The overlap sweep moves a horizontal line down the page, and its time
per event grows with the number of segments on that line. A long, low
path, such as rows of a zigzag, puts a whole row on the line at once.
stroke_region, unite_all_with and try_unite_all_with now estimate how
many segments the sweep holds at once in each direction, from up to
4,096 segment boxes (one from each run of segments, at a position that
changes from run to run). When the estimate for the input turned a
quarter turn is under two thirds of the upright one (2,048 segments or
more), the operation runs on the turned input and the result is turned
back. A quarter turn swaps and negates coordinates, so no point moves
by rounding. Offset Path is not turned. Callers of unite_all,
try_unite_all and outline_stroke take the turned sweep through these
functions.
Tidy's lookups of the input curves around a piece and of the kept
anchors can search along y: once lookups with more than 64 candidates
along x have scanned 16 per item, a lookup sorts the list along y once
and uses that window when it holds fewer candidates (under a quarter as
many for curves). Where the y window rarely helps, only one lookup in
32 measures it. is_kept counts the few points within its tolerance
directly instead of searching for the window's end.
In the sweep_axis tests, a turned result has as many subpaths as the
result without the turn, and the area where the two differ is no
larger than the refit precision times the outline's length.
On one open path of 200,000 anchors in rows of 1,000, Outline Stroke
takes 17.3 s of processor time instead of 31.3 s and Unite 2.9 s
instead of 6.5 s in a release build. Unite of 2,000 circles takes the
same time as before.
The contour passes of the two-path boolean operations, Offset Path and
the many-shape boolean (`boolean_n`, used by compound shapes and
Intertwine) built a flag for every input shape for each winding they
tested. They now test the list of inputs that cover the face, ascending,
refilled in one buffer per pass, so a test costs the size of the cover
rather than the number of shapes. `boolean_n` and `try_boolean_n` take
that list; the compound shape and Intertwine tests read it directly.
The contours, their order and the output are the same: a new test
compares every pass's test with the flag form on overlapping, nested,
touching and apart squares and on 72 overlapping squares, and another
checks that a pass over 100 squares apart hands its test one input at
most.
Finding the distinct regions of an arrangement built a flag for every
input shape at each edge and kept them in a sorted set, so it cost edges
times shapes in time and memory. Each coverage is now the sorted list of
the shapes that cover it, deduplicated in a hash set and sorted into the
order the flag masks had, so regions, their order and the output of
Divide, Hard Mix, Soft Mix, Shape Builder and the flattener are the same.
The face lookup and Pathfinder merging take the same lists.
Tests compare the lists and the regions with the flag masks for
overlapping, nested, touching and disjoint shapes and for 72 shapes.
normalize_bez ran the sweep without the catch_unwind guard that Sweep::new has,
so an outline_stroke width of 1e200 or more with a round join panicked inside
linesweeper (Left == Ish). The panic is now a Degenerate error and the art is
left unchanged. Adds hostile-number tests for the pathops edit, stroke and
offset functions.
A merge found the contours of the faces it unites with one contour pass
over the whole arrangement for each coverage among them, and the search
for slivers next to them did the same for each neighboring coverage.
Each in-place remnant of a merge is such a merge too, so a click on
many overlapping shapes made one pass over the whole arrangement for
each coverage, for the merged faces and again for each remnant. Both
now use the batching that Divide's regions use: one pass for each batch
of coverages whose faces share no vertex. The batching looks a face's
coverage up by a hash of the inputs that cover it, computed as it walks
them. The search for slivers looks up the merged faces' coverages the
same way.
Merging a frame no longer takes in an island of the same coverage in
its hole: each face is its outer contour and the holes directly in it,
as the regions are.
Boxes that touch at -0.0/+0.0 now count as overlapping, so the pair cap cannot be skipped. try_boolean_n, try_unite_all and try_area report a refused input, and the Pathfinder shape-mode unite and the Blob Brush use them to show an error. When the planar map is over its budget, the fallback leaves out paths that fill nothing so a flood of cutting lines cannot take the filled shapes down with it.
CodeRabbit findings #6, #19, #53, #59, #62, #167:
- Before a sweep builds a planar map (Boolean, normalize, Pathfinder, Offset Path, Live
Paint, Shape Builder), the pairs of segments whose bounds overlap are counted exactly in
O(n log n); past 4M the operation fails as `PathOpsError::TooComplex` (a crossing can only
lie where two segments' bounds overlap).
