Skip to content

STATUS.md

Latest commit

 

History

History
204 lines (160 loc) · 11.1 KB

STATUS.md

File metadata and controls

204 lines (160 loc) · 11.1 KB
Raw

CoreLibrary status

Last updated: 2026-09-05.

This is the only dated file in the repository. It is rewritten, never appended — there are no ## 2026-xx-xx update sections here and none should be added. If you want to know what changed and when, that is what git log is for. Everything durable — how to author content, what the gates mean, how to release — lives in README.md, CONTRIBUTING.md, docs/AUTHORING.md and docs/3d-placement-convention.md, and none of those carry dates.

Negative knowledge, recorded once so nobody spends an afternoon on it: the original CoreLibrary expansion spec and the ~/.claude/plans/* files that several older documents cited as "source of truth" are gone from disk. They are not recoverable and they are not in git. Don't go looking for them.

Inventory

241 components · 237 symbols · 168 footprints · 158 3D models. Ten footprints are no3d.

By category: ic 75 · connector 42 · transistor 32 · power 23 · diode 20 · passive 13 · mechanical 9 · opto 7 · sensor 7 · crystal 5 · switch 3 · relay 2 · battery 2 · audio 1.

Work in flight is on branch feat/corelib-hardening, uncommitted (the hardening pass, see below). master is unchanged since the docs consolidation.

Last verified gate run

Verified green on feat/corelib-hardening with bun run shared:status reporting all eight packages linked to ../shared — deliberately, because the pass depends on shared changes that are not yet tagged (see "The held release"). CI, which resolves the pinned tags, stays red until those tags exist.

Gate Result
bun run typecheck clean
bun test 186 pass
bun tools/validate.ts --release --strict OK, 241 components (G1–G12)
bun tools/audit-3d-placement.ts --release 158 ok / 0 errors / 0 warnings
bun tools/audit-components.ts --no-render 241 / 0 issues
bun tools/check-datasheet-links.ts structural, 180/180 OK
bun tools/pack.ts --version=0.0.0-dev builds (full pack)
… --no-step --out=dist/glb-only builds (GLB-only core + STEP companion)
../shared kicad-parsers, kicad-import bun test 51 + 52 pass

The hardening pass, in one paragraph

Every component was audited as if it were going into a product: symbol pin order against the manufacturer datasheet of the stated MPN in the stated package, package against footprint, datasheet against manufacturer, and sourcing. The per-part record, including what could not be verified and why, is docs/audit/2026-09-hardening-findings.md. Defects fixed: SS8050/SS8550 base–emitter swap in TO-92 (per-footprint pin map, SOT-23 variant added), BAV99 pin names, RP2040 on the wrong QFN-56 exposed-pad variant, DS3231 and PCF8574 on the narrow SOIC-16 instead of the 7.5 mm body, ATtiny85 on the JEDEC instead of the EIAJ SOIC-8, ESP32-WROOM-32E flash pins exposed as signals, LM741 identity, 1N4148/1N5819/SS24/SS34 package mismatches, slotted drills and thermal-pad paste dropped by the importer, and ~40 datasheets that belonged to another manufacturer. Every function part now carries parameters, a verified primary MPN with LCSC code where stocked, alternates in the same package, and an official datasheet. Seven gates (G6–G12) enforce all of that from now on; CONTRIBUTING.md explains each.

The held release

The release is still deliberately held, pending an explicit go-ahead. The desktop app bundles a seventeen-component pack built on 2026-06-02; none of the content grind has reached a user.

The unblock sequence, in order:

  1. Commit the uncommitted ../shared change on main: kicad-parsers parses (drill oval …) into drillSlot; kicad-import emits drillSlotMm, keeps paste-only sub-pads as role: "paste", and counts only numbered pads in padCount (a semantic change consumers of InspectFootprintItem.padCount will see). Tests are in both packages.
  2. Cut four tags in shared: step-to-glb-v0.1.5, kicad-import-v0.2.0 (now also carrying step 1), rendering-core-v0.1.4, contracts-v0.3.1. The commits for all four are already on main; none of the tags exist.
  3. Re-pin CoreLibrary/package.json (kicad-import → 0.2.0, step-to-glb → 0.1.5) and OpenPCB/package.json. Push. Confirm CI green.
  4. Only then tag v0.2.0, and re-pin the app bundle.

Signing prerequisites, still outstanding

release.yml fails closed, so v0.2.0 will not build until both of these exist:

gh secret set OPCLIB_SIGNING_KEY < keys/openpcb-core.priv.pem
gh variable set OPCLIB_KEY_ID --body openpcb-core-2026

Then delete the local .priv.pem. See keys/README.md.

Release checklist residue

  • Verify the signature and SHA256SUMS on the published artifacts.
  • Bump the app's bundled library via OpenPCB/scripts/fetch-core-library.ts and verify the boot import in a packaged build, not a dev run.
  • App smoke test once the pack lands: place an LM324 (four separate op-amps, no OUTPUT_OUTPUT_SHORT); place SS8050 and check the TO-92 variant lands E/B/C on pads 1/2/3; place the barrel jack and confirm the pads have holes; enable the assembly-view preset and confirm courtyards render.

What is still not verified

Recorded here so nobody re-derives it; details and what was tried are in the audit record.

