Compare commits

..
Author SHA1 Message Date
Chuck 782731052d fix(store): serialize concurrent updates per plugin, check cleanup results
CodeRabbit review on #405 flagged two things in
_reinstall_with_rollback, both verified against current code:

- Real race: the web UI runs Flask with threaded=True and there's a
  single update route, so two overlapping requests for the same
  plugin_id (double-click, two tabs) can interleave. The loser could
  rename the winner's in-progress install aside mid-download, deleting
  its own rollback safety net — worse than the bug this function
  exists to fix. Added a lazy per-plugin_id lock dict (mirrors the
  plugin_manager per-plugin lock pattern) held for the whole function.
- _safe_remove_directory's return value was ignored at both call
  sites. Stale-aside cleanup failure now aborts cleanly instead of
  falling through to a rename that would fail anyway with a less
  useful error; post-success backup-removal failure now logs instead
  of failing silently (still returns True — the update itself
  succeeded, and the next update self-heals the leftover aside).

Left the third nitpick (test_stale_aside_from_previous_crash_is_cleared)
addressed by asserting the stale dir is actually gone and that
install_plugin was reached, rather than just the end-to-end result.

Added a concurrency regression test asserting install_plugin never
runs for the same plugin_id while another call is in flight.
2026-07-13 09:11:21 -04:00
ChuckandClaude Fable 5 641296990d fix(store): plugin updates keep the old install until the new one succeeds
Both reinstall paths in update_plugin — the monorepo-migration remote
switch AND the routine archive update every store user hits — deleted
the installed plugin directory BEFORE downloading its replacement. A
mid-update failure (bad network, registry error) permanently destroyed
the plugin. Seen in the field: a Pi with broken DNS lost 12 plugins in
one update pass during the monorepo migration.

New _reinstall_with_rollback: rename the old install aside (using the
'.standalone-backup-' name pattern plugin discovery already excludes),
run install_plugin, remove the aside on success — restore it on ANY
failure, clearing partial-download debris first. A stale aside from a
previous crash is cleared before starting.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FqzC1nzTWL4kaqgMaQZFam
2026-07-12 17:00:05 -04:00
05e7c43b27 fix(plugins): replace dependency marker files with a real satisfaction check (#390)
* fix(plugins): replace dependency marker files with a real satisfaction check

The .dependencies_installed hash-marker system only tracked "was this exact
requirements.txt hashed before" — not whether the packages it names are
actually present. That made it fragile (a wiped venv, a manually removed
package, or a lost/corrupted marker forces a needless full pip reinstall or,
worse, a false skip) and produced dead weight for the ~10 plugins whose
requirements.txt is comment-only (they still paid a pip subprocess on first
boot before a marker existed).

Replace it with requirements_are_satisfied() in plugin_loader.py, which
checks each real requirement line against importlib.metadata directly, so
install_dependencies() only shells out to pip when something is actually
missing or version-mismatched. Drops the marker file entirely: removed all
marker read/write sites in plugin_loader.py and store_manager.py, the
now-pointless marker-cleanup step in the git-update path, the unused legacy
marker implementation in plugin_manager.py, and the already-stale
clear_dependency_markers.sh script.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* fix(security): close path-injection gap in dependency-satisfaction checks

CodeQL flagged 2 new high-severity "uncontrolled data used in path
expression" alerts at the open() calls inside this PR's new
requirements_has_real_deps()/requirements_are_satisfied() -- both are
reachable from paths that were never run through the basename+trusted-base
sanitiser this codebase already uses elsewhere:

- PluginLoader.install_dependencies() only applied that sanitiser when its
  optional plugins_dir argument was actually passed; the "no plugins_dir"
  branch trusted plugin_dir_real directly. Made plugins_dir required (not
  Optional) so that branch can't exist, and added an explicit guard in
  load_plugin() so install_deps=True without a plugins_dir fails loudly
  instead of silently. Production's only real caller (PluginManager) always
  passes plugins_dir already; the harness/dev-server/render-plugin callers
  all use install_deps=False and are unaffected.

