0 start's -0.0); a rigid map's turn and mirror measured whatever its shift, and
the identity; the sheet-metal and mirror tests unchanged.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 14:59:59 -05:00
CamWightandClaude Opus 5.5 664c9812d2 Paper scene: a sheet's pieces on paper are capped, refused by name past 4,194,304 (#4890 A1) (#11709)
A run's glyphs count once in a scene's records, for a writer that embeds its font, but the
walk onto a writer's grid copies every placed glyph's whole outline, so a sheet dense with
text could reach any size. The walk now counts the pieces it places and refuses a scene past
MAX_GRID_PIECES (1 << 22), saying to move some text to another sheet. place() is public so
a writer that draws text as outlines (PDF) counts with the same rule.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 13:20:44 -05:00
CamWightandClaude Opus 5.5 08cd8d6442 Drawings: title blocks on sheets, and a drawing can start from or be restyled by a template (#3875) (#7208)
* Drawing sheets: a copy of a title block on the drawing border, saved with the sheet (#3875)

`Sheet.title_block` holds a copy of its definition, as `border` does. Each sheet's check lays it
out on the drawing border, so a sheet without a border, a block wider or taller than the border,
or a change of paper that leaves no room is refused. `drawing.sheet.add` and `drawing.sheet.set`
take `title_block` (a definition, or null for none), a new sheet takes the block of the sheet it
follows, and `drawing.inspect` lists it. A sheet without a block saves no `title_block`, so an
older drawing reads as having none (pending format file `sheet-title-block.rs`).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: the built-in title block, on every new drawing (#3875)

`title_block::default_block()` is the 180 x 54 mm block of twelve cells approved on #3875, with
the legal owner read from the drawing's `organization` and the general tolerances as text to
change. `drawing.create` starts its sheet with it. The geometry tests read the built-in block
instead of their own copy of it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: a title block's lines on paper, drawn after the views (#3875)

`TitleBlockLayout::paper` draws the outline, then every cell edge inside it, each once:
neighbours' shared edges, and edges that meet along one line, are one stroke. The sheet scene
paints the block after the views, and counts its records with the border's, so no view takes
its room. The view tests' drawing leaves the block off, so their scene is still the border and
the views.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: the title block's projection symbol, in the drawing's projection (#3875)

ISO 5456-2 (Annex A): a truncated cone and its end view, h across the cone's
small end from table A.1's sizes (3.5 to 20 mm), the largest that fits the
cell, lines h/10 wide. The end view is on the right in first angle and on
the left in third. The block only says a cell holds the symbol; the angle
comes from drawing.projection, and the sheet scene redraws when it changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: a title block's fields, and text typed over its cells (#3875)

A title block cell reads its fields from the component the sheet documents (its
name, part properties, material and mass from the bill of materials), the
drawing, the sheet (number and count among the counted sheets) and the newest
revision row; a field with no value shows "-". A key its source doesn't have is
refused when a block is given or read. Sheets save text typed over their cells by
cell name; drawing.inspect lists each text cell with its value beside an override.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates as data: a drawing with no model, and ISO built-ins (#3875)

A template is a drawing with no sources, views, dimensions, balloons, revision
rows or documented components, so it is saved and read as a drawing is. A new
drawing is a copy with ids of its own. The built-ins, ISO A4 to A0 for parts and
for assemblies, are made in code when asked for and differ only in their general
tolerances.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates: drawing.create from a template, and the preferred ones (#3875)

drawing.create takes template? (default ISO A3, for a part or, for a design with
components, an assembly) and projection? over the template's. The new
drawing.template.list lists the built-ins and marks the one a new drawing starts
from. Preferences add a template for part drawings and one for assembly drawings;
the app fills in the preferred template and projection before drawing.create
runs, so a template named keeps its own projection, instead of changing the
projection after the drawing is made. A saved template this version doesn't know
reads as the default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates: apply one to the drawing open, moving nothing (#3875)

drawing.template.apply {template, dry_run?} gives each sheet the paper, drawing
border and title block of the template's sheet in its place (its last for the
sheets past its own). Text typed over a cell stays while the new block has a
text cell of that name; the cells whose text is dropped are listed. Nothing
moves: the views a new paper leaves outside the drawing space are listed. A dry
run answers the same and changes nothing; applying is one undo step, none when
nothing changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: views a title block covers are listed, never moved (#3875)

A view lying over its sheet's title block is listed with the ones the drawing
space leaves out, after a change of paper or block (drawing.sheet.set) and when
a template restyles a sheet. Where a block sits is one rule, the one a sheet is
checked by (Sheet::title_block_outline), and the overlap of two boxes is one
rule in Aabb2::overlaps, which the title block's own cell check now uses too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates: save one as a file, and start or restyle drawings from it (#3875)

drawing.template.save {path} writes the drawing open as a template: its format,
and nothing it draws or documents. drawing.create and drawing.template.apply take
a .json template file as well as a built-in's name. A file is read as a drawing
is, then held to the template rules; files are read and written through the same
file check and atomic write as the sheet export. A drawing made from a file keeps
its own copy, so a later save changes no drawing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: the built-in title block is 180 x 49 mm (#3875)

The top row is 15 mm, Sheet is 30 mm wide, lined up with Mass, and Date of
issue is 32 mm. The top row now holds one line of the legal owner, two of the
general tolerances and the 3.5 mm projection symbol.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: one operation sets a sheet's title block, and drawing.sheet.set names the typed text it drops (#3875)

A sheet's block was set in two places: drawing.sheet.set kept the typed
text of the text cells the new block still has, and drawing.template.apply
did the same and listed what it dropped. Sheet::set_title_block is now the
one rule both use. drawing.sheet.set's reply gains `lost`, the cells whose
typed text a new block drops, as drawing.template.apply's reply has it, and
its help says that a view over the title block is listed too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates: the built-ins and a new drawing start with the border only; the built-in title block is added by name (#3875)

A template now holds only the ISO border by default. A title block,
tables, references and text are added to a drawing, and a template saved
from it keeps them.

- The ten built-ins (iso-a4-part ... iso-a0-assembly) become five,
  iso-a4 to iso-a0: one sheet with the ISO border. Without the block, the
  part and assembly versions were the same.
- drawing.create with no template starts from iso-a3. The two
  preferences for parts and assemblies become one, drawing_template,
  "Template for new drawings".
- drawing.sheet.add and drawing.sheet.set take title_block: "built-in".
  The built-in block's general tolerances are a part's or, for a design
  with components, an assembly's, through Sheet::set_title_block.
- An old name such as iso-a3-part is refused with the list of names;
  drawings haven't shipped.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing title blocks: a drawing with no number of its own is numbered by the component its sheet documents (#3875)

The identification number cell read the drawing's part number property,
and showed "-" for a drawing that has none, which is every new one. Now
such a drawing shows the name of the component the sheet documents, so a
new drawing of a part is numbered by the part until it is given a number.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing templates: a template saved from a drawing keeps typed text only over plain text cells (#3875)

A template saved from a drawing kept all the text typed over its title
block's cells, including text over a cell that shows a field, such as
the material or the title, which belongs to the design the drawing
documents. Now it keeps the text typed over cells of plain text, such as
the general tolerances, and leaves the rest behind.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing title blocks and tables: a field without a value shows nothing (#3875)

A title block cell or a table row whose field has no value (a property
nobody gave, a row field its row doesn't have, an uncounted sheet's
number) now shows nothing instead of "-".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing title blocks: the built-in block's general tolerances start empty (#3875)

The general tolerances cell of the built-in title block now starts empty for
parts and assemblies alike, rather than ISO 2768-m in the block and a part's
or an assembly's text added by drawing.sheet.set. A shop types its own, or
its template carries them, and an empty cell shows nothing.

- template::PART_TOLERANCES, ASSEMBLY_TOLERANCES, for_assembly and
  built_in_block go: "built-in" is title_block::default_block.
- drawing.template.list's help no longer speaks of part and assembly
  built-ins.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 11:01:13 -05:00
CamWightandClaude Opus 5.5 c5dab0b9a8 Paper scene: ellipses, hyperbolas and higher Béziers as cubics on the grid (#4890 A3b) (#9846)
A conic of weight other than 1 and a Bézier above degree 3 become cubic Béziers within
1e-4 mm of the curve (`paper::cubic_pieces`), so the walk takes them and every writer
draws them with the cubics it already writes. Each piece's cubic has the curve's ends
and end tangents; its distance is bounded from the Bernstein form, not sampled: for a
Bézier, by the control points of its difference with the cubic raised to its degree;
for a conic, by Q·D − N over the denominator's least value. A piece over the error is
halved exactly and tried again, up to 4096 a curve and 1,000,000 a scene; past that the
curve is refused, never drawn coarser. The walk counts the cubics it made. Circular arcs
are left to the shared `paper::cubics` helper.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 09:27:36 -05:00
CamWightandClaude Opus 5.5 cca77b176f Drawings: a shaded view is drawn on the sheet as a picture under its lines (#5624) (#10926)
* A drawing's list of one kind of record: ById, with balloons first (#5426)

Each kind on a sheet writes out the same add, find by halves, cap, order and check loop. ById holds
those, so a kind's own file holds only its own rules, and the file's bytes don't change: it saves as
the plain list. Views and dimensions stay on their own lists for now, since other work is on them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* A drawing keeps an image's file once, however many places it sits (#5426 W38e-1)

images holds the files and sheet_images each place one sits, so a logo on every sheet costs what one
does. drawing.image.add reads the file (bytes and pixels checked from its header before anything
decodes it), reuses a file with the same bytes, and saves the bytes without the path. A file goes
with the last thing that uses it, which Drawing::images_in_use alone says, and reading a drawing
refuses a file nothing uses or a record its file disagrees with. A sheet's list and delete take the
placements on it. Deleting one placement waits for selection.delete to take a drawing's objects. Its
pixels on paper are W38e-2.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sheet images: an image file's pixels on any paper grid (#5425, W38e-2)

A PNG or JPEG decoded once, then stretched over its paper rectangle onto the pixel grid a writer
asks for: a box filter in linear light with alpha premultiplied, so a shrunk logo keeps its edges
and gains no dark fringe. The size is read from the header and refused over 64 megapixels before
anything is decoded. An EXIF orientation is applied; colour profiles are not read.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sheet images: a placement in the sheet's scene and its SVG (#5425, W38e-2)

Each image placed on a sheet is a group of the new kind `sheet_image`, between the drawing border
and the views, so a view's lines draw over a photo. `drawing_pictures` holds both that group and
the provider a writer asks for its pixels, so the two read the same placement.

