fix: bound the wait for a silent list server

A server that took the connection and never answered held line mode
93 s, three waits of 30 s for a first byte. A file list now waits 15 s
per answer and asks a silent server twice. Downloads keep their own
timeout between chunks.

install.ps1

@@ -11,7 +11,7 @@
 
     $ErrorActionPreference = "Stop"
     $defaultInstallUrl = "https://raw.githubusercontent.com/Abdess/retrobios/main/install.py"
-    $defaultInstallSha256 = "1f6b90760303d04cb6beef7fd294371b2e1d4ba733578c59a95e1b0395c2f376"
+    $defaultInstallSha256 = "aef82904ca3356ded55f7b43bc03fe80dc30da178a28bc598caef8367b3f9766"
     $maximumInstallerBytes = 2MB
     $installerExit = 0
     $installer = if ($PSScriptRoot) { Join-Path $PSScriptRoot "install.py" } else { $null }

install.py

@@ -3322,16 +3322,34 @@ def install_bounded_opener() -> None:
     urllib.request.install_opener(urllib.request.build_opener(*handlers))
 
 
+# The wait for each answer of a file list, its first byte included: a
+# server that takes the connection and never answers (a captive portal, a
+# stuck proxy) held the run 93 s on three waits of 30 s. A list is a few
+# MB, never a slow stream.
+LIST_TIMEOUT = 15.0
+# Attempts after a server that took the connection went silent: once more,
+# a stall is rarely a blip.
+STALLED_ATTEMPTS = 2
+
+
+def _stalled(exc: BaseException) -> bool:
+    """Whether exc is a wait that ran out, the connection made."""
+    reason = getattr(exc, "reason", None)
+    return isinstance(exc, socket.timeout) or isinstance(reason, socket.timeout)
+
+
 def _fetch_json(
-    url: str, limit: int, label: str, attempts: int = MAX_RETRIES, timeout: float = 30,
+    url: str, limit: int, label: str, attempts: int = MAX_RETRIES,
+    timeout: float | None = None,
 ) -> object:
     """A JSON document, tried MAX_RETRIES times like every file download.
 
     A file list fetched once ended the run on one timeout over flaky Wi-Fi.
     A 4xx answer or an invalid document is final: asking again changes
-    nothing.
+    nothing, and a server gone silent is asked once more only.
     """
     _network_allowed(label)
+    timeout = LIST_TIMEOUT if timeout is None else timeout
     for attempt in range(1, attempts + 1):
         try:
             with urllib.request.urlopen(url, timeout=timeout) as resp:
@@ -3344,9 +3362,12 @@ def _fetch_json(
         except (urllib.error.URLError, OSError, http.client.HTTPException) as exc:
             code, problem = network_words(exc)
             # A wrong clock or no certificate list fails every attempt alike.
-            if attempt == attempts or code != "network":
+            if (attempt == attempts or code != "network"
+                    or (_stalled(exc) and attempt >= STALLED_ATTEMPTS)):
                 if isinstance(exc, http.client.HTTPException):
                     raise ConnectionError(f"connection cut: {exc!r}") from exc
+                # How many were made, for the line that says so.
+                exc.attempts = attempt  # type: ignore[union-attr]
                 raise
             delay = _retry_delay(attempt, None)
         # Said, or three 30-second attempts read as a frozen program.
@@ -3416,7 +3437,8 @@ def fetch_manifest(plat: str) -> dict:
             print(f"  Could not fetch the file list for {plat}.", file=sys.stderr)
             print(f"  {words}.", file=sys.stderr)
             sys.exit(1)
-        print(f"  Could not fetch the file list for {plat} after {MAX_RETRIES} "
+        tries = getattr(exc, "attempts", MAX_RETRIES)
+        print(f"  Could not fetch the file list for {plat} after {tries} "
               f"attempts: {words}", file=sys.stderr)
         for line in _address_lines(url):
             print(line, file=sys.stderr)

install.sh

@@ -18,7 +18,7 @@ esac
 TEMP_INSTALLER=""
 TEMP_DIRECTORY=""
 DEFAULT_INSTALL_URL="https://raw.githubusercontent.com/Abdess/retrobios/main/install.py"
-DEFAULT_INSTALL_SHA256="1f6b90760303d04cb6beef7fd294371b2e1d4ba733578c59a95e1b0395c2f376"
+DEFAULT_INSTALL_SHA256="aef82904ca3356ded55f7b43bc03fe80dc30da178a28bc598caef8367b3f9766"
 MAX_INSTALLER_BYTES=2097152
 
 cleanup() {

tests/test_install.py

@@ -17,6 +17,7 @@ import subprocess
 import sys
 import tempfile
 import threading
+import time
 import unittest
 from types import SimpleNamespace
 import unittest.mock
@@ -2369,6 +2370,48 @@ class FileListsAreRetried(unittest.TestCase):
         self.assertIn(f"after {install.MAX_RETRIES} attempts", err.getvalue())
 
 
+class ASilentServerIsBounded(unittest.TestCase):
+    """A server that takes the connection and never answers held line mode
+    93 s: three waits of 30 s for a first byte that never came."""
+
+    def test_two_short_waits_then_the_reason(self):
+        import socket
+        import threading
+
+        listener = socket.socket()
+        listener.bind(("127.0.0.1", 0))
+        listener.listen(8)
+        self.addCleanup(listener.close)
+        held = []
+
+        def accept():
+            while True:
+                try:
+                    conn, _addr = listener.accept()
+                except OSError:
+                    return
+                held.append(conn)
+
+        threading.Thread(target=accept, daemon=True).start()
+        self.addCleanup(lambda: [conn.close() for conn in held])
+        port = listener.getsockname()[1]
+        err = io.StringIO()
+        start = time.monotonic()
+        with unittest.mock.patch.object(install, "LIST_TIMEOUT", 0.5), \
+                unittest.mock.patch.object(install, "_retry_delay", return_value=0), \
+                unittest.mock.patch.object(install, "MANIFEST_URL",
+                                           f"http://127.0.0.1:{port}/{{platform}}.json"), \
+                contextlib.redirect_stderr(err):
+            with self.assertRaises(SystemExit):
+                install.fetch_manifest("retroarch")
+        elapsed = time.monotonic() - start
+        self.assertLess(elapsed, 5)
+        self.assertEqual(len(held), install.STALLED_ATTEMPTS)
+        self.assertIn(f"after {install.STALLED_ATTEMPTS} attempts", err.getvalue())
+        # A download keeps its own, longer wait between chunks.
+        self.assertGreater(install.LIST_TIMEOUT, 1)
+
+
 class TargetListFailureIsNotAnAnswer(unittest.TestCase):
     """A timeout is not 'no targets': it sent users to the full pack."""