Merge pull request #577 from trails-cool/feat/gpx-parser-robustness-timestamps

feat(gpx): post-parse timestamp repair

openspec/changes/gpx-parser-robustness/tasks.md

@@ -11,9 +11,9 @@
 
 ## 3. Timestamp repair
 
-- [ ] 3.1 Create `packages/gpx/src/timestamp-repair.ts`: per-segment pass — >50% invalid → drop all; else linear interpolation between valid neighbors with leading/trailing clamp; output ISO 8601 strings
-- [ ] 3.2 Wire the repair pass into `parseTracks` output in `parse.ts`
-- [ ] 3.3 Unit tests (`timestamp-repair.test.ts`): sparse invalid interpolated, >50% invalid dropped (and `movingTime` returns null), leading/trailing clamp, all-valid untouched, no-timestamps untouched
+- [x] 3.1 Create `packages/gpx/src/timestamp-repair.ts`: per-segment pass — >50% invalid → drop all; else linear interpolation between valid neighbors with leading/trailing clamp; output ISO 8601 strings
+- [x] 3.2 Wire the repair pass into `parseTracks` output in `parse.ts`
+- [x] 3.3 Unit tests (`timestamp-repair.test.ts`): sparse invalid interpolated, >50% invalid dropped (and `movingTime` returns null), leading/trailing clamp, all-valid untouched, no-timestamps untouched
 
 ## 4. Metadata fallback
 

packages/gpx/src/parse.ts

@@ -1,5 +1,6 @@
 import type { Waypoint } from "@trails-cool/types";
 import type { GpxData, TrackPoint, ElevationProfile, NoGoArea } from "./types.ts";
+import { repairTimestamps } from "./timestamp-repair.ts";
 
 /**
  * Parse a GPX XML string into structured data.
@@ -31,7 +32,7 @@ function parseGpxWithParser(parser: DOMParser, xml: string): GpxData {
   const name = doc.querySelector("metadata > name")?.textContent ?? undefined;
   const description = doc.querySelector("metadata > desc")?.textContent ?? undefined;
   const waypoints = parseWaypoints(doc);
-  const tracks = parseTracks(doc);
+  const tracks = repairTimestamps(parseTracks(doc));
   const noGoAreas = parseNoGoAreas(doc);
   const { totalDistance, ...elevation } = computeElevation(tracks);
 

packages/gpx/src/timestamp-repair.test.ts

@@ -0,0 +1,56 @@
+import { describe, it, expect } from "vitest";
+import { repairSegmentTimestamps } from "./timestamp-repair.ts";
+import { movingTime } from "./moving-time.ts";
+import type { TrackPoint } from "./types.ts";
+
+const pt = (time?: string, lat = 52.5, lon = 13.4): TrackPoint => ({ lat, lon, time });
+
+describe("repairSegmentTimestamps", () => {
+  it("interpolates a sparse invalid timestamp between valid neighbours", () => {
+    const out = repairSegmentTimestamps([
+      pt("2026-01-01T10:00:00Z"),
+      pt("garbage"),
+      pt("2026-01-01T10:02:00Z"),
+      pt("2026-01-01T10:03:00Z"),
+    ]);
+    // index 1 is midway (by index) between valid index 0 and index 2.
+    expect(out[1]!.time).toBe("2026-01-01T10:01:00.000Z");
+    // Valid points keep their original strings untouched.
+    expect(out[0]!.time).toBe("2026-01-01T10:00:00Z");
+    expect(out[3]!.time).toBe("2026-01-01T10:03:00Z");
+  });
+
+  it("drops all timestamps when more than half the segment is invalid", () => {
+    const out = repairSegmentTimestamps([
+      pt("2026-01-01T10:00:00Z"),
+      pt("garbage"),
+      pt(undefined),
+      pt("nope"),
+    ]);
+    expect(out.every((p) => p.time === undefined)).toBe(true);
+    // A dropped time channel means the segment reads as untimed.
+    expect(movingTime([out])).toBeNull();
+  });
+
+  it("clamps leading and trailing invalid runs to the nearest valid timestamp", () => {
+    const out = repairSegmentTimestamps([
+      pt("garbage"),
+      pt("2026-01-01T10:01:00Z"),
+      pt("2026-01-01T10:02:00Z"),
+      pt(undefined),
+    ]);
+    expect(out[0]!.time).toBe("2026-01-01T10:01:00.000Z"); // leading clamp
+    expect(out[3]!.time).toBe("2026-01-01T10:02:00.000Z"); // trailing clamp
+  });
+
+  it("leaves an all-valid segment untouched", () => {
+    const input = [pt("2026-01-01T10:00:00Z"), pt("2026-01-01T10:01:00Z")];
+    const out = repairSegmentTimestamps(input);
+    expect(out.map((p) => p.time)).toEqual(["2026-01-01T10:00:00Z", "2026-01-01T10:01:00Z"]);
+  });
+
+  it("leaves an untimed segment (no timestamps) untouched", () => {
+    const out = repairSegmentTimestamps([pt(undefined), pt(undefined), pt(undefined)]);
+    expect(out.every((p) => p.time === undefined)).toBe(true);
+  });
+});