The provider holds what one sheet uses, decodes a file when it is first asked, once however many
placements show it, and refuses a sheet whose files hold more than 64 million pixels, from their
headers, before anything decodes. `drawing.export` passes it to the SVG writer in place of
`NoPictures`, and a file that won't decode refuses its sheet by name.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: pin the product image's output before it is drawn in tiles (#5624, W17b S1)

The decoded rows of five product renders (orthographic on a transparent background, perspective on a
solid one, a PBR surface, two glass layers sorted back to front, triangles crossing every edge of the
picture), taken on this tree. The next change reshapes ImageTask into a prepare stage, tiles and a PNG
sink; they must not move. The PNG bytes are not pinned: their compression depends on the deflate
backend the build unifies (flate2's zlib-rs when the UI crate is built alongside, miniz_oxide
without), not on this code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: pictures drawn tile by tile into any pixel window (#5624, W17b S1)

The CPU product renderer, reshaped so drawings can use it: one rasteriser (`Raster`) prepares a
scene once (appearances, projected triangles, glass sorted back to front), then draws each tile of
the picture on a canvas of its own. A sample's position comes from its whole-picture index, so tiles
put side by side are the picture drawn whole, to the byte.

- `ImageTask` (render.image) is that rasteriser over one whole-picture tile from the product camera,
  then its PNG: the same output, pinned by the previous commit's rows.
- `PictureTask` draws a `PictureRequest`: a `Projection` (the product camera, or an `OrthoFrame`,
  W17a's view frame as plain data: a `Plane` basis, a scale and a paper shift, on a
  `geom::pixel_grid::PixelGrid`), any list of `PixelWindow` tiles, `Supersample::One` (pixel centres)
  or `Two` (i + 1/4 and i + 3/4), a `BodyRole` per body (`DepthOnly` hides what lies behind it and
  draws nothing itself) and an optional background (none: transparent). Each tile comes back as
  straight-alpha sRGB RGBA rows, blended in linear light.
- The orthographic map is the frame's paper map, onto the grid, so lines drawn from the same frame
  register with the pixels (to 1e-9 mm in the test).
- Limits: 65,536 px a side, no total cap; each tile under the resolver's `RenderLimits`, as the
  product image is; at most 65,536 tiles; the prepared lists as a product image's; pixel tests up to
  8 per sample of the picture (128 M for a product image, as before), counted at admission by the
  one projection the raster uses, as main counts a product image's.

Not here: the drawing light, stored per-tile bins (S2, measured), the paper scene's picture item
(S3), 4x sampling, and `raster::render`, the snapshot rasteriser, which stays as it is.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: time the tile renderer (#5624, W17b S2)

An ignored timing test, as `simplify_timing` is: a 65,548-triangle front view drawn as the
product image, as one 300 dpi picture, and in 64-row strips at 300 and 1200 dpi, each the median
of three runs. It is the measure S2's speed-ups are judged by.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: a tile passes over a triangle outside it before painting it (#5624, W17b S2)

Each tile placed and painted every triangle of the picture, then dropped those outside it: the
appearance lookup, the PBR normals and the colour decode ran for every triangle in every tile. A
triangle's work is now three steps, `corners`, `look` and `place`. Preparation runs them in
today's order, so its checks and errors are the same. A tile places a triangle first and paints
only those inside it. What is drawn doesn't change: the pin and the tile tests stay as they were.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: one light per picture (#5624, W17b S2)

`raster::shade` normalised its key light and highlight directions again for every sample,
although both depend only on the view. `raster::Light` works them out once per view and shades
with them. The tile renderer and the snapshot rasteriser (`raster::render`) both use it, so the
product light is written in one place. The arithmetic is the same, so every pixel is too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: sRGB codes from tables (#5624, W17b S2)

Each shaded sample was stored as sRGB through a power (three per sample), and each output pixel
was averaged from its samples through twelve more and written back through three: the powers
were most of a sample's arithmetic. Decoding is now a 256-entry table of `geom::srgb`'s own
values (#7162), which also ends this file's own copy of the formula. Encoding counts the code
starts (the linear values halfway, in sRGB, between two codes) at or below the value. Within a
billionth of a start, where the formula's rounding could fall either way, `srgb::encode8`
decides, so every code is the one it gives. A test checks both against `geom::srgb` at every
start, the numbers around it, the band's edges and a million values in and outside 0 to 1.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: a tile's passes run in loops (#5624, W17b S2)

The raster went back through its step loop for every sample: the cancel check, the stage match,
the tile's window and its canvas bounds, once per sample cleared, tested or averaged. Clearing,
the opaque and glass passes and the downsample now each loop on their own up to the step's budget,
counting a unit as before; preparation and the glass sort still take one unit a call. A step
still checks for cancellation before each stage's run and still does at most 16384 units, so the
host's slices and the progress counts are the same.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: tests for a picture's light, glass order and test budget (#5624, W17b S2)

Three planted faults went through the image tests; each now has one
that fails:
- an orthographic picture lit from another direction than its eye: a
  face square-on to the view is the colour the product image gives it;
- glass placed by a triangle's first corner, not its middle: a tilted
  square whose first corners lie in front of the near one, its middles
  behind;
- a picture held to the product image's pixel-test budget: a face
  across most of an A0 sheet at 1200 dpi is prepared.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing sheets: a shaded view's picture on paper (#5624, W17 V9b)

A saved view in a shaded style draws on its sheet and in drawing.export:
draw puts one picture per entry of the view's parts under its lines, and
ViewPictures renders each on the writer's grid through the frame that
placed the lines. SheetPictures answers for a sheet's images and views.
The styles with lines stay refused until W06's exact lines.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Render: a picture's count of its tiles and its fixed limit, as #6483 has them (#5624)

A triangle counts only where it meets the tiles asked for, and a picture has a fixed 400 M pixel tests,
as #6483 at bf93e890e. V9b's stack holds an older copy of that slice; this drops away when drawings
takes main. PictureTask says what admission counted, for the view's running sum.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing views: a shaded view's pictures add up to one limit (#5624, W17 V9b)

A writer asks for a picture a band at a time, and each request is limited on its own, so ViewPictures
keeps what each view's requests have counted and refuses the next one past the limit, by name, before it
paints. That makes 400 M a bound on a whole view.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 08:21:51 -05:00
CamWightandClaude Opus 5.5 1bd0040d25 Drawing export: SVG and PNG cut a picture to a detail view's boundary (#4890) (#10969)
* Paper scene: clips of lines and arcs, cut once before every writer (#4890 A8a)

Scene::prepare_clipped(clips) cuts each group of an object to that
object's Clip, a rectangle or a circle (a detail view's), and keeps what
lies inside; prepare() is prepare_clipped with none. Prepared holds a
Cow: a scene without clips is borrowed as it is, and Group and Scene are
unchanged, so today's scenes are the same bytes.

- Lines and arcs are cut where they cross the edge; a part is kept when
  its middle is inside, the edge included. Parts no longer than EPS go.
- Kept parts that follow one another stay one open stroke. A stroke
  wholly inside is kept whole, closed if it was; a closed one cut open
  rejoins across its seam.
- A dashed stroke keeps its pattern: each part's phase is where it
  starts along the stroke. Across a closed one's seam the pattern runs
  on only when its length is whole patterns, so the parts rejoin only
  then.
- Checked: one clip an object, a rectangle from its lower-left to its
  upper-right corner or a circle with a radius above 0, within 1e5 mm;
  a clipped object's items strokes of lines and arcs. What the cuts
  made is validated again.

SVG and PNG draw the cut scene through Prepared as they are.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: clips of every curve, and fills kept or dropped whole (#4890 A8b)

A clipped object's strokes may now hold any curve, not only lines and
arcs, and its fills are checked against the clip.

- A line or an arc is still cut where its equation says. Any other
  curve is halved while its control points lie either side of the edge
  (widened by EPS), until they span at most CUT_MM (1e-9 mm), and is cut
  there. A curve that needs more than MAX_HALVINGS halvings (one that
  runs this close along the edge) is refused as not supported yet.
  Each part is the same kind of curve as the whole.
- Seg2::part gives the piece between two parameters, starting and
  ending exactly where split_at puts those points; a conic's tail is
  re-normalised, so its second parameter is carried into it.
  Seg2::MIN_SPLIT names split_at's distance from the ends.
- A line or a circle within EPS of touching a circle only touches it:
  one cut there, never a sliver of chord.
- A fill wholly inside its clip is kept and one wholly outside is
  dropped. One that crosses the edge, or lies round the whole clip, is
  refused as not supported yet.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures in a clipped object, cut by the SVG and sheet PNG writers (#4890 A8c)

A picture wholly outside its object's clip is dropped when the scene is prepared. One crossing the clip's edge reaches its writer with the clip, on the grid item, since pixels can't be cut as geometry: SVG cuts it with a clipPath, the sheet PNG by whether each pixel's middle is inside. Both draw only the pixels that meet the clip's box. Text in a clipped object stays refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 07:58:27 -05:00
CamWightandClaude Opus 5.5 49a3ba1c03 Paper scene: dashed lines, in SVG and PNG (#4890 A5) (#7430)
A stroke can carry a pattern, `Dashes { lengths_mm, phase_mm }`: dash and
gap lengths in turn, in paper millimetres, and how far into the pattern
the stroke starts. Without one a stroke is solid, and a solid stroke
reads and writes as before.