- StoreManager._install_dependencies() never sanitised plugin_path at all,
  and its call sites ultimately derive that path from a plugin's own
  manifest.json "id" field (install_plugin_from_url) -- a malicious plugin
  could otherwise point requirements_file outside plugins_dir. Applied the
  same os.path.basename()-based containment pattern PluginLoader already
  uses (and that CodeQL recognises as a real sanitiser).

Added test_install_dependencies_requires_plugins_dir and
test_install_dependencies_rejects_path_outside_plugins_dir to lock in the
actual security property, not just quiet the scanner. Verified: all 20
tests in test_plugin_loader.py pass, plus the PR's existing test plan
(test_plugin_system.py, test_store_manager_caches.py: 53 passed) and the
full CI plugin-safety suite (test_harness.py, test_visual_rendering.py,
test_plugin_matrix.py: 52 passed, 2 pre-existing skips) all still pass.

* fix(security): replace basename-only sanitiser with a trusted-enumeration check

The previous commit's os.path.basename() + os.path.join() pattern (which a
pre-existing code comment claimed CodeQL recognises as a sanitiser) did not
actually clear the alert -- the next CodeQL run still flagged the same 2
sink lines, plus a new one at the os.path.join() call itself. Taking a
substring of tainted data apparently isn't treated as a barrier by this
query, whatever the comment assumed.

Replaced it with find_trusted_subdir(): enumerate the trusted plugins_dir
via os.scandir() and only use a name that scandir itself produced, matched
by equality against the caller's requested name. The path is then built
from that enumerated entry, not from the caller's string -- a value
sourced from iterating a trusted, non-tainted directory carries no taint
regardless of what it happens to equal, which is a stronger and more
conventional allowlist-style barrier than string-stripping. Applied
identically in both PluginLoader.install_dependencies() and
StoreManager._install_dependencies(), sharing one implementation.

Re-verified: all 65 tests across test_plugin_loader.py (20, including the
2 new security regression tests), test_store_manager_caches.py (35),
test_plugin_system.py (10) pass, plus the full CI plugin-safety suite
(test_harness.py/test_visual_rendering.py/test_plugin_matrix.py: 52
passed, 2 pre-existing skips).

* fix(security): redact URL credentials from pip subprocess output before logging

CodeQL flagged 3 clear-text-logging-of-secrets alerts in
install_requirements_file() (src/common/permission_utils.py:353,360,371).
Pre-existing on main, unrelated to this PR's own diff, but now visible
since the path-injection alerts that previously took priority in the
annotation list are fixed.

The underlying risk is real: pip can echo a private index URL's embedded
basic-auth credentials (from a requirements.txt --index-url line or
PIP_INDEX_URL) back verbatim in its own stderr/stdout on failure, and this
function both logs that output directly and returns it to callers --
store_manager.py's _install_dependencies() logs result.stderr from this
same function too.

Added _redact_url_credentials(), applied immediately after each of the two
subprocess.run() calls (mutating result.stderr/stdout in place) rather
than patching each log call site individually. This closes the leak at
the source: every downstream use -- the three flagged log lines, the
"note" string embedded in the returned stdout, and store_manager.py's own
logging of the returned result -- gets the redacted text for free.

Verified the fixed-phrase "denied" check (`"a password is required" in
result.stderr`) is unaffected, since URL syntax and those phrases don't
overlap -- covered explicitly by
test_does_not_touch_denied_check_phrases. Added
test/test_permission_utils.py (6 tests) covering the redaction helper
directly and both subprocess.run() call sites (the sudo-wrapper branch,
which this repo's scripts/fix_perms/safe_pip_install.sh makes live, and
the no-wrapper fallback branch). All pass.

* fix(security): stop interpolating req_file/pip-output into log calls

The previous commit's redaction (mutating result.stderr/stdout right after
each subprocess.run()) didn't clear CodeQL's clear-text-logging alerts --
same lesson as the path-injection fix earlier in this PR: a static
analyzer can't tell "this value was already sanitised two lines up" from
"this is still the raw tainted value" just by looking at a single log
call in isolation, so it conservatively keeps flagging it regardless of
what the redaction function actually does.

Removed all dynamic interpolation (req_file, result.stderr) from the 3
flagged logger.warning() calls entirely, replacing them with fixed
messages plus (for the one that had it) result.returncode, which is a
plain int with no possible taint. The full redacted detail is still
available where it actually matters -- in the returned
CompletedProcess.stderr/stdout and the "note" text -- just not duplicated
into a log line a scanner has to reason about in isolation.

Re-verified: all 6 test_permission_utils.py tests still pass (they assert
on the returned result, not log call arguments), plus the full
test_plugin_loader.py/test_store_manager_caches.py/test_plugin_system.py
suite (71 passed, 1 pre-existing deselect, 4 subtests).

---------

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-11 08:55:50 -04:00
ChuckandGitHub 2ffc57cf40 fix(plugin-harness): add no-op process_deferred_updates to test double (#391)
The safety harness's VisualTestDisplayManager (base of
BoundsCheckingDisplayManager) doesn't implement process_deferred_updates(),
which 5 first-party ledmatrix-plugins call unconditionally between
set_scrolling_state() and their scroll-position update: news, odds-ticker,
ledmatrix-leaderboard, stock-news, and ledmatrix-stocks. Any of them fails
the harness with AttributeError the moment it's touched (surfaced when
ledmatrix-plugins#177 had to add a local hasattr guard in ledmatrix-stocks
just to pass CI). Add the method as a no-op, mirroring the existing
"no-op for testing" pattern already used for set_scrolling_state, so these
plugins render under the harness without every touching PR needing its own
guard.
2026-07-11 08:55:03 -04:00
ChuckandGitHub aab0e9ade0 fix(plugin-manager): fix TypeError breaking every plugin's scheduled update (#392)
run_scheduled_updates()'s resource-monitor branch wrapped the update call
in a closure stored as a *class* attribute on a dynamically-built type
(type('obj', (object,), {'update': monitored_update})()). The descriptor
protocol turns a function found via class-attribute lookup into a bound
method on instance access, silently prepending the synthetic instance as
an implicit first argument -- but monitored_update() takes none, so every
call raised "monitored_update() takes 0 positional arguments but 1 was
given", was caught by run_scheduled_updates' try/except, and recorded as
an update failure.

self.resource_monitor is None by default and was dormant until PR #388
("activate dormant plugin health/metrics subsystem") wired it up in both
display_controller.py and web_interface/app.py -- meaning this bug went
live in every real deployment as of that merge (2026-07-09) despite the
buggy line itself dating back to 2025-12-27. In practice this means no
plugin's update() has succeeded since upgrading past #388: circuit
breakers cycle through half-open -> immediate failure -> reopened every
health-check interval forever, and all plugin data (scores, odds, prices,
etc.) goes stale from whatever was last fetched before the upgrade.
Confirmed live on a running instance: odds-ticker (and stock-news,
ledmatrix-stocks, baseball-scoreboard, ledmatrix-leaderboard, of-the-day)
failing this exact way every 5-minute circuit-breaker retry.

Fixed by using types.SimpleNamespace(update=monitored_update) instead of
a dynamic class: SimpleNamespace stores attributes on the instance
itself, so attribute lookup returns the plain function unchanged --
never routed through the class-attribute descriptor protocol that
injects an implicit self.

