Improve sweep-line triangulation

Author: Agapanthus

.gitignore

@@ -1,4 +1,8 @@
 /target
 /target-bin
 bench_results.jl
+bench_results.jl.back
 graph.png
+flamegraph.svg
+*.old
+*.data

Cargo.lock

@@ -10,7 +10,7 @@ license = "MIT OR Apache-2.0"
 readme = "README.md"
 repository = "https://github.com/bevy-procedural/modelling"
 documentation = "https://docs.rs/procedural_modelling"
-rust-version = "1.81.0"
+rust-version = "1.80.0"
 include = [
     "src/**/*",
     "doc/**/*",
@@ -61,6 +61,7 @@ ab_glyph = { version = "0.2.29", optional = true }
 nalgebra = { version = "0.33.0", optional = true }
 num-traits = "0.2.19"
 criterion = { version = "0.5.1", features = ["html_reports"], optional = true }
+web-sys = "0.3.72"
 
 [features]
 default = ["nalgebra", "netsci", "fonts"]
@@ -105,3 +106,7 @@ inherits = "release"
 opt-level = "z"
 lto = "fat"
 codegen-units = 1
+
+[profile.profiling]
+inherits = "release"
+debug = true

Cargo.toml

@@ -213,25 +213,10 @@ The package supports different triangulation algorithms. The robustness and rend
 -   **Heuristic** Heuristic algorithm that tries to find a compromise between the speed of `Sweep` and the quality of `MinWeight`.
 -   **Auto** (default) Automatically choose the "best" algorithm based on the input. The edge-weight will be the same as Delaunay or better. Uses specialized fast implementations for small polygons to quickly generate min-weight triangulations. Falls back to Delaunay for larger polygons.
 
-<!--
-| Algorithm    | Requirements | Worst Case | Circle 10        | Circle 100         | Circle 1000       | Circle 10000      | ZigZag 1000        | ZigZag 10000      |
-| ------------ | ------------ | ---------- | ---------------- | ------------------ | ----------------- | ----------------- | ------------------ | ----------------- |
-| Fan          | Convex       | $n$        | 0.258µs (195fps) | 2.419µs¹ (154fps)² | 71.0µs (52.4fps)  | 161.8µs (15.7fps) | -                  | -                 |
-| EarClipping  | Simple       | $n^2$      | 0.746µs (196fps) | 21.75µs (155fps)   | 1.746ms (70.8fps) | 3.276s (15.9fps)  | 49.10ms (77.1fps)  | 46.03s (17.3fps)  |
-| Sweep        | None         | $n \log n$ | 1.584µs (196fps) | 13.58µs (161fps)   | 142.4µs (73.4fps) | 1.556ms (15.6fps) | 402.3µs (77.3fps)  | 4.334ms (17.2fps) |
-| Delaunay     | Simple       | $n \log n$ | 2.778µs (194fps) | 29.89µs (178fps)   | 308.5µs (178fps)  | 3.296ms (172fps)  | 3.002ms (77.0fps)  | 158.7ms (17.2fps) |
-| EdgeFlip     | Simple       | $n^3$      |                  |                    |                   |                   |                    |
-| SweepDynamic | Simple       | $n^3$      | 4.087µs (196fps) | 2.320ms (181fps)   | 1.817s (177fps)   |                   | 684.74µs (77.3fps) | 7.550ms (17.2fps) |
-| MinWeight    | Simple       | $n^3$      | 4.087µs (196fps) | 2.320ms (181fps)   | 1.817s (177fps)   |                   |                    |                   |
-| Heuristic    | Simple       | $n \log n$ |                  |                    |                   |                   |                    |
-| Auto         | Simple       | $n \log n$ |                  |                    |                   |                   |                    |
- -->
-
 <img src="assets/fps_vs_render.svg" alt="FPS Boxplot" width="800"/>
 
 -   Time for the triangulation on a Intel i7-12700K (single threaded). Run the benchmarks using `cargo bench --features benchmarks --profile release`.
 -   FPS when rendering 100 large, transparent instances with the bevy 0.14.2 pbr shader on a Nvidia GeForce RTX 4060 Ti in Full HD. See `cargo run -p fps_bench --profile release` and `julia --project=./playground/fps_bench/FPSBench ./playground/fps_bench/FPSBench/src/main.jl`. For the non-Delaunay algorithms, the rendering time deteriorates for the larger circles since the edge length is not minimized causing significant overdraw.
--   The `Delaunay` algorithm is currently significantly slowed down due to a inefficient algorithm for the interior detection. This will be fixed in the future.
 
 ## Supported Bevy Versions