- Checks: 2 to 12 lengths, an even count, each above 0 and at most
  1000 mm; the phase from 0 up to the pattern's length; at most a
  million dashes along the stroke, counted from its pieces' own lengths.
- The walk puts the pattern on the writer's grid: a dash or gap under
  half a step is one step, counted, as widths are.
- SVG: stroke-dasharray, and stroke-dashoffset when the phase isn't 0.
- PNG: tiny-skia's dashing on the flattened line; its points count
  against the sheet's, a line in a gap draws nothing, and one tiny-skia
  won't cut is refused rather than drawn solid.
- The paper fuzz cases carry a dashed arc.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 06:52:45 -05:00
CamWightandClaude Opus 5.5 d78895de4d Where a drawing stands: drawing.status in one read (#4894 WE2) (#11352)
drawing.status says each view's status (current, out of date, placeholder or
failed, with why) from what the sheet page last drew, and each dimension's in the
words inspect uses, reading each body once for the call. Nothing is drawn.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 04:41:33 -05:00
CamWightandClaude Opus 5.5 112e0a3d9c Paper text in SVG: a run's glyph outlines on the writer's grid, filled nonzero (#4890 B2a) (#9708)
The walk puts a text run on the grid as its glyphs' outlines, placed by the run
(`TextRun::outlines`), each loop walked as a fill's: `GridItem::Text`, with the run's
index among its group's items, so a writer that sets real text (PDF) finds it in paint
order. SVG writes one path a run, filled black with its default nonzero rule, so text
looks the same without the font and never reflows. The sheet PNG refuses text by name
until B2b, and its refusal test says so.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 02:09:25 -05:00
CamWightandClaude Opus 5.5 eb0b427a36 Drawings WE3-WE5: commands report where each object sits on a sheet, and select and delete views, dimensions and balloons (#4894) (#11047)
* Where a sheet's objects sit: bounds beside the scene (#4894 WE3)

The sheet scene now says where its border and each view sit on paper: the box
round what each draws, and a view's boundary, so one that can't be drawn can
still be picked. It is worked out once per assembly, beside the scene rather
than in it, so no writer sees it and the page needn't inspect each view.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Selected drawing objects: a view, dimension or balloon in the selection (#4894 WE4)

select.set takes {type: drawing, drawing, object} for the drawing open. It adds no
undo step and changes nothing saved, and a selected object leaves the selection
when an undo takes it or its drawing closes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Delete on a sheet: views, dimensions and balloons in one step (#4894 WE5)

selection.delete now works with a drawing open: a view goes with its dimensions
and balloons as one undo step, and one with views projected from it is refused,
naming them, unless they are selected too or contents: delete takes them. A sheet's
delete and this one share the record's one removal rule.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a view's box holds what it draws, and what's chosen goes in any order (#4894 WE3, WE5)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 00:59:41 -05:00
CamWightandClaude Opus 5.5 ba69e5488c Drawings WE1: Inspect Drawing lists each sheet's views, with their names, dimensions and balloons (#4894) (#11350)
* Where a sheet's objects sit: bounds beside the scene (#4894 WE3)

The sheet scene now says where its border and each view sit on paper: the box
round what each draws, and a view's boundary, so one that can't be drawn can
still be picked. It is worked out once per assembly, beside the scene rather
than in it, so no writer sees it and the page needn't inspect each view.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Selected drawing objects: a view, dimension or balloon in the selection (#4894 WE4)

select.set takes {type: drawing, drawing, object} for the drawing open. It adds no
undo step and changes nothing saved, and a selected object leaves the selection
when an undo takes it or its drawing closes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Delete on a sheet: views, dimensions and balloons in one step (#4894 WE5)

selection.delete now works with a drawing open: a view goes with its dimensions
and balloons as one undo step, and one with views projected from it is refused,
naming them, unless they are selected too or contents: delete takes them. A sheet's
delete and this one share the record's one removal rule.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a view's box holds what it draws, and what's chosen goes in any order (#4894 WE3, WE5)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* What's on each sheet: drawing.inspect lists its views (#4894 WE1)

Each sheet in drawing.inspect now lists its views: each base view by id, followed
by those projected from it, with what it is called and its dimensions and balloons.
It is one pass over the record with no model read, so the Browser can read it once
per revision.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 00:59:36 -05:00
CamWightandClaude Opus 5.5 9308095744 Paper text in the browser: Japanese through the font the web app fetches (#4890 F6) (#10516)
The browser has no font folders, so `font_dirs` is empty there. The first
kana or Han a text's font lacks asks for the font the web app already
fetches on demand (BIZ UDPGothic); no new file. The text is pending until
it comes, then sets whatever it has a glyph for. If the fetch fails, or
the font lacks a character, that character is refused by name, and the
refusal says the web covers Japanese for now.

The web app starts the fetch through its existing loader, once, when a
text waits for the font, and hands the font over when it arrives.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-02 21:59:34 -05:00
CamWightandClaude Opus 5.5 5bf3940c04 Sheet tables on paper: a table's lines (#3874, #4890 W10.7, D3a) (#7745)
TableLayout::paper gives a placed table's lines as one paper group of
the kind it is (parts_list or revision_table), so both sheet tables reach
paper through one function:

- the outline as one closed stroke, outline_line_mm wide;
- a stroke cell_line_mm wide across the whole table at each row boundary
  inside the outline, the heading's too, from the top whichever side the
  heading is on;
- a stroke down the whole table at each column boundary inside it.

Each line is drawn once, centred on its edge, on the layout's own
corners. The cells' text follows once the paper scene has text (D3b,
after #4813); hatch clearance waits for hatching.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-02 21:53:06 -05:00
CamWightandClaude Opus 5.5 2f483a00c0 Drawing text F4: a character missing from the chosen font is drawn with another installed font (#4890) (#10515)
* Drawings: the font a drawing's text is set in, by family (#4890 F1)

A drawing names the font family its text is set in (`font`, saved only when
set; a pending drawing-format file). `drawing.create` and `drawing.set` take
it, and `drawing.inspect` reports it, with `font_note` when this computer sets
the text in the built-in font instead ("Futura isn't on this computer: set in
Liberation Sans"). Drawings name fonts by family, so they travel: a font file
is refused.

Paper text finds the family among the installed fonts as sketch text does
(`paper_text_in` for given folders), admits each face once a process, and
sets a family that is missing or can't be set on paper in the built-in font.
Bold takes the family's bold face; the serif flag comes from the face's OS/2
family class. The fixed list of drawing fonts is gone.

New drawings made in the app take the interface font when it is a family.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper text: any face with outlines is set, and says how a PDF may carry it (#4890 F2)

A face is admitted for paper whatever its tables lack:
- a variable face is set at its default instance, which is what a PDF draws;
- its cap height is the one sketch text scales it by (OS/2, else the top of
  its `H`, else 70 % of the em);
- its PostScript name is what a PDF name can hold of its own, else of its
  family and style, else `Font-` and the start of its id;
- without a `post` table it has no italic angle, and a fixed pitch when every
  glyph advances alike;
- its stem width is measured across its `l`, else taken from its weight.
A bitmap or colour-only face, with no outlines to fill, is refused by name.

`Font.embed` says how a PDF may carry the face, from its embedding permissions:
a subset, the whole face (no subsetting), or not at all (restricted, bitmap
embedding only, or CFF2 outlines), when the PDF draws the runs' outlines with
their text beside them. Which characters a PDF name may hold is now one rule,
`is_postscript_char`, that the record's check and sketch share.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper text: a font's family, for readers that pick fonts by family (#4890 F2)

`Font.family` is the face's own family name: its typographic family (name
ID 16), else its family (ID 1), read from the face itself (in a collection,
not the first face), without control characters and at most 255 characters;
empty when it has neither. A DXF text style picks a font by it; `Font.id`
stays the identity, so two faces of one family are still two fonts
(@0xlillith on #4890). The name lookup PostScript names use is one function
now, shared by both.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fonts: one list of fallback fonts per script, in sketch (#4890 F3)

The interface's lists of fonts to try for a script move down to
`sketch::fonts::fallback`, so drawings can use the same ones. Each row is
a family and the files it's known to be in. The interface reads the files
in the order it always did, so what it finds is unchanged.

- `CRAFT_FONTS_DIR` joins `font_dirs`, as the interface already has it:
  `SOLIDCRAFT_FONT_DIRS`, else the pack, then the system's folders.
- `font_files` sorts the files within each folder, so a search doesn't
  depend on how the system lists a folder.
- The Windows folder search takes its names from the Japanese list
  instead of its own copy of them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper text: a character its font lacks falls back to a font that has it (#4890 F4)

A text on paper was refused for any character its font lacks. Now each
such character falls back to an installed font that has it: the script's
list in `sketch::fonts::fallback` first, Han by the drawing's language,
then the first face in sorted order with outlines and the character.

- Runs split by direction, then by font. Every font takes the same em
  size, on the line's baseline and pitch. A space, a digit or punctuation
  stays with the run it follows when that font has it; a mark stays with
  its letter. A font fallen back to is shaped, with the drawing's
  language. Its nearest style is used: bold or italic isn't refused.
- A line its own font has all of is set as it was.
- A character no font has is still refused by name.
- The last-resort search reads each file's character map alone, stays
  within `MAX_FONT_FILES`, and keeps its answers, misses too, for the
  process.
- `paper_text_now` is for the window's thread: the search runs on a thread
  of its own (`font-search`) and it refuses with `FontPending` until the
  answers are in. `paper_text` searches where it's called.
- `paper_fallback_notes` names what fell back, for a drawing's status,
  from the answers so far: it looks for no font.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-02 21:27:16 -05:00
Nathan Dane d5fd14135f Add the native Drawing workspace UI scaffold (#5349)
* docs: plan drawing workspace UI implementation

* assets: add drawing icon drafts for owner review

* feat(ui): add drawing workspace scaffold with shared ribbon and docked settings

Expose planned Drawing tools without introducing drawing documents, paper rendering or CAD behavior. Reuse existing source BOM, properties and flat-pattern services; retain placeholders for unapproved icon drafts.