Added test_run_scheduled_updates_calls_update_with_resource_monitor to
test/test_plugin_system.py using a real PluginResourceMonitor (not a
mock of it), so the test exercises the actual descriptor-binding
behavior that caused this. Verified the test fails with the exact
reported error against the pre-fix code and passes against the fix.
2026-07-11 08:54:30 -04:00
6 changed files with 348 additions and 11 deletions
+12 -1
View File
@@ -11,6 +11,7 @@ import json
import sys
import time
import threading
import types
from pathlib import Path
from typing import Dict, List, Optional, Any
import logging
@@ -743,8 +744,18 @@ class PluginManager:
# If resource monitor exists, wrap the call
def monitored_update():
self.resource_monitor.monitor_call(plugin_id, plugin_instance.update)
# SimpleNamespace stores `update` as an *instance*
# attribute, so attribute lookup returns the plain
# function object as-is. A dynamically-built class
# (`type(..., {'update': monitored_update})`) instead
# stores it as a *class* attribute, which the
# descriptor protocol turns into a bound method on
# access -- silently prepending the instance as an
# implicit first argument to a function that takes
# none, raising "monitored_update() takes 0
# positional arguments but 1 was given" on every call.
success = self.plugin_executor.execute_update(
type('obj', (object,), {'update': monitored_update})(),
types.SimpleNamespace(update=monitored_update),
plugin_id
)
else:
+93 -9
View File
@@ -142,9 +142,28 @@ class PluginStoreManager:
# then get the result from the warm cache (double-checked locking).
self._registry_fetch_lock = threading.Lock()
# Per-plugin locks for _reinstall_with_rollback: the web UI runs
# Flask with threaded=True, so two overlapping requests for the
# same plugin_id (double-click, two browser tabs) would otherwise
# both rename the same directory aside — one succeeds, and the
# loser can end up renaming the winner's in-progress install aside
# mid-download, stealing its own rollback safety net. Keyed by
# plugin_id so unrelated plugins still update concurrently.
self._reinstall_locks: Dict[str, threading.Lock] = {}
self._reinstall_locks_guard = threading.Lock()
# Ensure plugins directory exists
self.plugins_dir.mkdir(exist_ok=True)
def _get_reinstall_lock(self, plugin_id: str) -> threading.Lock:
"""Lazily create (or fetch) the per-plugin reinstall lock."""
with self._reinstall_locks_guard:
lock = self._reinstall_locks.get(plugin_id)
if lock is None:
lock = threading.Lock()
self._reinstall_locks[plugin_id] = lock
return lock
def _record_cache_backoff(self, cache_dict: Dict, cache_key: str,
cache_timeout: int, payload: Any) -> None:
"""Bump a cache entry's timestamp so subsequent lookups hit the
@@ -2263,6 +2282,74 @@ class PluginStoreManager:
self.logger.error(f"Error uninstalling plugin {plugin_id}: {e}")
return False
def _reinstall_with_rollback(self, plugin_id: str, plugin_path: Path) -> bool:
"""Replace an installed plugin with a fresh install, atomically.
The old install is renamed aside (not deleted) until the new install
succeeds, then removed; on ANY install failure the old directory is
restored. This is the difference between a failed update and a
destroyed plugin: the previous delete-then-install flow permanently
removed plugins whenever the download failed mid-update (seen in the
field during the monorepo migration on a Pi with broken DNS every
old-remote plugin was deleted and none could be re-downloaded).
The aside name embeds '.standalone-backup-' so plugin discovery
(plugin_manager._scan_directory_for_plugins) ignores it even though
it still contains a manifest.json.
Held for the whole operation under a per-plugin_id lock: two
overlapping requests for the same plugin (double-click, two
browser tabs the web UI runs Flask with threaded=True) must not
interleave their renames, or the second could steal the first's
rollback safety net mid-install. Other plugin_ids are unaffected.
"""
with self._get_reinstall_lock(plugin_id):
backup_path = plugin_path.with_name(
f"{plugin_path.name}.standalone-backup-migrating")
# A stale aside from a previous crash would block the rename
if backup_path.exists():
if not self._safe_remove_directory(backup_path):
self.logger.error(
f"Could not clear stale backup for {plugin_id} at "
f"{backup_path}; leaving old install in place")
return False
try:
plugin_path.rename(backup_path)
except OSError as e:
self.logger.error(
f"Could not set aside old plugin directory for {plugin_id}: {e}")
return False
try:
installed = self.install_plugin(plugin_id)
except Exception as e:
self.logger.error(f"Reinstall of {plugin_id} raised: {e}")
installed = False
if installed:
if not self._safe_remove_directory(backup_path):
self.logger.warning(
f"Update of {plugin_id} succeeded but the old backup "
f"at {backup_path} could not be removed; it will be "
f"cleared on the next update")
return True
