[LorbusChris]

Change

Fixes #11034. An edge flange picked on one segment of a stepped base outline ran the base face's whole length: a 20 mm flange on a left edge that steps out 2 mm between Y 10.5 and 388.3 bent over 403.5 (Y 5.4–408.9), not the 377.8 of the segment.

The cause: add_flanges_each trims the picked edge back to the bend line, and on a stepped outline that line is where the base's own edges already run beyond the step. The trimmed edge merged with them, and the flange took the merged edge. Now, before trimming, the flange notes where the base's outline already runs along the bend line on either side of the picked edge (SheetBody::bend_line_runs), and keeps off those stretches afterwards. So it spans the segment picked, with no relief there: the base is already back at the bend line. A flange on another flange's edge, or on an outline without steps, is unchanged, and so is the outline made by cutting notches into a rectangle, which gave the right flange already.

Changes a default: no

Validation

• New engine test in crates/engine/src/cmd/sheet_tests.rs, a_flange_on_part_of_a_stepped_outline_spans_that_part: the outline from the issue, 1 mm thick (R 1, K 0.44), a 20 mm flange on the X −108.9 segment. Expected: one bend 377.8 long from Y 10.5 to 388.3, and a flat area of the trimmed base (215.8 × 403.5) plus the bend allowance and the 18 mm leg along 377.8, to 1e-9: no relief cut. Without this change the test fails with one bend 403.5 long, Y 5.4–408.9.

Run on Linux x86_64 (Fedora 44, Rust 1.99.1), --profile ci, on this branch on main at da50117:

• cargo xtask ci: fmt, assets, icons, licences, lists, skills, layering, UI containers, clippy --workspace --all-targets -D warnings, the wasm check and the doc tests pass. 81 of the 84 test binaries pass, among them solidcraft-engine lib (3191 passed, 0 failed) and solidcraft-doc lib (530 passed). cargo xtask ci runs the test binaries at once (a load average above 400 on this machine's 24 threads), so timing-sensitive tests can fail inside the full run and pass alone. The three that fail have failures unrelated to this change:

  • solidcraft-gpu lib: 29 of 165 tests, all path-tracer tests on this machine's AMD Radeon 890M (RADV). On main at da50117, 29 fail as well.
  • solidcraft-ui-egui lib: 4132 passed and 2 failed. render_workspace::tests::progressive_image_controls_route_to_the_registered_host is a path-traced render ("inconsistent glass medium paths") and fails on main too. op_auto::tests::primitives_and_revolves_join_cut_or_stand_apart failed only inside the full run; run alone, it passes on this branch and on main.
  • solidcraft-ui-egui test scenarios: 146 passed and 1 failed, s15_start_page_sample_save_reopen ("step 3: timed out: start page open"). Run alone, it passes on this branch and on main.
  • The web offline test wasn't reached. With no Chrome on this machine, it would have been skipped anyway.

• Rebased onto main at 29a6bbd, whose new commits include Sheet metal: a fold crosses slanted or curved edges of any panel (#10621) #10922 (a fold across slanted or curved edges, in crates/doc/src/sheet.rs); the patch is unchanged. cargo fmt --check and clippy -p solidcraft-doc -p solidcraft-engine --all-targets -D warnings pass there; cargo test -p solidcraft-doc --lib: 538 passed, and -p solidcraft-engine --lib flange sheet fold: 271 passed, 0 failed.

• Reproduction with solidcraft-cli run, built from main at da50117 and from this branch, the issue's two scripts; the flat pattern's bends:

  Script	main	this branch
  stepped outline, flange on the X −108.9 segment	one bend 403.5 long, Y 5.4–408.9	one bend 377.8 long, Y 10.5–388.3
  rectangle with the steps cut away as notches	one bend 377.8 long, Y 10.5–388.3	the same

Screenshots of the behavior (when observable)

The issue's stepped outline with a 20 mm flange picked on the stepped segment, in 3D and in the flat pattern.

Before (main): the flange runs the base face's whole length, past both ends of the segment; in the flat pattern its bend line spans the whole 403.5.

After: the flange spans the segment, Y 10.5–388.3: in 3D it stops 5.1 short of the near corner and 20.6 short of the far one, and in the flat pattern the base's outline steps back at both ends of the bend.

Captured from desktop builds (--profile ci) of main at da50117 and of this branch on Linux x86_64 under Xvfb, driven through the control channel (engine.execute, ui.screenshot).

Integration status

Rebased on main at 29a6bbd (2026-11-02); no conflicts. The full run above was at da50117. Open PRs changing the same function, add_flanges_each in crates/doc/src/sheet.rs, each add a value to the same found tuple and its two loops, so whichever lands second has a small mechanical conflict there:

• Edge Flange: start and end set-backs for a flange shorter than its edge (#3516) #4340 (edge flange set-backs): its notched panel flag.
• Sheet metal: an edge flange runs along the edge a cut left (#2009) #7278 (an edge flange runs along the edge a cut left): its notch flag.

Other open sheet metal PRs (#7652, #10356, #9966, #9714, #6855, #8512, #7582) merge cleanly alongside it. #5943 (partial flanges) asks for set-backs and more; this only makes a flange on a stepped outline's segment stay on that segment.

Assisted-by: LLM