Streckenverläufe kompakt speichern und nur bei Bedarf lesen

pfaedle erzeugt für die Schweiz eine 2.6-GB-shapes.txt; die wird jetzt
in shapes.bin (670 MB, mmap) umgewandelt. Halt-Positionen kommen aus
shape_dist_traveled. RAM-Empfehlung im README ergänzt.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X48pE4bzjHej8w9HyZXCjb

trainpulse/prepare.py

@@ -15,6 +15,7 @@ import shutil
 import subprocess
 import time
 import zipfile
+from array import array
 from pathlib import Path
 
 import httpx
@@ -168,6 +169,38 @@ def run_pfaedle(work: Path, rail: Path) -> Path | None:
     return out
 
 
+def pack_shapes(d: Path) -> None:
+    """shapes.txt ist für die Schweiz ~2.6 GB (56 Mio. Punkte) – viel zu gross für den Speicher.
+    Umwandeln in shapes.bin (float32: lon, lat, Meter ab Start) + shapes.json (id -> [Start, Anzahl]),
+    der Server liest daraus nur die Strecken, die er gerade braucht."""
+    src = d / "shapes.txt"
+    if not src.exists():
+        return
+    index, cur, pts, n = {}, None, [], 0
+    with open(d / "shapes.bin", "wb") as out:
+        def flush():
+            nonlocal n
+            if cur is not None:
+                pts.sort()
+                array("f", [v for p in pts for v in p[1:]]).tofile(out)
+                index[cur] = [n, len(pts)]
+                n += len(pts)
+        with open(src, encoding="utf-8-sig", newline="") as f:
+            r = csv.reader(f)
+            h = {k: i for i, k in enumerate(next(r))}
+            sid, lat, lon, seq = h["shape_id"], h["shape_pt_lat"], h["shape_pt_lon"], h["shape_pt_sequence"]
+            dist = h["shape_dist_traveled"]
+            for row in r:
+                if row[sid] != cur:
+                    flush()
+                    assert row[sid] not in index, "shapes.txt nicht nach shape_id gruppiert"
+                    cur, pts = row[sid], []
+                pts.append((int(row[seq]), float(row[lon]), float(row[lat]), float(row[dist])))
+            flush()
+    (d / "shapes.json").write_text(json.dumps(index, separators=(",", ":")))
+    src.unlink()
+
+
 def publish(src: Path, meta: dict) -> None:
     (src / "meta.json").write_text(json.dumps(meta))
     dest = C.FEEDS_DIR / meta["feed_version"]
@@ -212,6 +245,8 @@ def main() -> None:
 
     shaped = run_pfaedle(work, work / "rail")
     meta["shapes"] = shaped is not None
+    if shaped:
+        pack_shapes(shaped)
     meta["source_file"] = name
     meta["prepared_at"] = int(time.time())
     publish(shaped or work / "rail", meta)