Skip to content

Baseplates: a bowtie or puzzle key put in the wall from above leaves a pair of folds or four-use edges at some magnet sizes #89

Open

Description

@Oliver-Johnson

Found by the sweep for #85 (PR #88). All three are on main (c0dc948), on a solid floor, with the key put in the wall from above. #88 does not touch them (same bytes on both trees).

What happens. At these sizes a piece builds with a pair of faces folded back on each other (or two edges used four times) next to a key cup, and the page gives no warning: Checks is empty and Download stays on. The page's own manifold check only counts edges, so it does not see folds.

Repro (Baseplates page, two 2 x 2 pieces, default settings except as in the link; counts are per piece, folds as test/orientation.js counts them, bad edges as the page counts them):

Link Result
#pi=37.5&w=150&d=75&mm=custom&ml=0&mr=0&mf=0&mb=0&bw=90&bd=400&cn=bowtie&km=wall&ki=top&mg=1&md=7.8&to=tight B1: 2 edges used four times
#pi=39.5&w=158&d=79&mm=custom&ml=0&mr=0&mf=0&mb=0&bw=95&bd=400&cn=puzzlekey&km=wall&ki=top&mg=1&md=11.2&to=loose B1: 2 folds (11.2 mm is the magnet cap there)
#pi=42.5&w=170&d=85&mm=custom&ml=0&mr=0&mf=0&mb=0&bw=100&bd=400&cn=puzzlekey&km=wall&ki=top&mg=1&md=5.8&to=loose A1: 2 folds

The first reproduces as 2 bad edges on B1, the others as 0 bad edges and 2 folds. Neighbouring sizes and the other tolerances were not swept for these; the sweep that found them covered wall keys from above at three tolerances, 30 to 46 mm every 0.5 mm, magnets from 5 mm to the cap.

Possible cause. The same kind as #85: a magnet pocket's plane lands within microns of a face near the key cup, and the seam repair leaves a split a few microns off its face. #88's re-cut of a cell's pockets closes wall-key cups that come out open, but these come out closed with a fold, which that path leaves alone on purpose (re-cutting folded-but-closed housings changed the bytes of designs that fold nothing in the finished piece).

Possible fix. Find which faces fold, then either re-cut when a housing folds and the re-cut is better in the mesh (the measure to use is test/orientation.js, not the STL read-back), or find the near-coincidence and keep the faces apart. Add audit rows for the three links.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

bugSomething isn't working

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions