robert/PedalPebble

Watch

Fork 0

Unit tests: snapping on out-and-back routes, off-route hysteresis #39

New issue

Open

opened 2026-08-31 17:14:50 +02:00 by robert · 2 comments

robert commented

2026-08-31 17:14:50 +02:00

(Migrated from git.butzei.de)

Goal

The navigation logic most likely to fail in a way that is confusing on the road.

Acceptance criteria

Snapping tested on an out-and-back route: must not jump to the return leg
Snapping tested on a lollipop route and on a route that crosses itself
Re-acquisition after a GPS gap tested
Off-route enter and exit hysteresis tested at the thresholds
Rejoining further along the route tested
Distance-to-turn and ETA tested against hand-computed values

Files

companion/src/test/

## Goal The navigation logic most likely to fail in a way that is confusing on the road. ## Acceptance criteria - [ ] Snapping tested on an out-and-back route: must not jump to the return leg - [ ] Snapping tested on a lollipop route and on a route that crosses itself - [ ] Re-acquisition after a GPS gap tested - [ ] Off-route enter and exit hysteresis tested at the thresholds - [ ] Rejoining further along the route tested - [ ] Distance-to-turn and ETA tested against hand-computed values ## Files - `companion/src/test/`

robert added a new dependency

2026-09-03 19:01:21 +02:00

#31 Navigation engine: windowed snapping to the route polyline

robert referenced this issue

2026-09-03 19:04:11 +02:00

CI: JVM tests, host C tests and both watchapp builds on every push #67

robert referenced this issue

2026-09-05 17:47:38 +02:00

Navigation engine: windowed snapping to the route polyline #31

robert referenced this issue from a commit

2026-09-05 22:39:11 +02:00

Test audit of #39: lollipop/self-crossing snapping, fix a real self-crossing bug

robert referenced this issue from a pull request that will close it,

2026-09-05 22:39:37 +02:00

Test audit of #39: lollipop/self-crossing snapping, fix a real self-crossing bug #116

robert added a new dependency

2026-09-05 22:39:44 +02:00

#32 Off-route detection with hysteresis and auto-recovery

robert added a new dependency

2026-09-05 22:39:44 +02:00

#33 Next turn, then-turn, remaining distance and ETA

robert commented

2026-09-05 22:40:01 +02:00

Owner

Audited this issue's 6 acceptance criteria against #31's RouteSnapper (merged tonight as PR #114) and opened PR #116 with the real, missing test scope. Not closing this issue — real scope remains (see below), leaving that call to Robert.

Already covered by RouteSnapperTest (#31), verified non-superficial (tight tolerances):

Out-and-back not jumping to the return leg — covered.
Re-acquisition after a GPS gap — covered (tight tolerance on resulting distance, not just "found something").
Rejoining further along the route — covered (same test also demonstrates the widened-window reacquisition).

No near-duplicate tests added for these three.

Genuinely missing, now added in PR #116:
2. Lollipop and self-crossing routes. This surfaced a real bug in RouteSnapper, not just an untested case: on a route that crosses itself, ordinary GPS noise (3 m, well inside normal accuracy) near the crossing could make the windowed search jump the cursor irrecoverably to a much later, chronologically-wrong visit of the same physical point. Reproduced concretely on an 8 m half-size figure-eight; fixed with a two-phase search in RouteSnapper.snap() that is verified to change nothing on any previously-covered path (full existing test suite + real-komoot fixture pass unchanged) while closing the gap. A lollipop fixture is also added and confirmed to already track correctly without needing the fix.

Out of this issue's honest scope, not fabricated, dependency edges added instead:
4. Off-route enter/exit hysteresis at the thresholds — belongs to #32, which is separate and still open. RouteSnapper deliberately only reports a raw offset for #32 to build hysteresis on; there is no OffRouteDetector.kt in the repo yet. #39 now depends on #32.
6. Distance-to-turn and ETA against hand-computed values — belongs to #33 ("Next turn, then-turn, remaining distance and ETA"), also separate and open, targeting a NavEngine.kt that doesn't exist yet. The pieces it would combine (Cue.distanceAlongRouteMeters, RouteSnap.distanceAlongRouteMeters) both already exist independently, but nothing combines them yet, and no ETA logic exists anywhere. #39 now depends on #33.