  • Pin order behind a WAF: the six STM32 LQFP parts (st.com unreachable), AD620, ADUM1201, DS1307, DS3231 (analog.com), ULN2803A, the Molex USB-A / micro-B receptacles, the C&K slide switch, the Panasonic tactile switch, the TDK buzzer, both Keystone holders, the Phoenix screw terminals. Pin counts, packages and KiCad provenance all match; nothing suggests a defect.
  • Bridge rectifier ABS: neither Diotec nor Diodes numbers the terminals, so which pad is + rests on KiCad's convention.
  • RJ45 jack LED polarity (LED1A/LED1K…) and the Sanyou SRD relay pin map were derived from KiCad symbol wiring, not a numbered manufacturer table.
  • CH340C pin 8 (NC vs an output) is disputed between WCH's page and vendor symbols.
  • AMS1117 vin_max is 15 V (Advanced Monolithic absolute maximum) while the new UMW primary specifies 12 V.

Known, accepted exceptions: MT3608's datasheet is hosted by Olimex (Aerosemi publishes none); ME6211 has no manufacturer PDF (an AP2112K alternate carries one); the DB107 bridge datasheet sits on the primary entry because MDD's host is not in the G8 table.

Content backlog, in value order

Target for v1 stays a jellybean set of roughly 250–300 components. Delivered this round: six shielded power-inductor packages on the generic inductor, six Hirose FH12 FFC/FPC sizes, VSON-10, WSON-8 (two bodies), SSOP-20/28 and MSOP-8 with real consumers (TPS63001, TMP1075, W25Q32, ENC28J60, MCP2200, MCP6002). Still open:

  1. 0201, 1812, 2010 passive sizes.
  2. 1.27 mm and 2.00 mm headers.
  3. Power electronics: DPAK / D²PAK / TO-247, gate drivers, Kelvin shunts.
  4. QFN and BGA.
  5. FT232RL and TMP117 were dropped from the DFN/SSOP wave because their KiCad symbols carry no pin for datasheet-NC pads and --strict cannot express "intentionally unmapped". See the tools follow-up below.

Constraint on all of it: every new footprint needs a component that references it, and every new function part must pass G6–G12 before it lands.

Wave-2 blockers — still genuinely open

Part Missing
DHT22 no symbol in KiCad 10
74HC125 only 74LS125 / 74LVC125 / 74AHCT125 exist
ESP-12F only the ESP-12E footprint exists
Solder Jumper Jumper.pretty has footprints but no schematic symbol
JST-XH 3D footprints present, no STEP in Connector_JST.3dshapes
MPU-6050 InvenSense_QFN-24 has no STEP
HC-SR04 no dedicated symbol; generic 4-pin header only
MP1584 / MP2307 no symbol in KiCad 10 stock libraries
EC11 encoder needs external/authored assets

Tools follow-ups

  • unmappedPads in manifests: a way to declare datasheet-NC footprint pads as intentionally unmapped, with a reason, so --strict stops forcing NC pads onto live pins (blocked FT232RL, TMP117; the W25Q32 WSON EP is grounded per its datasheet instead).
  • Provenance refresh: 12 symbols and 14 components carry a sourceHash from an older KiCad checkout and no upstreamCommit. Re-importing them against the pinned 10.0.4 tree changes only the stamps, but do it deliberately and diff the geometry.
  • topology vs type on power parts: twelve manifests still emit topology; the key is now redundant with G6's type. Change manifest and content together.
  • check-datasheet-links does not walk entry-level datasheets (manufacturerParts[].datasheet).
  • ACS712 uses the plain SOIC-8 land; Allegro recommends widened pads for the current path.

Data model

Typed parameter values — {value, unit, min, typ, max} — remain the next data-model step, for real parametric filtering. They require a component.schema.json change plus work in opclib-pack and the app. parameters is now populated and vocabulary-checked (G5, G6) on every function part, so the app can start reading it.

Sourcing fields do not reach the pack. manufacturerParts[] entries carry lcsc, jlcpcbAssemblyType, package and role in source; @openpcb/opclib-pack's packed-manifest schema admits only {manufacturer, mpn}, so tools/pack.ts strips the rest. Relaxing that schema (and reading lcsc into the app's existing lcscPartNumber column) is the follow-up that makes the sourcing data useful in a BOM.

App-side follow-ups

Tracked here because CoreLibrary work created them, though the work itself is in OpenPCB:

  • Render drillSlotMm as a slot (today a 1 × 3 mm slot draws as a 1 mm round hole) and treat role: "paste" pads as paste, not copper (r3f-eda-canvas footprint-render-layer.tsx).
  • Read manufacturerParts[].lcsc into lcscPartNumber / supplier; surface alternates in BOM export; show and search keywords, subcategory, parameters.
  • Per-unit placement UX — U1A / U1B plus physical-part grouping for ERC, PCB and BOM.
  • The courtyardPolygon producer in board-snapshot.ts.
  • opclib-importer.ts: assert there is no render-ref when transformBaked is set.

Deferred decisions

  • Ratify the connector sidecars' scaleMm: {y: -1} (pin headers, JST XH, IDC, and now Hirose FH12) via a contact sheet.
  • shared/step-to-glb: return a post-bake bounding box and an orientation-mismatch warning so the gate can compare against what was actually baked.
  • Whether npn-sot-23-ebc / pnp-sot-23-ebc should stay: they encode a pin order no SOT-23 BJT in the library uses; every real SOT-23 BJT here is B-E-C.

Git

feat/corelib-hardening carries the whole pass uncommitted. ../shared has an uncommitted parser and importer change on main (step 1 of the release sequence). The 2026-07-12 branch feat/corelib-npth-pending is archived as tag archive/corelib-npth-pending; release/v0.1.0-beta is a stale, unrelated branch that predates the content grind and is not the vehicle for v0.2.0.