Validation: full UI, kernel, icon, clippy, wasm and layout checks passed. Full cargo xtask ci reports three geometry failures that also reproduce on unchanged drawings base 3b982ca: counterbored chamfered-disc ring cut, robustness seed 36 and seed 71. No geometry source or golden fixtures changed.

* Preserve default icon palette in Drawing scaffold tests

* Preserve Drawing providers and complete scaffold translations
2026-11-01 21:45:05 -05:00
CamWightandClaude Opus 5.5 85be50df0d Reference parts: a view shows them by percent, not by a margin (#5866 V5c) (#10713)
reference_margin_mm becomes reference_shown_percent, 0 to 100. At 0 the
boundary is the model's box, as before, and no reference part is measured.
At 100 it holds the model and all the view's reference parts. Between, each
side moves on its own, so a part on one side grows only that side.

The model's middle now places a view, not the middle of what it shows, so a
view stays lined up with its projected views as its box grows. With nothing
grown the two are the same point and the picture is the same.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 18:36:53 -05:00
CamWightandClaude Opus 5.5 71edae80df Drawing views made by command: drawing.view.base and drawing.view.projected (#4816) (#9254)
* Drawing views made by command: drawing.view.base and drawing.view.projected (#4816)

drawing.view.base {sheet, view?, scale?, position_mm?, the view settings?} adds a base view of the
drawing's design to a sheet. Without a scale it takes the largest ISO 5455 preset at which it lies
inside the sheet's drawing space, centred where it goes, its margin included, as the snapshot fits
it; that scale is saved as the view's own (1:1 with nothing to measure).

drawing.view.projected {parent, side, place?} adds a view projected from a saved view, on its
parent's sheet, following its parent's scale and display settings. Without a place it is lined up
with its parent, its boundary 20 mm past the parent's (DEFAULT_GAP_MM, until a drawing style
gives it).

Each is one undo step through edit_drawing, refused with nothing changed when the record's rules
refuse it; the reply gives the view, its sheet, its scale, its boundary and, when it leaves the
sheet's drawing space, itself as outside (it isn't moved back in). Unknown fields are refused.

One rule, one place: whether a view leaves its drawing space is drawing_view::leaves, and the sheet
of an id is drawing_view::sheet, used by outside() and both commands.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing views made by command: a build fix, and each view one undo step, tested (#4816)

From the review of the view create commands:
- drawing.view.base measures the shown bodies (`measure` takes them, not the session), so it builds.
- The nothing-to-measure case is a design with one sketch and no bodies, as the other view tests
  make it: drawing.create refuses an empty design.
- The schema test gives drawing.view.base its own key list (it takes no snapshot-only keys).
- Every view a test makes is checked to be one undo step, not only the first: a second step made
  under the first (a nested edit) still let one undo take the view away.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 17:12:38 -05:00
CamWightandClaude Opus 5.5 93baf88e2d Paper scene: pictures in the sheet PNG and SVG (#5624, W17b S5) (#7591)
* Paper scene: one walk onto a writer's number grid, for every writer (#4890, C4)

The SVG writer kept its own number grid: each coordinate whole nanometres, every joint written once,
a closed path ending exactly on its first point, slivers merged into the next piece and a path's
last piece kept. PDF, PNG and the Drawing workspace's canvas need the same rules, so they move into
`geom::paper::grid`, and the SVG writer only formats what the walk gives it.

- `Grid`: `per_mm` steps a millimetre. `Grid::NANOMETRE` (six decimals of a millimetre) is SVG's.
  `Grid::new` takes 1 to 1e9; a power of two keeps every number exact as a binary float, which a
  writer of 32-bit numbers such as PDF can use.
- `Prepared::on_grid(grid) -> OnGrid`: each group's items as paths of `Piece`s (lines, arcs of at
  most a quarter turn, cubics, parabolas), y up as in the scene, a closed path's closing line left
  to the writer's close, and a report of widths snapped and pieces merged. Slivers are measured in
  the grid's own steps, so a coarser grid merges more.
- The SVG writer flips y on the whole numbers it's given and writes the same bytes as before.
  `SvgReport` is the grid's report.
- Ellipses, hyperbolas and Béziers above degree 3 are still refused (A3b), now by the walk.

DXF stays on its own number rule for now: it writes exact entities (ELLIPSE, SPLINE) and is being
reworked under #5318 to #5321; whether its polylines take a grid is decided there.

Tests: every SVG test unchanged, bytes included; the grid's rounding and its bounds; the A3 border
on the grid (y up, closing line left out, fill loops closed); slivers in the grid's own steps, with
the merge boundary both ways (10 and 11 steps, a 6-8 and a 7-8 step); an arc's quarters with their
radius and turn in steps; a width under half a step made one step; an ellipse refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: a sheet PNG at a chosen dpi and colour (#4890 A4)

A new writer, io::paper::paper_png, that reads the shared walk on the
nanometre grid, the one the SVG writer uses, so a PNG draws the same joints,
closures and merges. Neither the walk nor the SVG writer changes.

- Curves are flattened within 1/8 px by geom::paper::flatten, which turns
  each walked piece back into its Seg2 (an arc's centre from its ends, radius
  and direction) and takes Seg2::chords' steps. chords is the bound
  Seg2::polyline already used (the chord's bulge for arcs, Wang's bound for
  the rest), moved out of polyline unchanged so both read one rule.
  polyline itself doesn't fit: it floors an arc's tolerance at a millionth of
  its radius and caps a piece at 4096 steps, coarser than asked without a
  word (0.1 mm, 9 px, on a 100 m arc at 2400 dpi). Over 2^20 points a path,
  or 2^24 a sheet, is refused instead. tiny-skia only gets lines: its own
  flattening stops at 64 lines a curve, more than a pixel off on a large arc
  at a high dpi.
- 25 to 2400 dpi (300 by default). 8-bit greyscale (the default), 1-bit
  black and white (no anti-aliasing, every line at least a pixel wide), 24-bit
  colour, or 32-bit colour with transparent paper.
- The file records its dpi (pHYs) and sRGB, so it prints at the sheet's size.
- Each line is stroked once, then the sheet is drawn in bands of 16 MiB of
  pixels streamed into the file, so memory doesn't grow with the sheet. Each
  band also draws a row above and below that isn't written: tiny-skia starts
  an edge a little differently where a clip cuts it (one subsample, 16
  levels, in a few pixels), so no written row sits on a cut and the bytes are
  the same whatever the band.
- A line under a pixel wide (in greyscale and colour) is drawn as tiny-skia's
  hairline, its coverage scaled by its width: its stroked outline would fall
  between the rows the rasteriser samples and vanish.

Tests: the pixel size and pHYs of A4 at 300 and 25 dpi; a line's and a fill's
pixels at 254 dpi (butt ends, the fill's hole); each colour's PNG type, ink
and paper; a hairline between two rows' centres that is light grey in
greyscale and stays one pixel in black and white, and a bar a fifth of a pixel
wide that black and white inks by its centre; a closed outline joined at its
start; a 40 mm circle round to the pixel at 600 dpi; the same bytes drawn 1, 3
or 7 rows at a time or all at once, and again; the refusals; the flattener's
bounds and counts, and its point cap at the exact boundary; the scene-JSON
fuzz through the PNG writer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Geom: the pixel grid of a paper rectangle (#5624, W17b S1)

One rule for where a picture's pixels lie on paper, shared by the picture renderer and the writers
that place it: anchored at the rectangle's lower-left corner, rounded outward (W = ceil(w / p - 1e-9)
with p = 25.4 / dpi), so the overhang is at the top and the right; rows run top-down. At most
65,536 px a side; nothing caps the total, as pictures are drawn in tiles and strips.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures, and a sheet's pixel grid and windows (#5624, W17b S3; #4890 A9)

A picture on a sheet is `Item::Picture`: the rectangle it sits in, the name of its pixels
(`source`, opaque as a group's object id is) and its opacity. The pixels stay out of the scene: a
writer asks a `Pictures` provider for them on the grid it draws on. A PNG at 1200 dpi then gets a
shaded view drawn at 1200 dpi, not one stretched, and a scene stays small, plain data.

The pixel grid gains the two rules a sheet needs, beside the rectangle's own grid:
- `PixelGrid::sheet`: the whole sheet's grid, anchored at its top-left corner, as an image of
  the sheet is drawn.
- `PixelGrid::window`: the columns and rows a rectangle meets; a side on a pixel edge takes no
  sliver beyond it. A picture drawn into its window of the sheet grid is placed by copying whole
  pixels.

`Picture::grid` is a picture's own grid at a dpi, for SVG and PDF to embed. `Scene::validate`
counts a picture as one record. It refuses a rectangle outside the coordinate bound or without
area, and an opacity that isn't above 0 and at most 1. `Aabb2` is read and written as
`{min, max}`.

On the walk (C4), which S3 follows here: `GridItem::Picture` passes a picture through with its rectangle in
millimetres, since the pixel grid places its pixels, and the SVG and PNG writers refuse it ("not supported yet:
pictures (A9)") until their own slices. `GridItem`, `GridGroup` and `OnGrid` are `PartialEq` only, as a
picture holds numbers that aren't `Eq`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures in a sheet PNG (#5624, W17b S5)

The sheet PNG writer draws the scene's pictures (S3's Item::Picture) where
the scene orders them, between its lines and fills, from a Pictures provider.