PR: #116. Real ./gradlew :companion:core:test --rerun-tasks and koverVerify both green.

Audited this issue's 6 acceptance criteria against #31's `RouteSnapper` (merged tonight as PR #114) and opened PR #116 with the real, missing test scope. Not closing this issue — real scope remains (see below), leaving that call to Robert. **Already covered by `RouteSnapperTest` (#31), verified non-superficial (tight tolerances):** 1. Out-and-back not jumping to the return leg — covered. 3. Re-acquisition after a GPS gap — covered (tight tolerance on resulting distance, not just "found something"). 5. Rejoining further along the route — covered (same test also demonstrates the widened-window reacquisition). No near-duplicate tests added for these three. **Genuinely missing, now added in PR #116:** 2. Lollipop and self-crossing routes. This surfaced a **real bug** in `RouteSnapper`, not just an untested case: on a route that crosses itself, ordinary GPS noise (3 m, well inside normal accuracy) near the crossing could make the windowed search jump the cursor irrecoverably to a much later, chronologically-wrong visit of the same physical point. Reproduced concretely on an 8 m half-size figure-eight; fixed with a two-phase search in `RouteSnapper.snap()` that is verified to change nothing on any previously-covered path (full existing test suite + real-komoot fixture pass unchanged) while closing the gap. A lollipop fixture is also added and confirmed to already track correctly without needing the fix. **Out of this issue's honest scope, not fabricated, dependency edges added instead:** 4. Off-route enter/exit hysteresis at the thresholds — belongs to #32, which is separate and still open. `RouteSnapper` deliberately only reports a raw offset for #32 to build hysteresis on; there is no `OffRouteDetector.kt` in the repo yet. **#39 now depends on #32.** 6. Distance-to-turn and ETA against hand-computed values — belongs to #33 ("Next turn, then-turn, remaining distance and ETA"), also separate and open, targeting a `NavEngine.kt` that doesn't exist yet. The pieces it would combine (`Cue.distanceAlongRouteMeters`, `RouteSnap.distanceAlongRouteMeters`) both already exist independently, but nothing combines them yet, and no ETA logic exists anywhere. **#39 now depends on #33.** PR: #116. Real `./gradlew :companion:core:test --rerun-tasks` and `koverVerify` both green.

robert referenced this issue from a commit

2026-09-05 22:42:28 +02:00

Merge pull request 'Test audit of #39: lollipop/self-crossing snapping, fix a real self-crossing bug' (#116) from area/navigation-snapping-test-audit into main

robert commented

2026-09-05 22:42:38 +02:00

Owner

PR #116 merged. Independently re-verified the claimed bug fix before merging: reverted just RouteSnapper.kt's two-phase-search change (keeping the new tests), re-ran ./gradlew :companion:core:test --tests "*RouteSnapperTest*" — the new figure-eight test genuinely fails against the pre-fix code, confirming this is a real bug catch, not a test written to match already-correct behaviour. Restored the fix, full :companion:core:test suite green again.

Leaving this issue open as Sentinel recommended — real scope remains on #32 and #33, now correctly tracked as dependencies. Will pick those up as separate issues in due course.

PR #116 merged. Independently re-verified the claimed bug fix before merging: reverted just `RouteSnapper.kt`'s two-phase-search change (keeping the new tests), re-ran `./gradlew :companion:core:test --tests "*RouteSnapperTest*"` — the new figure-eight test genuinely fails against the pre-fix code, confirming this is a real bug catch, not a test written to match already-correct behaviour. Restored the fix, full `:companion:core:test` suite green again. Leaving this issue open as Sentinel recommended — real scope remains on #32 and #33, now correctly tracked as dependencies. Will pick those up as separate issues in due course.

robert referenced this issue

2026-09-05 23:03:37 +02:00

Issue #40: phone-side map slice geometry (clip, project, simplify) #117

robert referenced this issue

2026-09-05 23:45:06 +02:00

Next turn, then-turn, remaining distance and ETA #33