packages/gpx/src/timestamp-repair.ts

@@ -0,0 +1,67 @@
+import type { TrackPoint } from "./types.ts";
+
+// Post-parse timestamp repair (spec: gpx-parser-robustness "Timestamp
+// repair"). Recorders from flaky devices emit runs of missing or garbage
+// `<time>` values; passed through raw they silently shrink moving time and
+// skew start-time derivation. This is a pure per-segment pass with an
+// explicit policy, so `movingTime`/`computeDays`/`startTime` stay correct
+// without each re-implementing (and drifting on) the same rules.
+//
+// Policy:
+//   - Validity = `Date.parse` yields a finite epoch.
+//   - A segment with NO valid timestamps is left untouched (an untimed
+//     track — downstream already handles that).
+//   - > 50% of a segment's points invalid → drop ALL of that segment's
+//     timestamps (a mostly-broken time channel is noise, not signal).
+//   - Otherwise → linearly interpolate invalid runs between their valid
+//     neighbours; a leading/trailing invalid run clamps to the nearest
+//     valid timestamp. Only repaired points are rewritten (as ISO 8601);
+//     valid points keep their original string untouched.
+//
+// Monotonicity is deliberately not enforced — `movingTime` already skips
+// non-positive intervals, and reordering recorded data would be fabrication.
+
+function epochOf(time: string | undefined): number {
+  return time == null ? NaN : Date.parse(time);
+}
+
+export function repairSegmentTimestamps(points: TrackPoint[]): TrackPoint[] {
+  const epochs = points.map((p) => epochOf(p.time));
+  const validIdx = epochs.map((e, i) => (Number.isFinite(e) ? i : -1)).filter((i) => i >= 0);
+
+  // Nothing to anchor to, or the time channel is mostly broken → treat the
+  // segment as untimed. (When there are zero valid timestamps this leaves
+  // the already-absent times as-is.)
+  if (validIdx.length === 0) return points;
+  const invalidCount = points.length - validIdx.length;
+  if (invalidCount === 0) return points; // all valid → untouched
+  if (invalidCount / points.length > 0.5) {
+    return points.map((p) => (p.time === undefined ? p : { ...p, time: undefined }));
+  }
+
+  const firstValid = validIdx[0]!;
+  const lastValid = validIdx[validIdx.length - 1]!;
+
+  return points.map((p, i) => {
+    if (Number.isFinite(epochs[i]!)) return p; // valid → keep original string
+    let t: number;
+    if (i < firstValid) {
+      t = epochs[firstValid]!; // leading run clamps forward
+    } else if (i > lastValid) {
+      t = epochs[lastValid]!; // trailing run clamps back
+    } else {
+      // Between two valid anchors: linear interpolation by point index.
+      let prev = i;
+      while (!Number.isFinite(epochs[prev]!)) prev--;
+      let next = i;
+      while (!Number.isFinite(epochs[next]!)) next++;
+      const frac = (i - prev) / (next - prev);
+      t = epochs[prev]! + (epochs[next]! - epochs[prev]!) * frac;
+    }
+    return { ...p, time: new Date(t).toISOString() };
+  });
+}
+
+export function repairTimestamps(segments: TrackPoint[][]): TrackPoint[][] {
+  return segments.map(repairSegmentTimestamps);
+}