- The PNG is now the sheet's pixel grid (PixelGrid::sheet, S1): anchored at
  the sheet's top-left corner and rounded outward, so nothing is clipped.
  210 mm at 300 dpi is 2481 px (it was 2480, rounded to the nearest). A side
  over 65,536 px is refused and the refusal names the limit ("a sheet PNG
  side of ... px at ... dpi is over the limit of 65536 px"); an A0 sheet
  fits up to about 1400 dpi.
- A picture fills exactly its window of that grid, so it registers with the
  lines drawn over the same rectangle. Each band asks the provider for just
  its own rows, as straight-alpha sRGB RGBA, and the bytes are checked: too
  few or too many, or the provider's error, refuse the PNG with the picture's
  source named. paper::NoPictures refuses every picture.
- Pictures are asked for at most MAX_PICTURE_DPI (1200): above it, on the
  grid k = ceil(dpi / 1200) times coarser, whose pixels each fill exactly
  k x k of the sheet's (both grids share the top-left anchor), so a 2400 dpi
  export doesn't ask a renderer for four times the pixels.
- Source-over in sRGB, the picture's alpha times its opacity; transparent
  paper keeps the alpha. In black and white each picture pixel is dithered
  with an 8 x 8 ordered (Bayer) matrix at its place on the sheet, so bands
  don't show; lines stay solid. PngReport says whether anything was
  dithered.
- In the writer a picture is one more shape a band draws, beside A4's
  outlines, fills and hairlines; the walk's picture arm passes pictures
  through, and the SVG writer still refuses them (A9).

Tests: a picture's exact window and its register with a fill over the same
rectangle, also on a sheet 100.8 px high (the grid anchored at its top);
pictures keep their place in the paint order; opacity scales the alpha (on
white and transparent paper, and the picture's own half alpha); above 1200
dpi a coarser picture's pixels are each copied 2 x 2; black and white
dithers half of mid grey and keeps a line over it solid, whatever the band;
a provider's short rows, its error and no provider are refused, and a
picture off the sheet asks for nothing; the bands test draws three pictures
1, 3 or 7 rows at a time, the same bytes and each picture row asked for
once; the sheet sizes on the outward grid and the 65,536 px refusal; the PNG
fuzz with a picture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures in a sheet SVG (#5624, W17b S5)

A picture is an <image> holding a PNG as a data URI, on the picture's own
pixel grid (Picture::grid) at the picture resolution the caller gives,
25 to 1200 dpi: paper_svg(scene, pictures, picture_dpi), the same shape as
paper_png. Only its window over the sheet is embedded and asked for, so it
never costs more pixels than the sheet holds; one off the sheet is left out.
Its x, y, width and height are the grid's own pixel edges put on the
nanometre grid every line is on, so a line on its edge has its number.
Opacity below 1 is written, in millionths as lengths are.

The PNG writer gives the embed its encoder settings (sRGB, the dpi
recorded) and its provider check, now one function each, and streams the
picture a band of rows at a time, refused once it passes the room left
under the file's 64 MiB.

Tests: the 1182 x 591 px grid at 300 dpi and its edges, a line on the
edge, paint order and opacity, the crop and the picture off the sheet, the
bands, and each refusal; the SVG fuzz case gains a picture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 14:01:08 -05:00
CamWightandClaude Opus 5.5 4060069fcd Drawing views: shaded snapshots and their pictures' parts (#5624, W17b S4) (#9798)
* Paper scene: one walk onto a writer's number grid, for every writer (#4890, C4)

The SVG writer kept its own number grid: each coordinate whole nanometres, every joint written once,
a closed path ending exactly on its first point, slivers merged into the next piece and a path's
last piece kept. PDF, PNG and the Drawing workspace's canvas need the same rules, so they move into
`geom::paper::grid`, and the SVG writer only formats what the walk gives it.

- `Grid`: `per_mm` steps a millimetre. `Grid::NANOMETRE` (six decimals of a millimetre) is SVG's.
  `Grid::new` takes 1 to 1e9; a power of two keeps every number exact as a binary float, which a
  writer of 32-bit numbers such as PDF can use.
- `Prepared::on_grid(grid) -> OnGrid`: each group's items as paths of `Piece`s (lines, arcs of at
  most a quarter turn, cubics, parabolas), y up as in the scene, a closed path's closing line left
  to the writer's close, and a report of widths snapped and pieces merged. Slivers are measured in
  the grid's own steps, so a coarser grid merges more.
- The SVG writer flips y on the whole numbers it's given and writes the same bytes as before.
  `SvgReport` is the grid's report.
- Ellipses, hyperbolas and Béziers above degree 3 are still refused (A3b), now by the walk.

DXF stays on its own number rule for now: it writes exact entities (ELLIPSE, SPLINE) and is being
reworked under #5318 to #5321; whether its polylines take a grid is decided there.

Tests: every SVG test unchanged, bytes included; the grid's rounding and its bounds; the A3 border
on the grid (y up, closing line left out, fill loops closed); slivers in the grid's own steps, with
the merge boundary both ways (10 and 11 steps, a 6-8 and a 7-8 step); an arc's quarters with their
radius and turn in steps; a width under half a step made one step; an ellipse refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: a sheet PNG at a chosen dpi and colour (#4890 A4)

A new writer, io::paper::paper_png, that reads the shared walk on the
nanometre grid, the one the SVG writer uses, so a PNG draws the same joints,
closures and merges. Neither the walk nor the SVG writer changes.

- Curves are flattened within 1/8 px by geom::paper::flatten, which turns
  each walked piece back into its Seg2 (an arc's centre from its ends, radius
  and direction) and takes Seg2::chords' steps. chords is the bound
  Seg2::polyline already used (the chord's bulge for arcs, Wang's bound for
  the rest), moved out of polyline unchanged so both read one rule.
  polyline itself doesn't fit: it floors an arc's tolerance at a millionth of
  its radius and caps a piece at 4096 steps, coarser than asked without a
  word (0.1 mm, 9 px, on a 100 m arc at 2400 dpi). Over 2^20 points a path,
  or 2^24 a sheet, is refused instead. tiny-skia only gets lines: its own
  flattening stops at 64 lines a curve, more than a pixel off on a large arc
  at a high dpi.
- 25 to 2400 dpi (300 by default). 8-bit greyscale (the default), 1-bit
  black and white (no anti-aliasing, every line at least a pixel wide), 24-bit
  colour, or 32-bit colour with transparent paper.
- The file records its dpi (pHYs) and sRGB, so it prints at the sheet's size.
- Each line is stroked once, then the sheet is drawn in bands of 16 MiB of
  pixels streamed into the file, so memory doesn't grow with the sheet. Each
  band also draws a row above and below that isn't written: tiny-skia starts
  an edge a little differently where a clip cuts it (one subsample, 16
  levels, in a few pixels), so no written row sits on a cut and the bytes are
  the same whatever the band.
- A line under a pixel wide (in greyscale and colour) is drawn as tiny-skia's
  hairline, its coverage scaled by its width: its stroked outline would fall
  between the rows the rasteriser samples and vanish.

Tests: the pixel size and pHYs of A4 at 300 and 25 dpi; a line's and a fill's
pixels at 254 dpi (butt ends, the fill's hole); each colour's PNG type, ink
and paper; a hairline between two rows' centres that is light grey in
greyscale and stays one pixel in black and white, and a bar a fifth of a pixel
wide that black and white inks by its centre; a closed outline joined at its
start; a 40 mm circle round to the pixel at 600 dpi; the same bytes drawn 1, 3
or 7 rows at a time or all at once, and again; the refusals; the flattener's
bounds and counts, and its point cap at the exact boundary; the scene-JSON
fuzz through the PNG writer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Geom: the pixel grid of a paper rectangle (#5624, W17b S1)

One rule for where a picture's pixels lie on paper, shared by the picture renderer and the writers
that place it: anchored at the rectangle's lower-left corner, rounded outward (W = ceil(w / p - 1e-9)
with p = 25.4 / dpi), so the overhang is at the top and the right; rows run top-down. At most
65,536 px a side; nothing caps the total, as pictures are drawn in tiles and strips.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures, and a sheet's pixel grid and windows (#5624, W17b S3; #4890 A9)

A picture on a sheet is `Item::Picture`: the rectangle it sits in, the name of its pixels
(`source`, opaque as a group's object id is) and its opacity. The pixels stay out of the scene: a
writer asks a `Pictures` provider for them on the grid it draws on. A PNG at 1200 dpi then gets a
shaded view drawn at 1200 dpi, not one stretched, and a scene stays small, plain data.

The pixel grid gains the two rules a sheet needs, beside the rectangle's own grid:
- `PixelGrid::sheet`: the whole sheet's grid, anchored at its top-left corner, as an image of
  the sheet is drawn.
- `PixelGrid::window`: the columns and rows a rectangle meets; a side on a pixel edge takes no
  sliver beyond it. A picture drawn into its window of the sheet grid is placed by copying whole
  pixels.

`Picture::grid` is a picture's own grid at a dpi, for SVG and PDF to embed. `Scene::validate`
counts a picture as one record. It refuses a rectangle outside the coordinate bound or without
area, and an opacity that isn't above 0 and at most 1. `Aabb2` is read and written as
`{min, max}`.

On the walk (C4), which S3 follows here: `GridItem::Picture` passes a picture through with its rectangle in
millimetres, since the pixel grid places its pixels, and the SVG and PNG writers refuse it ("not supported yet:
pictures (A9)") until their own slices. `GridItem`, `GridGroup` and `OnGrid` are `PartialEq` only, as a
picture holds numbers that aren't `Eq`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures in a sheet PNG (#5624, W17b S5)

The sheet PNG writer draws the scene's pictures (S3's Item::Picture) where
the scene orders them, between its lines and fills, from a Pictures provider.

