What happens
On narrow-body quad variants the corner silkscreen marks are drawn on top of the pads, not near them:
fp.string("tqfp32_w7").circuitJson()
// pcb_silkscreen_path_top-left route: [[-2.4, 3.5], [-3.5, 3.5], [-3.5, 2.4]]
// pad 32: x[-3.075, -2.525] y[3.400, 4.875]The horizontal arm runs from x = -3.5 to x = -2.4 at y = 3.5. Pad 32 occupies x[-3.075, -2.525], and y = 3.5 is inside its y[3.400, 4.875] span — so the silkscreen passes straight through copper. Measuring segment-to-pad distance minus half the stroke width:
| footprint |
min silk-to-pad clearance |
tqfp32_w7 |
−0.050 mm (overlap) |
tqfp48_w7 |
−0.050 mm (overlap) |
tqfp32 (default) |
0.575 mm |
lqfp32 |
0.450 mm |
qfn32 |
0.583 mm |
Cause
In src/fn/quad.ts the corner is placed at (±w/2, ±h/2) — the pad-row centre line — and each arm is a fixed csz = pw * 2 long:
const corner_x = (parameters.w / 2) * dx
const corner_y = (parameters.h / 2) * dy
const csz = parameters.pw * 2
route: [{ x: corner_x - csz * dx, y: corner_y }, { x: corner_x, y: corner_y }, ...]Because the corner sits on the centre line, an arm running inward along one axis always passes through the band the outer pads occupy on the perpendicular side. It only escapes when the arm happens to be short enough to stop before the outermost pad — which is why the default tqfp32 (body 10 mm, arm ends at x = -3.7, outermost pad starts at x = -3.075) looks fine while the 7 mm-body variant does not. The arm length never consults the pads.
KiCad reference
KiCad's TQFP-32_7x7mm_P0.8mm puts the corner at (-3.61, 3.61) with arms only 0.275 mm long, leaving exactly 0.200 mm to the nearest pad. Same for TQFP-48, TSSOP-10, SOT-457T — all measure exactly 0.200 mm, i.e. IPC silk-to-pad clearance.
Why no test catches it
tests/kicad-parity/tqfp32_kicad_parity.test.ts compares copper via boolean difference and asserts courtyardDiffPercent. Both are blind to silkscreen, so it reports Diff: 0.00% while the silkscreen sits on the pads. This is the same blind spot as #732 — I found this one by extending that scan across all 90 parity footprints.
Other findings from the same scan
Reporting these separately so they can be triaged independently — this issue is only about the quad corners:
Happy to submit PRs for the rest if the approach in the quad fix looks right.
What happens
On narrow-body quad variants the corner silkscreen marks are drawn on top of the pads, not near them:
The horizontal arm runs from
x = -3.5tox = -2.4aty = 3.5. Pad 32 occupiesx[-3.075, -2.525], andy = 3.5is inside itsy[3.400, 4.875]span — so the silkscreen passes straight through copper. Measuring segment-to-pad distance minus half the stroke width:tqfp32_w7tqfp48_w7tqfp32(default)lqfp32qfn32Cause
In
src/fn/quad.tsthe corner is placed at(±w/2, ±h/2)— the pad-row centre line — and each arm is a fixedcsz = pw * 2long:Because the corner sits on the centre line, an arm running inward along one axis always passes through the band the outer pads occupy on the perpendicular side. It only escapes when the arm happens to be short enough to stop before the outermost pad — which is why the default
tqfp32(body 10 mm, arm ends atx = -3.7, outermost pad starts atx = -3.075) looks fine while the 7 mm-body variant does not. The arm length never consults the pads.KiCad reference
KiCad's
TQFP-32_7x7mm_P0.8mmputs the corner at(-3.61, 3.61)with arms only 0.275 mm long, leaving exactly 0.200 mm to the nearest pad. Same forTQFP-48,TSSOP-10,SOT-457T— all measure exactly 0.200 mm, i.e. IPC silk-to-pad clearance.Why no test catches it
tests/kicad-parity/tqfp32_kicad_parity.test.tscompares copper via boolean difference and assertscourtyardDiffPercent. Both are blind to silkscreen, so it reports Diff: 0.00% while the silkscreen sits on the pads. This is the same blind spot as #732 — I found this one by extending that scan across all 90 parity footprints.Other findings from the same scan
Reporting these separately so they can be triaged independently — this issue is only about the quad corners:
sod123andsot723emit no silkscreen at all where KiCad has 3 paths each (sot23was sot23 (3-pin) generates no silkscreen body outline or pin 1 marker, unlike every sibling SOT #732/fix: draw the sot23 body outline and pin 1 marker (#732) #733).tssop10_w3mm_p0.5mm(−0.025 mm) andsot457_w2_pl1.5_pw0.6(−0.005 mm) also overlap, from different code paths.0402,0603,soic*_legsoutside, …) sit between 0.045 mm and 0.15 mm — tight versus IPC's 0.2 mm but not overlapping.Happy to submit PRs for the rest if the approach in the quad fix looks right.