fix: merge generated net-label connector rails with same-net traces (#34)

[DPS0340]

/claim #34

Fixes the remaining #34 symptom: close parallel same-net rails left unmerged when one of them is a generated net-label connector trace.

Reproduce in an example file (as requested)

On current main, example45 shows the exact symptom from the issue screenshot: the generated SDA net-label connector runs a vertical rail at x=10.42 while the same-net R10.2 -> CONN1.3 trace runs a parallel rail 0.166 apart at x=10.58, with 0.5 of overlapping length — two rails that should be one line.

trace R10.2-CONN1.3               path: (10.15,0.70) -> (10.58,0.70) -> (10.58,0.10) -> (11.02,0.10)
trace available-net-orientation-1-SDA path: (10.37,0.70) -> (10.42,0.70) -> (10.42,0.20) -> (10.32,0.20)

Root cause

The aligning_same_net_rails cleanup pass added in #663 (traceCleanupSolver2) restricts eligibleTraceIds to traces that existed at the first cleanup pass:

eligibleTraceIds: new Set(
  instance.traceCleanupSolver!.getOutput().traces.map((trace) => trace.mspPairId),
)

Net-label connector traces are generated later by AvailableNetOrientationSolver, so they're never eligible for rail alignment — the merging machinery from #663 simply never sees them.

Fix

Add the generated available-net-orientation-* traces from the pre-alignment collision output to the eligible set. The existing rail-group merge (alignSameNetRails) then pulls them onto the shared coordinate — no changes to the alignment algorithm itself. All existing safety checks (obstacle collisions, label anchor preservation, readability scoring) still apply.

Demonstrate fix before/after (as requested)

After the fix, the SDA connector rail merges onto x=10.42 with the same-net trace:

trace R10.2-CONN1.3               path: (10.15,0.70) -> (10.42,0.70) -> (10.42,0.10) -> (11.02,0.10)

Snapshots updated for the three boards whose rails now merge: example45, example49, bug-report-20260707T134549Z. I scanned all example and bug-report boards with a close-parallel-rail detector before/after: example45's finding is eliminated and no new ones appear anywhere.

Verification

• New regression test tests/repros/repro-same-net-rail-merge-34.test.ts asserts no close parallel same-net rails remain in example45 (fails on main)
• bun test — 141 pass, 0 fail
• bunx tsc --noEmit / bunx biome check — clean

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
schematic-trace-solver	Ready	Preview, Comment	Jul 25, 2026 6:13am

[DPS0340]

Following the repo's repro-first workflow: opened #708 as the standalone reproduction PR for the remaining #34 symptom (pins the unmerged SDA/R10.2-CONN1.3 rail pair on example45 with a visual snapshot). This PR is the fix that flips that repro's assertion to .toEqual([]) — happy to rebase once the repro lands.

[DPS0340]

Rebased onto current main (through #714) — CI is green and this is mergeable.

If you're triaging my open PRs on this repo, this is the cheapest one to review first: 5 files, +83/-6, one behavioral change in SchematicTracePipelineSolver.

The regression test is meaningful in both directions:

# on current main
(fail) close parallel same-net rails are merged onto the same coordinate

# on this branch
(pass) close parallel same-net rails are merged onto the same coordinate

Full suite: 145 pass / 0 fail.

For ordering, my open PRs here go smallest-first: #700 (+83/-6) → #707 (+78/-0, characterization only) → #699 (+462/-324) → #703 (+712/-50) → #702 (largest, mostly regenerated snapshots). Happy to split #702 further if the size is what's holding it up — just say the word.

[DPS0340]

@imrishabh18 @mohan-bee — you've both merged in this repo recently, so tagging you rather than leaving this sitting: would one of you be able to review?

This is the smallest of my open PRs here (5 files, +83/-6, one behavioral change in SchematicTracePipelineSolver) and it's rebased onto current main through #714.

The regression test is meaningful both ways:

# current main
(fail) close parallel same-net rails are merged onto the same coordinate
# this branch
(pass) close parallel same-net rails are merged onto the same coordinate

Full suite 145 pass / 0 fail, CI green.

If it's easier to batch, my open PRs here smallest-first are: #700 (+83/-6), #707 (+78, characterization only), #699, #703, #702. Happy to split #702 if its size is the blocker.

[mohan-bee]

LGTM cc @seveibar

[DPS0340]

Thanks for the review @mohan-bee. @seveibar — this one is approved and MERGEABLE / CLEAN with all checks green, ready whenever you have a moment.

Quick summary for the merge decision: it's +83/-6 across 2 files. LongDistancePairSolver was emitting a generated net-label connector rail parallel to an existing same-net trace rail, so the same net got drawn twice a fraction of a millimetre apart. The fix merges the generated rail into the existing one when they're on the same net and same axis. Regression test fails on main, passes here; full suite green.

No rush — flagging it because it's been sitting approved and I didn't want it to get lost in the queue.

[DPS0340]

@mohan-bee gentle nudge — this has been APPROVED and MERGEABLE/CLEAN since 06:37 UTC, about 4½ hours now.

Flagging it because that's unusual for this repo rather than to rush anyone: #694 went approved → merged in 4 minutes, and you merged #712 yourself, so I don't think this is waiting on permissions. It looks more like it just slipped off the list after the approval.

Recap so it's cheap to action:

• +83/-6, one source file plus its test
• Fixes generated net-label connector rails not merging with same-net rails
• The regression test fails on main and passes here
• All checks green, no conflicts

If you'd rather someone else press the button, @MustafaMulla29 has merged most of the recent ones. And if there's something you spotted after approving that you'd like changed, I'm happy to take it — just say so.

[DPS0340]

@MustafaMulla29 @techmannih — one small ask, not a review request for everything I have open.

#700 has been APPROVED and MERGEABLE/CLEAN since 06:37 UTC (~5½ hours). mohan-bee approved it with "LGTM cc @seveibar" and then it seems to have slipped off the list rather than hitting any objection.

For calibration, the usual gap in this repo is minutes — #694 went approved → merged in 4 minutes, and #697/#695 merged with no approval at all. So this one sitting is unusual rather than me being impatient.

It's the smallest thing I have open: +83/-6, one source file plus its test. Fixes generated net-label connector rails not merging with same-net rails; the regression test fails on main and passes on the branch. All checks green.

I know I have a lot of PRs open here and I don't expect them reviewed as a batch — I'd rather land one and let that inform the rest. If any of them are unwelcome, say so and I'll close them; I'd genuinely prefer that to leaving a queue nobody wants.

If there's something you'd like changed on #700 that wasn't caught in the approval, happy to take it.