- The PNG is now the sheet's pixel grid (PixelGrid::sheet, S1): anchored at
  the sheet's top-left corner and rounded outward, so nothing is clipped.
  210 mm at 300 dpi is 2481 px (it was 2480, rounded to the nearest). A side
  over 65,536 px is refused and the refusal names the limit ("a sheet PNG
  side of ... px at ... dpi is over the limit of 65536 px"); an A0 sheet
  fits up to about 1400 dpi.
- A picture fills exactly its window of that grid, so it registers with the
  lines drawn over the same rectangle. Each band asks the provider for just
  its own rows, as straight-alpha sRGB RGBA, and the bytes are checked: too
  few or too many, or the provider's error, refuse the PNG with the picture's
  source named. paper::NoPictures refuses every picture.
- Pictures are asked for at most MAX_PICTURE_DPI (1200): above it, on the
  grid k = ceil(dpi / 1200) times coarser, whose pixels each fill exactly
  k x k of the sheet's (both grids share the top-left anchor), so a 2400 dpi
  export doesn't ask a renderer for four times the pixels.
- Source-over in sRGB, the picture's alpha times its opacity; transparent
  paper keeps the alpha. In black and white each picture pixel is dithered
  with an 8 x 8 ordered (Bayer) matrix at its place on the sheet, so bands
  don't show; lines stay solid. PngReport says whether anything was
  dithered.
- In the writer a picture is one more shape a band draws, beside A4's
  outlines, fills and hairlines; the walk's picture arm passes pictures
  through, and the SVG writer still refuses them (A9).

Tests: a picture's exact window and its register with a fill over the same
rectangle, also on a sheet 100.8 px high (the grid anchored at its top);
pictures keep their place in the paint order; opacity scales the alpha (on
white and transparent paper, and the picture's own half alpha); above 1200
dpi a coarser picture's pixels are each copied 2 x 2; black and white
dithers half of mid grey and keeps a line over it solid, whatever the band;
a provider's short rows, its error and no provider are refused, and a
picture off the sheet asks for nothing; the bands test draws three pictures
1, 3 or 7 rows at a time, the same bytes and each picture row asked for
once; the sheet sizes on the outward grid and the 65,536 px refusal; the PNG
fuzz with a picture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper scene: pictures in a sheet SVG (#5624, W17b S5)

A picture is an <image> holding a PNG as a data URI, on the picture's own
pixel grid (Picture::grid) at the picture resolution the caller gives,
25 to 1200 dpi: paper_svg(scene, pictures, picture_dpi), the same shape as
paper_png. Only its window over the sheet is embedded and asked for, so it
never costs more pixels than the sheet holds; one off the sheet is left out.
Its x, y, width and height are the grid's own pixel edges put on the
nanometre grid every line is on, so a line on its edge has its number.
Opacity below 1 is written, in millionths as lengths are.

The PNG writer gives the embed its encoder settings (sRGB, the dpi
recorded) and its provider check, now one function each, and streams the
picture a band of rows at a time, refused once it passes the room left
under the file's 64 MiB.

Tests: the 1182 x 591 px grid at 300 dpi and its edges, a line on the
edge, paint order and opacity, the crop and the picture off the sheet, the
bands, and each refusal; the SVG fuzz case gains a picture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing views: shaded snapshots and their pictures' parts (#5624, W17b S4)

drawing.view.snapshot takes `shaded` as a display or reference style and,
when a picture is drawn, says which: picture {pixels, dpi, rect_mm}, the
boundary's pixel grid at the sheet's dpi or 1200, the lower, since lines
may export finer. It renders nothing. A shaded part's own lines are left
out of the reply, as its picture draws it; it still hides what lies behind
it. shaded_visible and shaded_visible_hidden stay refused until the view's
exact lines (W06), and shaded with `scene` (and on a sheet) until the
paper scene draws pictures.

- engine::drawing_picture: the pictures a view draws, bottom first, and how
  each body takes part in each (painted, depth only, left out). The model
  always hides reference parts behind it; reference parts hide the model
  only when reference_hides_model is on; shaded reference parts that don't
  get a picture of their own under the model's, so they never cover it.
- product_scene::extract_where: the scene for the bodies a caller picks;
  extract keeps the browser's eye, a drawing draws every body.

Tests: every style and hide case, reference parts through a subassembly
and a part, the picture's size at 300, 1200 and 2400 dpi, a view with
nothing shaded, the hidden body in the filtered scene, and which lines a
shaded model or shaded reference parts leave.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 14:01:04 -05:00
CamWightandClaude Opus 5.5 a4c9bc380f Paper scene: text runs and their fonts, as records with their rules (#4890 B1a) (#9512)
`Item::Text` is a run of placed glyphs in one font, with its logical text and the clusters that
tie the two together, and `Scene.fonts` the table of the fonts the runs use (as posted on #4890,
with each glyph's `hmtx` advance). `Scene::validate` holds them to the posted rules. Writers never
parse a font.
- A run names its font by the font's id (a hash of its face's bytes and index), so groups laid
  out apart name a font alike whichever machine's font it is; `Scene::add_fonts` joins their
  tables at assembly, kept in order of id, so the same input gives the same bytes.
- A font's bytes never go through serde: a scene read from JSON is refused until they come from
  the font of the id it names.
- `TextRun::outlines` places the glyph outlines on the sheet, for the writers to fill nonzero.
The SVG writer refuses text by name until B2. `Scene::new` builds a scene without text, so the
next field doesn't touch every caller again.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 13:15:32 -05:00
CamWightandClaude Opus 5.5 f219d9b795 Paper text: a text laid out by sketch's rule, as the scene's runs and their drawing font (#4890 B1b) (#9513)
* Paper scene: text runs and their fonts, as records with their rules (#4890 B1a)

`Item::Text` is a run of placed glyphs in one font, with its logical text and the clusters that
tie the two together, and `Scene.fonts` the table of the fonts the runs use (as posted on #4890,
with each glyph's `hmtx` advance). `Scene::validate` holds them to the posted rules. Writers never
parse a font.
- A run names its font by the font's id (a hash of its face's bytes and index), so groups laid
  out apart name a font alike whichever machine's font it is; `Scene::add_fonts` joins their
  tables at assembly, kept in order of id, so the same input gives the same bytes.
- A font's bytes never go through serde: a scene read from JSON is refused until they come from
  the font of the id it names.
- `TextRun::outlines` places the glyph outlines on the sheet, for the writers to fill nonzero.
The SVG writer refuses text by name until B2. `Scene::new` builds a scene without text, so the
next field doesn't touch every caller again.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper text: a text laid out by sketch's rule, as the scene's runs and their drawing font (#4890 B1b)

`paper_text` lays a text out with `TextMetrics`, the rule sketch text draws and measures with, and
returns the paper scene's runs of placed glyphs (B1a) with the font they are set in, holding the
glyphs they place: each one's `hmtx` advance and its outline in font units.
- Drawings are set in bundled fonts only (`DRAWING_FONTS`: Liberation Sans for now). A font is
  admitted once: a static face with a PostScript name, a cap height and a `post` table, its stem
  width measured across its `l`, and its id hashed then. A run names its font by that id.
- A character the font has no glyph for is refused by name; bold is refused; an italic leans as
  sketch text's made-up italic does; text on paper follows no path.
- Lines break a pitch apart and align as sketch text's do. A line with right-to-left characters
  is a run per direction, shaped, with clusters from the shaper; any other is one run, a
  character to a glyph.
`TextMetrics` keeps one layout: it now reports each glyph's cluster and the line's runs, and the
glyph outline walk gives loops of `Seg2`, which sketch text turns into the geometry it drew before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Paper text: the fonts a text uses, each once (#4890 F0)

`paper_text` returns its runs with the list of fonts they use, each once in order of id (so the
same text gives the same list) and each holding exactly the glyphs its runs place, where it
returned one font. Still one drawing font today; the list is the shape a text set in several
fonts (a Latin, Arabic or CJK line) needs, and `Scene::add_fonts` already takes it as it is.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 13:15:28 -05:00
CamWightandClaude Opus 5.5 21888e8000 Drawing views: reference parts in a view's lines and thin on paper (#5866, V5) (#10520)
* Reference parts in a drawing view's lines (V5a, #5866)

A body placed through an occurrence that counts as reference is a reference
part: one rule, Document::reference_occurrences. Its lines say so in replies.
With reference_hides_model off (the default) the model's lines skip reference
parts at the one occlusion test, and reference lines test against everything.
The two roles merge apart, so a reference line never takes a model line's
place on one stretch. Extents come from the model parts when there are any. A
sketch curve is the model's: a reference part hides it only as it hides the model.

One thing the plan didn't name: reference lines stop at the view's boundary
(extents and reference margin) when model parts are there too, which is what
the margin is for; a part much bigger than the model would otherwise run off
the sheet.

Hidden reference lines in a scene are refused until dashed lines (V7b), as the
model's are.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Reference parts on paper: thin, under the model's lines (V5b, #5866)

Saved views and the snapshot's scene alike, 0.25 mm until the drawing styles
slice sets widths by kind of line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Reference parts: tests the planted faults asked for (#5866 V5)

A reference line against a model line on one stretch, and against another reference part; one part
placed through a reference assembly and on its own; a dimension on a reference part's edge; undoing
and redoing a change of type, set on the occurrence or its component; and a saved view's setting.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* A test reads any other paper item as no stroke (#5866 V5)

Naming Fill alone would stop compiling when #7590 adds a kind of item, with
no conflict for git to show.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 11:51:20 -05:00
CamWightandClaude Opus 5.5 9b73ec8d9d Display units: every length unit of the expression table shows its own numbers (#4896) (#7809)
A drawing's fixed unit may be any length unit of the expression unit
table (expr::UNITS: um, inch, yd, mil and the rest besides a design's
five), but the display functions knew only the five and showed the
others' measurements as millimetres. unit_scale now reads the same
table, so the two can't drift apart again; the five keep exactly the
scales they had (1 over 10, 1000, 25.4 and 304.8). The inch rule (a
fourth digit when it means something) follows an inch's size, so
`inch` reads as `in` does.

