fix: make hole circuit json parse against circuit-json schemas (#2843)

[DPS0340]

Closes #2843.

Two independent defects that both make core emit circuit JSON which fails any_circuit_element.parse().

                                          before   after
board-level <hole />                      invalid  valid
board-level <platedhole />                invalid  valid
<chip footprint="dip8" /> plated holes    invalid  valid

1. pcb_component_id: null on board-level holes

PrimitiveComponent initialises ids to null (pcb_component_id: string | null = null), and the holes resolve their owner with ??, which only falls through on undefined. A hole inside a footprint has an owner; one placed directly on the board doesn't, so null reached circuit JSON — and every id on these schemas is z.string().optional(), which rejects null.

Isolated it before fixing, so the diagnosis isn't a guess:

as-is                                     → false
{ ...hole, pcb_component_id: undefined }  → true
omitted entirely                          → true

Fixed with a small optionalId helper at the two sites where the schema actually permits an absent id.

2. circular_hole_with_rect_pad missing hole_shape / pad_shape

Found while verifying the first fix — the test still failed, and the culprit was a different element. That schema variant requires both fields:

var pcb_circular_hole_with_rect_pad = z.object({
  shape: z.literal("circular_hole_with_rect_pad"),
  hole_shape: z.literal("circle"),
  pad_shape: z.literal("rect"),
  ...
})

PlatedHole.ts omitted them at that one insertion site while the neighbouring pill variants in the same file already set them, so I matched the existing pattern. A plain <chip footprint="dip8" /> was enough to trigger this.

Deliberately out of scope

The sweep also found pcb_silkscreen_text and pcb_fabrication_note_text carrying pcb_component_id: null. I tried the same normalisation there and tsc rejected it — those schemas require a non-optional string. That makes it a design question (what should a board-level silkscreen text reference?) rather than a null→undefined fix, so I reverted it and reported it in #2843 instead of guessing.

Verification

The test bites. Reverting Hole.ts + PlatedHole.ts:

Expected: > 0
Received: 0
(fail) board-level holes produce valid circuit json

The test asserts both directions: board-level holes have no id, footprint holes keep theirs (typeof === "string"), and every hole passes any_circuit_element.safeParse. So it can't be satisfied by blanket-clearing the field.

Suite: 1260 pass / 0 fail, tsc --noEmit 0 errors, biome format clean.

One inline snapshot changed — plated-hole-pill-rotated2 had "pcb_component_id": null recorded, i.e. the bug was captured in an expectation. It now reads undefined.

Worth noting how this was found: validating core's own output against circuit-json's exported schemas across several board shapes. It's cheap and it also surfaced pcb_group.anchor_alignment: null and schematic_group.subcircuit_id: null, which I've left for separate triage.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
tscircuit-core-benchmarks	Ready	Preview, Comment	Jul 25, 2026 5:11pm

[DPS0340]

Housekeeping — rebased onto current main (d99c99f), still 1260 pass / 0 fail. No behaviour change.

Since I have several PRs open here now, a note on how they interact so none of this lands on you:

Five of the seven are fully independent — they touch different files and merge in any order:

PR	file
#2838	Inductor.ts
#2844	Hole.ts, PlatedHole.ts
#2846	Group/Group.ts (group insert sites)
#2840	Group/Group.ts (pcb_trace segment insert)
#2842	NormalComponent.ts, Group/Group.ts, Group_doInitialPcbComponentAnchorAlignment.ts

Two pairs conflict, and both are trivial:

1. fix: omit the voltage half of the fuse label when voltageRating is absent (#2833) #2834 ↔ fix: honour schShowRatings={false} on fuse (#2835) #2836 — both edit Fuse._getSchematicSymbolDisplayValue(). Whichever merges first, the other needs the two guards stacked:

   if (this._parsedProps.schShowRatings === false) return undefined   // #2836
   if (voltage === undefined || Number.isNaN(voltage)) return currentDisplay  // #2834
   return `${currentDisplay} / ${formatSiUnit(voltage)}V`

   They're orthogonal (one suppresses the whole label, the other drops just the voltage half), so the combined behaviour is well-defined. Both test files also touch fuse.test.tsx, appending separate tests.

2. fix: carry net highlightColor into pcb_trace.highlight_color (#2839) #2840 ↔ fix: write display_offset_x/y as display strings (#2841) #2842 — both add a field to nearby inserts in Group.ts; the resolution is to keep both lines.

Happy to do the merge myself: say the word and I'll combine any subset into a single PR, or rebase the losers as soon as the first one lands. I'd rather you not spend time on conflict resolution for my changes.

I verified the above rather than assuming it — checked all 21 branch pairs for conflicts and confirmed each rebases cleanly onto d99c99f.