test: pin net-label placement defects per bug report#720
Open
DPS0340 wants to merge 1 commit into
Open
Conversation
Adds helpers that count net labels rendered inside chip bodies (tscircuit#655) and labels overlapping each other, plus a test pinning the current count for every imported bug report. The inside-chip check mirrors CHIP_COLLISION_MIN_OVERLAP_FRACTION from NetLabelNetLabelCollisionSolver: a label counts as inside a chip only once the chip covers at least half of it. Counting any interior overlap instead reports incidental 1-6% grazes that the solver tolerates deliberately, so the metric would fight the implementation rather than guard it. Corpus state on main: 8 labels inside chips, 7 label-label overlaps. These are current values, not targets — several are open bugs. Verified the guard bites: changing the expected insideChips for bug-report-20260707T230831Z from 4 to 3 fails with a visible diff.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This was referenced Jul 25, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Same pattern as #718, applied to net-label placement instead of trace crossings.
What it measures
Two defects, per imported bug report:
Corpus state on
main: 8 labels inside chips, 7 overlapping pairs.The threshold matters, and I got it wrong first
My initial version counted any interior overlap. That reported 17 in-chip labels — but when I looked at what it was flagging, most were incidental grazes:
NetLabelNetLabelCollisionSolverhasCHIP_COLLISION_MIN_OVERLAP_FRACTION = 0.5and tolerates smaller overlaps deliberately — its comment says relocating them can violate orientation preferences. So a metric counting every graze would report defects the solver is intentionally leaving alone, and would fight the implementation rather than guard it.The helper now mirrors that constant. With it, the count is 8 rather than 17, and every remaining one is a genuine ≥50% cover.
Why pin it
These boards are verified by SVG snapshots today. A label sliding under a chip is a few pixels in a large regenerated SVG and easy to approve by accident. Pinning means a placement regression fails an assertion that names the board.
The numbers are current state, not targets. Several are open bugs and should come down — #699 already takes
bug-report-20260707T230831Zfrom 4 to 0.There's also a guard asserting the pinned set matches what's on disk, so a newly imported bug report can't go unmeasured.
Verification
insideChipsforbug-report-20260707T230831Zfrom 4 to 3 fails with a visible diff. A pinning test that can't fail is worse than none.tsc --noEmitclean, biome clean.netLabelNetLabelCollisionSolver.getOutput().netLabelPlacements— the output copy. The solver also exposesnetLabelPlacements, which is its input; reading that makes any fix look like a no-op.Diff: 2 new files, no existing file touched. Independent of #718 — either can merge first.