A design's own units are only the five, so nothing it shows changes.
The test that pinned `inch` to millimetres becomes one that checks
every length unit of the table shows its own numbers; angle names
(`deg`, `rad`) and other text keep the millimetre fallback.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-31 05:38:22 -04:00
CamWightandClaude Opus 5.5 0d4739fffb Drawing dimensions: between two cylinders' silhouettes (#4896, W20b) (#9251)
* Drawing dimensions: a hole's or an arc's centre as a point to measure from (#4896, W20c)

A linear dimension's point can be the centre of a circle or an arc, saved as its one edge's name
with `centre`. It is found again as construction geometry finds a round edge's centre
(`construct::as_circle`), in 3D, so any view can measure it; a split edge is ambiguous, and an edge
that is no longer round can't be measured. `drawing.dimension.add` takes `{centre: edge}` wherever
a corner goes, and refuses a straight edge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing dimensions: a chain of distances along one line (#4896, W20c)

A chain dimension runs through 3 to 64 points in order (corners or hole centres) and reads the
distance from each to the next along one line as the view shows it: horizontal, vertical, along a
direction, or aligned from the first point to the last. The line runs the way the first point to
the last goes, and a point out of order is refused, never moved, so the members add up to the
overall from the first point to the last (`drawing_measure::chain`). It reads projected only (true
lengths of points off one line don't add up) and shows its members' numbers, never a text instead.
`drawing.dimension.add` takes `{chain: {points, direction?}}` and counts the points before looking
any up; `drawing.dimension.edit` changes its direction; inspect lists its members, its value being
the overall.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing dimensions: a whole cylinder's diameter between its silhouettes (#4896, W20b)

A cylinder dimension saves a face by name (ModelFace, as ModelEdge names an edge) and reads Ø
between the cylinder's two silhouettes, which any view but one along its axis shows 2r apart
(`drawing_measure::silhouette_side`). The face, or the pieces it was cut into, must lie on one
cylinder (the kernel's own condition for faces on a common cylinder); every face of the body on
it counts, since the kernel builds a whole cylinder as two half faces; and both silhouettes must
lie on those faces, so a fillet's quarter round is refused. `Document::face_reference` picks a
face by a point on it or by its index; `drawing.dimension.add` takes `{cylinder: {face}}`, edit
refuses its settings, and inspect can list face candidates.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drawing dimensions: between two cylinders' silhouettes (#4896, W20b)

A dimension between a silhouette of one cylinder and a silhouette of another, as the view shows
them. Each is named by a face of its cylinder and by its side: the one nearer the other cylinder
on the view, or the one farther (SilhouetteSide). So it keeps to them whichever way round the
view shows the cylinders.

- geom: drawing_measure::between_silhouettes takes the two cylinders (Cylinder, now geom's: the
  type doc kept privately), their sides and a direction as a linear dimension takes it (aligned:
  square to the silhouettes). The silhouettes must be parallel on the view and the axes apart
  there. Foreshortened when the axes aren't parallel in the model, or the silhouettes lie at
  different depths.
- doc: DimensionKind::Silhouettes { a, b: Silhouette { face, side }, direction }, read as the
  view shows it. Each silhouette must lie on its cylinder's faces (Lookup::on_cylinder, now shared
  with the cylinder kind). Document::silhouettes_kind makes two silhouettes of one cylinder of
  one body that cylinder's diameter (which takes no direction), else the distance between them.
- engine: drawing.dimension.add takes {silhouettes: {a, b: {face, side}, direction?}};
  drawing.dimension.edit changes its direction only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-31 04:49:53 -04:00
CamWightandClaude Opus 5.5 693cf911f1 Recovery: unsaved drawing work survives a crash (#3890, K9′) (#9289)
* Recovery: a change to a drawing alone is autosaved and recovered (#3890, K9')

A design's drawings are part of its document (K3b), so the autosave already sees a drawing's
change as unsaved work and writes it with the design, and recovery reads it back. This test
holds that: with the model saved, a change to a drawing's properties alone is autosaved, and the
recovered design has the drawing as changed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Recovery: a drawing that breaks a rule is left out of the autosave and the crash copy, named (#3890)

Reading a design refuses it whole for one drawing that breaks a rule of the record, so an autosave
or a crash report's design holding one never opened again: all the unsaved work was lost. Both
copies now keep the rest of the design (drawing::openable) and name each drawing left out and the
rule: the autosave as its error, the crash report in its notes. A design's save error names the
drawing it couldn't save too (Drawing::label).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Recovery: a copy that must open keeps a design's drawing rules too (#3890)

drawing::openable now leaves out what a read refuses at the design's level as well: the later of
two drawings with one id, and the drawings past MAX_DRAWINGS, each named with its rule. So an
autosave never keeps a stale copy over a drawing problem. The id rule is take_id, which
check_in_design uses too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Recovery: a drawing too large to save is left out too, and the status line names it (#3890)

The autosave and the crash copy already leave out a drawing that breaks a rule of the record.
A drawing too large to save failed the whole autosave instead; `openable` now leaves it out the
same way, before its id, so the rest of the design is kept. Nothing is skipped silently: the
autosave's error names each drawing left out, and the status line shows it.

Tests: a drawing left out takes no id from a later one; an autosave leaves out one too large to
save; an autosave keeps what is hidden; the status line names a drawing the autosave left out.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-31 00:56:45 -04:00
CamWightandClaude Opus 5.5 b21c377296 A lost dimension keeps its last value after the design is reopened (#4896 W19d) (#9764)
Dimension::last_good saves the value the session last read, written only when
there is one, and a save takes it from the session's reads as it takes what is
hidden: no read touches the document, makes an undo step or asks to be saved.
drawing.dimension.inspect answers this session's read first, else the file's.
resolve_dimensions never sees it, so a lost dimension stays lost. A value past
the drawing format's limit is left out of the save, not written into a file that
then can't open.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 18:59:15 -04:00
CamWightandClaude Opus 5.5 878bc1b421 A point on a curved edge finds it, between the edge's chords (#4896) (#9765)
An edge given as a point looked along the edge's mesh lines, within a point's tolerance, so a point on a
circle or an arc that fell between two chords found no edge: a radial dimension on most points of a rim
said "no edge at". A point off every chord now takes the edge whose chords are within what its body was
meshed to, and only when no edge passes through the point itself, so a point on a straight edge beside
a curve is still that edge.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 18:59:11 -04:00
CamWightandClaude Opus 5.5 562bb379a5 Dimension layout: sizes for paper, and text above the line (#4896, W19c) (#9193)
* Dimension layout: sizes for paper, and text above the line (#4896, W19c)

A dimension's sizes become a parameter, so a drawing can lay its
dimensions out in paper millimetres with the same code as a sketch:
DimSizes { arrow, gap, overshoot, offset, above } and
dim_layout_sized. dim_layout and dim_layout_reaching keep their
signatures and use DimSizes::screen(px), today's 9, 3, 4 and 22 px;
the other default text places stay the same fractions of the offset
(18, 48 and 20 to 22), and default_text takes its 22 from the same
place. A size that isn't a finite number above 0 (or is beyond MAX_COORD,
the largest length a sketch takes) falls back to its 1 px value, as a
bad px does today.

`above` puts a linear dimension's text above an unbroken dimension
line (ISO 129-1), running along the line and reading from the bottom
or the right; DimLayout::text_dir gives its direction (level for text
in the line). The arrows go inside or out by the same test, with the
text's length along the line; a leader to text outside runs on under
it.

On main, #7056 moved the four default places into
default_text_offset(frame, px). This leaves those lines as they are,
so when main next comes into drawings the layout's arms call it with
this px (the offset's 22nd), and lib.rs's export line takes both lists.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Format the dimension layout's re-exports (#4896, W19c)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 13:59:49 -04:00
bherbruck 1a47ac13de Merge pull request #9467 from tangency-labs/sync/main-into-drawings-1030
Sync main into drawings (Oct 30)
2026-10-30 10:29:35 -04:00
bherbruckandClaude Opus 5.5 1c957c105d Translate the drawings' new messages into the complete catalogs (ar, he, vi)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 10:15:28 -04:00
CamWightandClaude Opus 5.5 ec527a32ab geom: arrowheads and leaders on paper (#4890 D1a) (#9466)
`paper::annotate` places the marks that point at things on a sheet as plain
strokes and fills in sheet millimetres: open, closed and filled arrowheads,
the oblique stroke and the dot, and leaders (points, a landing line in the
reading direction, a terminator at the first point). Every size is the
caller's; ISO 128-2's dot and landing ratios and ISO 129-1's 15° to 90°
arrowhead angle are constants beside them. A closed or filled arrowhead stops
its line at the base. Dimension layout stays W19c's (#4896) and the datum
triangle the datum symbol's (#7257).

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 09:59:02 -04:00
bherbruckandClaude Opus 5.5 d01de87552 Merge main into drawings (Oct 30)
Brings the 1,270 commits on main since the drawings branch last met it,
including the first drawings landing (#8249), into drawings.

Merged with the drawings snapshot that #8249 landed (ebafe0748) as the
merge base, so only the drawings work after that landing is replayed on
top of main. Main's structure wins; drawings' additions are kept and
adapted to main's APIs (metadata Owner/KEYS table, checked! without a
contract argument, Snapshot visibility, crash journal outcome, the
part-properties normalisation, model_body_components, io helpers).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 09:54:08 -04:00
github-actions[bot] ab67126eec Regenerate generated files and fold pending files [skip ci] 2026-10-30 12:31:20 +00:00
Nathan Dane 3b8cbc4e22 Viewport: surface analysis status items and legend keys (#8547) (#8556) 2026-10-30 12:12:19 +00:00
FrankandClaude Opus 5.5 f59f6698f1 Ribbon drop-downs as a grid of large icons, as an option (#4739) (#5012)
* Ribbon drop-downs as a grid of large icons, as an option (#4739)

Preferences > Interface gets two settings, both off by default:

- Menus: List | Large icons. With Large icons a panel's drop-down shows its
  commands as rows of up to five icons at 32 px (twice the list's), keeping
  the list's group headings and order. The name, shortcut and help are in
  the tooltip, as on a ribbon button. Arrow keys move across and between
  rows; a click starts the command; right-click and drag pin it as before.
  Resting on an icon for two seconds arms a pin in its corner; a quicker
  click anywhere starts the command, so nothing is pinned by accident. A
  pinned command shows its pin.
- "Leave pinned commands out of the drop-down menus": a drop-down (list or
  grid) leaves out the commands the ribbon shows as buttons; a panel too
  narrow for its buttons lists them again, and a menu is never left empty.

No icon artwork or colours change. Sketch > Create keeps its family list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sketch > Create in the large-icon grid: every variant under its family (#4739)

With Menus: Large icons, Sketch > Create showed its family sub-menus as
plain text rows (Line, Rectangle, Circle, ...), unlike every other menu. It
now shows every variant as a large icon under its family's heading, then the
other tools under "More". The list keeps its family sub-menus.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: groups side by side, no footer (#4739)

A menu of two or more headed groups (Sketch > Create's families, Model >
Modify's Edges, Faces, Bodies, Specialise) puts the groups side by side, each
heading over a column of its icons, at most two across. Menus without groups,
or one too wide for the window, keep rows of up to five. The pinning note
under the large-icon menus goes: the tooltips and right-click menus say it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: a separator line between groups side by side (#4739)

As between the ribbon's sections: a 1 px line in the theme's border colour.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: at most three rows, growing to the right (#4739)

A menu fills five columns, more when that would take over three rows (Model >
Create: 7 x 3); in grouped menus a group too big for three rows gets more
columns. Only a window too narrow for that wraps into more rows. A test checks
every panel's menu of every tab.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: group headings centred over their icons (#4739)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drop-down group headings: smaller than the drop-down's header, not in capitals (#4739)

Edges, Faces, Bodies (and Sketch > Create's families) were drawn in capitals
at font_small, which reads larger than the drop-down's own header
(font_caption). They now use font_micro, written as named, in the list and the
large-icon grid. Review feedback on #5012.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: start a command from the grid at its icon, as main's rows do (#4739)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Ribbon icon grid strings: a pending locale file, with ar, he and vi (#4739)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Large-icon menus: commands in no group stand apart, so a left-out group doesn't move them (#4739)

With Convert to Sheet Metal pinned and left out of Model > Modify, its
Specialise group went and Delete, in no group, moved under Bodies.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 11:59:08 +00:00
Rick MorgansandClaude Opus 5.5 bd1525dcfb Expressions: hyperbolic functions sinh, cosh, tanh and their inverses (#9332) (#9334)
* claim: #9332

* Expressions: hyperbolic functions sinh, cosh, tanh and their inverses (#9332)

`sinh`, `cosh`, `tanh` and `asinh` take a unit-less number and give one,
as `exp` does; `acosh` refuses x < 1 and `atanh` refuses |x| >= 1, worded
as `asin` and `ln` refuse what lies outside their domain. They are in the
expression help (`FUNCTIONS`, the parameter autocomplete), the
hostile-expression fuzz tokens, the parameters and equation-curve guides,
and the checking skill's list of functions. Horn flares (Salmon's hypex
`cosh h + T sinh h`), catenaries and the tractrix can now be written as
they are stated instead of expanded into `exp`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 11:13:59 +00:00
kaltenbrunnermarkusandClaude Opus 5.5 1e5eb1e3b8 Sketch: find profiles island by island, so dragging in a large imported logo stays smooth (#6896) (#9313)
* claim: #6896

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013WUFq6QFBqmZjxwaKy2J1M

* Sketch: find profiles island by island and say when the search is cut short (#6896)

Every live drag frame ran the profile search over the whole sketch: every
spline piece of a large imported logo was sampled, checked for crossings
and traced again, about 100 ms per frame at 6,000 degrees of freedom.

Profile curves are now split into islands whose enlarged piece boxes do
not touch (the margin exceeds every tolerance the search uses), each
island's loops are found on their own and cached by a fingerprint of its
geometry, and the loops are nested into profiles over the whole sketch in
the order a single search finds them, so profiles, ids, holes and
whole-boundary certificates stay the same. The end-near-crossing lookup
covers its whole reach, independent of the grid's cell size.

The 50,000-piece limit now holds per island, and a sketch that still
exceeds it gets a warning instead of silently missing profiles.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013WUFq6QFBqmZjxwaKy2J1M

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 11:13:52 +00:00
ABDULLAH a1483d63c1 Arabic: plural wording in the Discard autosaves title (#9354)
recovery.discard_all_title read "discard {count} of the autosave"; it now names the autosaves as a plural.
2026-10-30 11:10:28 +00:00
semmelrindeandClaude Opus 5.5 491687d4c1 Sketch: the cursor snaps On a spline, ellipse or conic while drawing, as on a line or circle (#9321) (#9343)
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 11:10:25 +00:00
Rick MorgansandClaude Opus 5.5 66bd64cc5e Engine: one contract check for a call, used by the preflight, the bridge and the app (#9339)
`schema::issues(id, params)` looks up the command's typed contract and
lists what is wrong with the parameters (none without a contract or for
an unknown id). The batch preflight, the MCP bridge's pre-send check and
the app's check of a design-switching command each did that lookup
themselves; they now call it. No behaviour change.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 10:21:06 +00:00
jeromehzl f9146cdd58 Hole: a Hole at sketch points hides its sketch as an Extrude does (#9082) (#9211)
sketch_consumed knew Extrude, Revolve and the sheet metal flanges but not
a Hole drilled at a sketch's points, so that sketch stayed drawn inside the
holes. A Multiple hole now counts as using its sketch; Edit Feature shows
its points while the dialog is open. Single holes keep their sketch shown,
since a hidden sketch offers no points to snap to.
2026-10-30 09:37:53 +00:00
github-actions[bot] 23e7ac767e Regenerate generated files and fold pending files [skip ci] 2026-10-30 09:58:59 +00:00
phanoobandClaude Opus 5.5 3ac7a196c9 Recover unsaved design?: Discard all, with a confirmation (#9299) (#9300)
* claim: #9299

* Recover unsaved design?: Discard all, with a confirmation (#9299)

A Discard all button at the left of the window's footer, away from Later.
It asks first ("Discard 7 autosaves? This cannot be undone.", Discard all /
Cancel), then discards every listed autosave the way its own Discard button
does (doc.recovery_discard on each id). Cancel goes back to the list
unchanged. The button publishes the handle recovery:discard_all.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-30 09:26:04 +00:00
Logan NightandClaude Code 202e713cf1 Preferences: search finds settings in rows without a label (#9265) (#9318)
* claim: #9265

Co-Authored-By: Claude Code <noreply@anthropic.com>

* Preferences: a setting in a row with no label is found by the words it shows (#9265)

A search hid a row by its label before its controls were drawn. Seven checkbox rows and two segmented rows on the Interface page have no label, so "status bar", "search results" and "reveal on hover" found nothing. A hidden row with no label now keeps the words of its controls, as a dropdown keeps its value, and is drawn when they match. Each such row has a key of its own, so its neighbours are not drawn with it.

Co-Authored-By: Claude Code <noreply@anthropic.com>

---------

Co-authored-by: Claude Code <noreply@anthropic.com>
2026-10-30 09:25:24 +00:00
N3MI-DGandClaude Opus 5.5 63923da2ca Viewport: an opened design is framed at the home view, in the middle of the view (#9195) (#9204)
* claim: #9195

* Viewport: an opened design is framed at the home view, in the middle of the view (#9195)

A fit the viewport cannot frame yet is kept and asked again on the next frame instead
of being consumed and lost, so a design opened from the Home screen or by launching on
a file no longer comes up at the default camera, wherever its own coordinates put it.

Framing then centres the model on the view rather than on the free rectangle between
the floating controls: `Layout` carries `middle`, the view without the panels, computed
before the compass, the navigation bar and the OK/Cancel corner join the obstacles. The
model is still sized to `drawable`, so it stays clear of every control, but only a panel
moves it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Viewport: keep the centred fit clear of the controls, and frame --sample (#9195)

The fit framed the model into a rectangle the size of `drawable` centred on `middle`, which is
not inside `drawable`, so the model could sit under the compass, the navigation bar or the
render product frame. The model is now sized to `drawable` as before, and only the room left
over in it moves the model towards the middle of the view.

`s27_move_triad` turns the Z ring on its far side: the model now sits further right, and the
Move dialog that docks on the right afterwards covered the ring's old grab point.

A launch with `--sample` builds the design straight into the session and never asked for a
fit; it now comes up at the home view, framed, like an opened design.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Viewport tests: the cursor orbit's pivot keeps the pixel it was picked on (#9195)

With the model now centred on the view, one of the test's perspective drag points lands within
the pick tolerance of the box's top edge, and the pick snaps the pivot onto that edge, a few
pixels from the cursor. The orbit keeps that point still, as it should; the test compared it
with the cursor instead of with where it started.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-10-30 09:14:18 +00:00
jeromehzl 7e22aed739 View ribbon: Visual style and Navigation tool offer their choices instead of an error (#8868) (#9127)
Both commands need a choice, but start_with_variant_memory only special-
cased the parameterless display toggles, so a click on the ribbon (or
command search, a shortcut, ui.start) ran them with {} and showed only
the engine's "style must be ..." / "tool must be ..." error.

They now open a short menu, the way Units does: under the ribbon row
that started it (a second click closes it), or in the middle of the view.
Visual style lists the four styles with the current one checked; the
Navigation tool lists Orbit, Pan and Zoom, and picking the latched one
releases it, as its navigation-bar button does. The four styles come
from one list shared with the navigation bar's menu.
2026-10-30 09:04:39 +00:00
jeromehzl 966b10e731 Sketch: a feature on a failed sketch names that sketch (#9075) (#9276)
A sketch that failed (a dimension set negative by a parameter) left every
feature built on it with "unknown sketch 1 (it must come earlier in the
timeline)", which points at the timeline order. The evaluation now notes
the features that failed earlier in the pass, and a feature whose input
sketch is one of them fails with "sketch `Sketch1` failed: <its error>".
That result is never cached, since the cache key holds neither the
sketch's name nor its error: a second failing value or a rename shows
up at once.
2026-10-30 09:03:46 +00:00