Pull progress: overall byte summary per tick

Per-layer events gave no sense of how much is LEFT. Downloading lines
are parsed per-layer and summed; every 3s the snapshot becomes
'⏳ 3.14/9.27 GB (34%) · 最大层 3f4a2b1c9d0e: 0.51/1.2 GB' instead of
a bare layer line. Key events (Pulling from / Status / errors) still
print immediately.

Co-ached-By: Claude <noreply@anthropic.com>

evalharness/sandbox/prefetch.py

@@ -117,16 +117,44 @@ def _pull_one(image: str, verbose: bool = True) -> None:
                                 text=True, bufsize=1)
         last_line = ''
         last_show = 0.0
+        layer_prog = {}   # layer id -> (done_bytes, total_bytes): the raw
+        # per-layer events give no sense of OVERALL progress -- sum them
         import os as _os
+        import re as _re
+
+        def _bytes(s):
+            s = s.strip()
+            mult = {'kB': 1e3, 'KB': 1e3, 'MB': 1e6, 'GB': 1e9, 'B': 1}
+            for suf, m in mult.items():
+                if s.endswith(suf):
+                    return int(float(s[:-len(suf)]) * m)
+            return int(float(s or 0))
+
         for line in proc.stdout:
             last_line = line.strip()
+            m = _re.match(r'^([0-9a-f]{12}): Downloading.*?([\d.]+[kKMG]?B)/([\d.]+[kKMG]?B)', line)
+            if m:
+                layer_prog[m.group(1)] = (_bytes(m.group(2)), _bytes(m.group(3)))
+            if 'Pull complete' in line:
+                lid = line.split(':')[0]
+                if lid in layer_prog:
+                    layer_prog[lid] = (layer_prog[lid][1], layer_prog[lid][1])
             now = _time.monotonic()
-            if verbose and (any(k in line for k in
+            key_evt = any(k in line for k in
-                                ('Pulling from', 'Pull complete', 'Status',
+                          ('Pulling from', 'Status', 'error', 'denied'))
-                                 'error', 'denied', 'Extracting', 'Verifying'))
+            if verbose and key_evt:
-                            or now - last_show > 3):   # Downloading snapshots
                 print(f'  {last_line[:100]}', file=_sys.stderr, flush=True)
                 last_show = now
+            elif verbose and now - last_show > 3 and layer_prog:
+                done = sum(v[0] for v in layer_prog.values())
+                tot = sum(v[1] for v in layer_prog.values())
+                if tot:
+                    pct = done / tot * 100
+                    big = max(layer_prog.items(), key=lambda kv: kv[1][1] - kv[1][0])
+                    print(f'  ⏳ {done / 1e9:.2f}/{tot / 1e9:.2f} GB ({pct:.0f}%)'
+                          f' · 最大层 {big[0]}: {big[1][0] / 1e9:.2f}/{big[1][1] / 1e9:.2f} GB',
+                          file=_sys.stderr, flush=True)
+                last_show = now
         proc.wait(timeout=3600)
         r = subprocess.CompletedProcess(proc.args, proc.returncode,
                                         stdout='', stderr=last_line)