11 rehearsal: a 4-layer board as if someone else drew it¶
Reference: examples/boards/stm32f103_4l (06), copied without its
board.py, so the only way in was boardwright existing and boardwright sch,
as on a board drawn in KiCad. The change: I2C pull-ups, 4.7 kohm 0402, R10 on
PB6 and R11 on PB7, to +3V3, beside the right-hand header.
Stresses: the whole existing-board loop on four layers, start to finish:
schematic edit, carry it to the board, place, route, text, DRC, verify.
KiCad 10.0.8 judged each step (kicad-cli through the session's wrapper).
What broke
1. Placing by eye. MOVE put R10 and R11 beside J3 on top of PB7 and PB8
tracks and inside SW2's courtyard; apply said nothing; DRC found 16
violations, three of them shorts (#24).
2. Two ROUTES from R11.1 each dropped a +3V3 via at the same point,
0.01 mm apart: hole_to_hole. It came back with a clean placement too (#23).
3. With copper clean, R10's reference sat over J3's silkscreen, and nothing in
the tool could move a footprint's text: verify NOT DONE on 4 silkscreen
findings (#25).
4. Generated boards carry no footprint-to-symbol link; renaming R1 to R12 in
the schematic made update add a new R12 and orphan the routed R1 (#26).
5. MOVE saved through kicad-tools and rewrote the file: a four-part move,
+198 / -5507 lines, nothing lost (#32).
6. The connectivity check the guide names, netspec snap / netspec diff,
could not reach KiCad from the session; the netlists were compared from
KiCad's own export instead. That is the machine's limit, not this tool's.
Tool changes
- existing apply checks every moved part (pads against other nets' copper
and the edge, courtyard against courtyards, real pad shapes) and writes
nothing on a clash; existing free REF [near PAD] lists clear spots.
- Routes change layer on their own net's via or through-hole pad instead of
adding one.
- A MOVE entry can place the part's reference and value text.
- gen links footprints to symbols; existing update links what it matches
by reference.
- MOVE edits only the moved footprints' blocks.
- On the way: previews crashed without rsvg-convert (now SVG), and
update wrote a sub-sheet part's sheetname wrongly.
Result: schematic netlist changed by exactly R10.1/R11.1 on +3V3, R10.2 on
PB6, R11.2 on PB7; ERC 0. After update, free, MOVE, four ROUTES (7 vias,
one shared) and the text move: DRC at every severity 0 violations,
0 unconnected, 0 parity issues; existing verify PASS against the baseline.
Left open
- Text placement is by hand: free finds room for copper and courtyards, not
for silkscreen. The first spot tried for R10's reference overlapped SW2.
- live apply --save saves without refilling: a part moved into a pour
keeps stale fill over its pads until live refill.