v3 stream-through; forward range to HF; cache on full request

app.py

@@ -307,12 +307,14 @@ class Handler(http.server.BaseHTTPRequestHandler):
                 remaining -= len(chunk)
 
     def _stream_from_hf_with_cache(self, key, dataset, file_path):
-        """Stream from HF dataset directly, caching the file simultaneously."""
+        """Stream from HF dataset — if full file requested, cache it; otherwise forward range."""
         dl_url = f"https://huggingface.co/datasets/{dataset}/resolve/main/{file_path}"
-        req_headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-
         client_range = self.headers.get("Range", "")
 
+        req_headers = {"Authorization": f"Bearer {HF_TOKEN}"}
+        if client_range:
+            req_headers["Range"] = client_range
+
         try:
             hf_req = urllib.request.Request(dl_url, headers=req_headers)
             hf_resp = urllib.request.urlopen(hf_req, timeout=300)
@@ -322,25 +324,21 @@ class Handler(http.server.BaseHTTPRequestHandler):
             return
 
         total = int(hf_resp.getheader("Content-Length", 0) or 0)
-
-        if client_range:
-            match = re.match(r"bytes=(\d+)-(\d*)", client_range)
-            if match:
-                cs = int(match.group(1))
-                ce = int(match.group(2)) if match.group(2) else total - 1
-                clen = ce - cs + 1
-                self.send_response(206)
-                self.send_header("Content-Range", f"bytes {cs}-{ce}/{total}")
-            else:
-                cs, ce, clen = 0, total - 1, total
-                self.send_response(200)
+        hf_ct = hf_resp.getheader("Content-Type", "video/mp4")
+        hf_cr = hf_resp.getheader("Content-Range", "")
+
+        if hf_cr:
+            self.send_response(206)
+            self.send_header("Content-Range", hf_cr)
+        elif client_range:
+            self.send_response(206)
+            if total:
+                self.send_header("Content-Range", f"bytes 0-{total-1}/{total}")
         else:
-            cs, ce, clen = 0, total - 1, total
             self.send_response(200)
 
-        ct = hf_resp.getheader("Content-Type", "video/mp4")
-        self.send_header("Content-Type", ct)
-        self.send_header("Content-Length", str(clen))
+        self.send_header("Content-Type", hf_ct)
+        self.send_header("Content-Length", str(total))
         self.send_header("Accept-Ranges", "bytes")
         self.send_header("Cache-Control", f"public, max-age={CACHE_TTL}")
         self.send_header("Content-Disposition", f'inline; filename="{key}"')
@@ -348,34 +346,30 @@ class Handler(http.server.BaseHTTPRequestHandler):
         self.send_header("Access-Control-Expose-Headers", "*")
         self.end_headers()
 
-        temp = CACHE_DIR / f"dl_{int(time.time()*1000)}_{os.urandom(2).hex()}"
+        do_cache = not client_range and total > 0
+        temp = CACHE_DIR / f"dl_{int(time.time()*1000)}_{os.urandom(2).hex()}" if do_cache else None
         written = 0
-        skipped = 0
         try:
-            with open(temp, "wb") as tmpf:
-                while True:
-                    chunk = hf_resp.read(65536)
-                    if not chunk:
-                        break
+            tmpf = open(temp, "wb") if temp else None
+            while True:
+                chunk = hf_resp.read(65536)
+                if not chunk:
+                    break
+                if tmpf:
                     tmpf.write(chunk)
-                    written += len(chunk)
-                    if skipped < cs:
-                        need = cs - skipped
-                        if len(chunk) <= need:
-                            skipped += len(chunk)
-                            continue
-                        chunk = chunk[need:]
-                        skipped = cs
-                    self.wfile.write(chunk)
-                    self.wfile.flush()
+                self.wfile.write(chunk)
+                self.wfile.flush()
+                written += len(chunk)
+            if tmpf:
+                tmpf.close()
         except (BrokenPipeError, ConnectionResetError):
             pass
         finally:
             hf_resp.close()
-
-        if total > 0 and written >= total * 0.8:
-            cache.put(key, temp, ct)
-        temp.unlink(missing_ok=True)
+            if temp and written > 0 and written == total:
+                cache.put(key, temp, hf_ct)
+            if temp:
+                temp.unlink(missing_ok=True)
 
     def _json(self, code, data):
         self.send_response(code)