v2.4.1

v2.4.1

Eight merges since v2.4.0. The headline is infrastructural, and it is worth stating plainly: CI had never once passed in this repository's history. 32 failed runs, 1 cancelled, 0 successes — and GitHub Actions had been disabled repo-wide since January. Seven months of merges landed with no automated gating at all.

That is fixed, and it paid for itself on the first real run.

CI now actually gates the test suite (#334)

The python-tests job was a no-op in four independent ways:

• Every gate ended in || echo "... not configured yet" — black, mypy, pylint and pytest all reported success regardless of outcome.
• pytest python/ pointed at the source tree. An explicit path argument overrides pytest.ini's testpaths, so it collected 7 tests from a stray file while the real suite never ran.
• No JRE was installed, so the freerouting suites would have failed environmentally.
• Actions was off, so none of it executed anyway.

All 1551 Python and 63 TypeScript tests now gate every push, across Python 3.9–3.12 on two Ubuntu versions, TypeScript on four OSes, and integration tests against KiCAD 8/9/10.

Format and type checks deliberately run once rather than per matrix entry: pip install black resolves a different build per interpreter (black ≥25.12 requires Python ≥3.10, so the 3.9 runner silently gets 25.1.0 and disagrees with 26.x on three files).

Three tools that were registered but had no backend now work

assign_net_to_class, check_clearance (#315) and set_layer_constraints (#323) each had a full schema and a router entry but no dispatch handler, so every call returned Unknown command. Found by a documentation-coverage audit rather than by anyone using them — thanks @boovaragan.

Per-layer constraints are written to a project-scoped .kicad_dru custom-rules file, which kicad-cli pcb drc and the GUI both auto-discover; there is no pcbnew API for them.

Silent failures removed

• create_zone was broken on the IPC backend — it assigned kipy's read-only Zone.fill_mode, raising AttributeError on every call. Invisible locally because kipy is absent from a bare dev environment, so mypy typed it Any. Caught by the new CI gate.
• autoroute was abandoned at 30 s (#251) while Freerouting was still running, reporting failure against a valid .ses that existed on disk. Its timeout now derives from the timeout and attempts you pass, rather than a fixed ceiling — timeout is per-attempt and best-of-N multiplies it.
• get_board_2d_view produced no file at all (#235) when no layers were given, because KiCad 9+ refuses pcb export svg without --layers. Fix authored by @stefangordon.
• replace_schematic_component corrupted lib_symbols while restoring fields, and delete_schematic_net_label silently removed the first match on an ambiguous bare-name delete (#332). Thanks @Euter2.

New part-sourcing tools

• Open parts registry (#329, closes #297): search_parts_registry, get_registry_part, download_registry_part — reuse a verified existing footprint or symbol instead of generating one. Downloads are host-allowlisted, extension-checked and capped at 50 MB. Thanks @mingyo186 for both the request and the implementation.
• Live JLCPCB lookup (#326): get_jlcpcb_part returns real-time stock and tiered pricing when Open Platform credentials are configured, falling back to the local snapshot with a source field showing which path answered. Thanks @ravishivt.

One note for anyone tracking assign_net_to_class

As first written it did not work on any real project: a freshly written .kicad_pro carries "netclass_assignments": null, and setdefault returns that None rather than replacing it, so the write raised and degraded to a silent no-op. It was caught by a round-trip test across the boundary this shares with #302's DSN exporter — two features, one on-disk key, and both sides had green suites that never crossed it. That seam now has a permanent test.

Issues closed

#235, #251, #297

Full details in CHANGELOG.md.