- Live Paint's winding crossings are sparse: the sweep keeps one per edge, and a vector as
long as the input count made disjoint shapes quadratic in memory.
- Gap detection groups the segments near a loose end by track once.
- 3D: ring containment checks bounds first and walks at most 50M edges; Inflate's mixed-cell
booleans take at most 4M outline segments together.
Trim, Merge and Crop walk the arrangement once per input shape, and every walk tested each edge
side with a coverage mask built from a winding vector with one slot per input. On 300 paths,
against main:
- Trim: 362 → 31 ms
- Merge: 366 → 31 ms
- Crop: 303 → 28 ms
- Unite, Intersect, Exclude, Minus Front and Minus Back: unchanged or a little faster
- Multi, the arrangement's winding, keeps only the non-zero windings as sorted (input, winding)
pairs, added in place. A face is covered by a few inputs, so a test reads a few entries
instead of one per input. The winding arithmetic is unchanged.
- Arrangement::covered lists the inputs that cover a face (the true entries of its coverage
mask). Trim, Merge, Crop, Minus Front, Minus Back, Intersect and Exclude test that list; each
test's old mask form is in a comment beside it. mask() still builds the same mask for the rest.
pathfinder_pinned hashes the output of all ten operations on 2, 5, 12 and 30 overlapping
ellipses, rectangles and stars (even-odd and non-zero, shared keys for Merge). The 40 hashes were
computed with main's dense windings, and the new code gives every one of them. A wrong Crop test
fails it. The pathops suite (103 tests) passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VmMmc1J8XeNLxhRihWv9cc
The sweep under Offset Path (linesweeper 0.4) never finished on some
miter offsets: on the reported glyph and on a synthetic outline it kept
allocating until the app ran out of memory. linesweeper 0.5 fixes that
infinite loop ("Fixed an infinite loop", 0.5.0 changelog), and builds
without changes.
0.5 can hand back a curve piece that doesn't follow the input curve it
was cut from. In the Shape Builder rosette test, a contour runs out along
one petal and back by such a piece, which became a 59 x 4 pt hole in the
merged shape. Tidy::repair rebuilds a piece that strays from the one input
curve through both its ends as that curve's exact sub-curve (input curves
are indexed by their bounds' left edge, so it stays cheap). The there-and-
back then encloses nothing and is dropped as a sliver. Pieces on their
curve, and pieces whose ends two input curves pass through (a real lens),
are left alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Live Paint Bucket now fills areas enclosed by any paths. Crossing
lines enclose faces it can fill, a line across a shape splits that shape's
face in two, and an area enclosed by paths becomes a face even when nothing
fills it (it starts unfilled). Edges are the paths cut wherever another path
meets them, keeping their stroke, so Shift-click paints each piece. Closed
shapes give the same faces as before.
Root cause: Live Paint built its faces from the Pathfinder regions, which
divide the filled areas of closed shapes. An open path counted only as the
area it would fill if closed, which for a straight line is nothing, so lines
divided nothing and the bucket found no face to fill. A new planar map in
vectorcraft-pathops (live_paint) sweeps every path as an edge: an open path
goes in traced there and back, with crossing counts that never cancel; faces
are walked on the resulting graph (holes included), and their fills and the
edges' sources are found geometrically. The sweep runs in an exactly sheared
frame, because it mishandles a horizontal line traced there and back.
Make, Merge, the bucket and Live Paint Selection all use it;
interior_point moves to vectorcraft-pathops with it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The sweep compares a curve with another segment over a y-span shorter than
its tolerance by bounding boxes only, so an edge rising by less than that
tolerance (e.g. a rectangle edge cut through a circle's cap, whose computed
intersection end sits ~1e-12 off the horizontal) was mis-ordered and the
result lost a corner: xor/difference/area disagreed by up to 10%. Every
sweep input now snaps on-curve y values within the tolerance to one level
(handles follow and are levelled with their anchor), so such edges become
exact horizontals, which the sweep handles exactly.
- tests/regressions.rs: the six minimal failures found by the properties,
the root cause in isolation, and the self-crossing "star" polygon.
- The star-polygon strategy rejected no angular step of half a turn or
more, so it sometimes drew self-crossing polygons (the shoelace failure).
- Property suites run from a fixed seed (PROPTEST_CASES / PROPTEST_RNG_SEED
still override), so CI cannot fail at random.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dotCooCoo