Rotation check: repeated pads, vendor vs KiCad pad numbering, alternate land patterns

Colby Knox · ·closed by Colby Knox

Split out of #1 (John Lauer, 2026-10-19). #1's own fix, keeping vendor access failures as failed verification, shipped in 0.1.1.

What's left. On the five-board electronic-load quote, full EasyEDA retrieval gave twelve pad-fit discrepancies. One was a real stale BOM identity (J10), since fixed. The other eleven are probably cases the rotation check can't tell apart from a wrong footprint:

  1. Repeated physical pads. Several pads share a number: thermal pads split into a grid, or mounting tabs.
  2. Vendor versus electrical terminal numbering. EasyEDA's pad numbers aren't KiCad's, for example a connector's shield or mechanical pads.
  3. Alternate land patterns. The vendor footprint is a valid alternative land with different pad dimensions, so the centres don't line up exactly.

Today rotation (src/rotation.rs, about line 214) fails every such part with "review terminal numbering, repeated pads, land dimensions, footprint identity and orientation". That's correct but unspecific, so all eleven need a manual drawing review.

Wanted. Checker support for each case that still never relaxes the tolerance:

  • match repeated pads as a set;
  • map vendor to KiCad numbering, from an explicit per-part mapping or by matching pad geometry, and record which;
  • say when a land is a dimensionally different alternative, not a rotation error.

Each part's result should say which case applied. Anything still unexplained stays a failure for manufacturer-drawing review.

Done when: the electronic-load board's eleven remaining parts are each classified (fits after the case is handled, or a real mismatch), with tests for each case and no tolerance change.

1 Reply

Colby Knox marked this in progress ·
Colby Knox linked pull request #4, which closes this when it merges ·
Colby Knox closed this in pull request #4 ·
Colby Knox ·

This is in adom-aiflow-jlcpcb 0.1.6. The rotation check now names how each part was matched (case in the record), and the tolerance is unchanged:

  • repeated-pads-as-set: every pad is kept now (before, one pad per number was), so a split thermal pad or mounting tabs are matched as a set.
  • explicit-pin-map: a per-part map in the overrides, {"<ref or footprint>": {"pins": {"<KiCad pad>": "<JLCPCB pad>"}}}. The map used goes in the record.
  • geometry-pin-map: the numbers differ, but the pad geometry lands at exactly one rotation, or the numbers decide between the ones that do. The implied map goes in the record.
  • alternate-land: the same pad arrangement at another size. The record says it's a dimensionally different land, not a rotation error, takes the rotation from the arrangement, and keeps the part flagged for the drawing review. It's never a pass.
  • unexplained: still a failure for the manufacturer-drawing review. A symmetric part whose numbers can't decide its orientation lands here, with a note asking for a pin map.

Each case has a test. I haven't re-run the electronic-load board's eleven parts, because their saved vendor responses aren't here. Running rotation on that board with 0.1.6 will classify each one.

Upgrade with adom-wiki pkg install adom/adom-aiflow-jlcpcb; adom-aiflow-jlcpcb --version then shows 0.1.6.

Log in to reply.