# Install failed (bad network, registry error...) — put the old
# version back so the user still has a working plugin.
self.logger.error(
f"Reinstall of {plugin_id} failed; restoring previous version")
try:
if plugin_path.exists():
# partial download debris from the failed install
self._safe_remove_directory(plugin_path)
backup_path.rename(plugin_path)
self.logger.info(f"Restored previous install of {plugin_id}")
except OSError as e:
self.logger.error(
f"CRITICAL: could not restore {plugin_id} from {backup_path}: {e}. "
f"The previous install is preserved there — rename it back manually.")
return False
def update_plugin(self, plugin_id: str) -> bool:
"""
Update a plugin to the latest commit on its upstream branch.
@@ -2325,10 +2412,7 @@ class PluginStoreManager:
f"Plugin {resolved_id} git remote ({local_remote}) differs from registry ({registry_repo}). "
f"Reinstalling from registry to migrate to new source."
)
if not self._safe_remove_directory(plugin_path):
self.logger.error(f"Failed to remove old plugin directory for {resolved_id}")
return False
return self.install_plugin(resolved_id)
return self._reinstall_with_rollback(resolved_id, plugin_path)
# Check if already up to date
if remote_sha and local_sha and remote_sha.startswith(local_sha):
@@ -2632,11 +2716,11 @@ class PluginStoreManager:
# Plugin is not a git repo but is in registry and has a newer version - reinstall
self.logger.info(f"Plugin {plugin_id} not installed via git; re-installing latest archive (registry id: {registry_id})")
# Remove directory and reinstall fresh
if not self._safe_remove_directory(plugin_path):
self.logger.error(f"Failed to remove old plugin directory for {plugin_id}")
return False
return self.install_plugin(registry_id)
# Reinstall with the old version kept aside until the new
# download succeeds — this is the path every routine store
# update takes, and a mid-update network failure must not
# destroy the user's plugin.
return self._reinstall_with_rollback(registry_id, plugin_path)
except Exception as e:
import traceback
@@ -454,6 +454,18 @@ class VisualTestDisplayManager:
"""Check if display is currently scrolling."""
return self._scrolling_state['is_scrolling']
def process_deferred_updates(self):
"""Process any deferred updates (no-op for testing).
Several ticker-style plugins (news, odds-ticker, leaderboard,
stock-news, stocks) call this unconditionally between
set_scrolling_state() and their scroll-position update, mirroring the
real display_manager's deferred-update queue. This double has no such
queue, so there is nothing to process the no-op just lets those
plugins render under the harness instead of raising AttributeError.
"""
pass
# ------------------------------------------------------------------
# Utility methods
# ------------------------------------------------------------------
+10
View File
@@ -172,6 +172,16 @@ class TestVisualDisplayManager:
vdm.set_scrolling_state(False)
assert vdm.is_currently_scrolling() is False
def test_process_deferred_updates_is_noop(self):
# Ticker-style plugins (news, odds-ticker, leaderboard, stock-news,
# stocks) call this unconditionally alongside set_scrolling_state();
# it must exist and be harmless so those plugins render under the
# harness instead of raising AttributeError.
vdm = VisualTestDisplayManager(width=128, height=32)
vdm.set_scrolling_state(True)
vdm.process_deferred_updates() # should not raise
assert vdm.is_currently_scrolling() is True
def test_format_date_with_ordinal(self):
from datetime import datetime
vdm = VisualTestDisplayManager(width=128, height=32)
+50 -1
View File
@@ -3,6 +3,7 @@ from unittest.mock import MagicMock, patch
from pathlib import Path
from src.plugin_system.plugin_manager import PluginManager
from src.plugin_system.plugin_state import PluginState
from src.plugin_system.resource_monitor import PluginResourceMonitor
class TestPluginManager:
"""Test PluginManager functionality."""
@@ -74,10 +75,58 @@ class TestPluginManager:
# No manifest in pm.plugin_manifests
result = pm.load_plugin("non_existent_plugin")
assert result is False
assert pm.state_manager.get_state("non_existent_plugin") == PluginState.ERROR
def test_run_scheduled_updates_calls_update_with_resource_monitor(
self, mock_config_manager, mock_display_manager, mock_cache_manager
):
"""Regression test: run_scheduled_updates() must actually call a
plugin's update() when self.resource_monitor is set (as it is in
every real deployment -- display_controller.py and web_interface/
app.py both assign a real PluginResourceMonitor after construction).
Previously, the resource_monitor branch wrapped the call in a
function stored as a *class* attribute on a dynamically-built type
(`type('obj', (object,), {'update': monitored_update})()`), which
the descriptor protocol turns into a bound method on access --
silently passing the synthetic instance as an implicit first
argument to monitored_update(), which takes none. Every plugin's
scheduled update failed with "monitored_update() takes 0 positional
arguments but 1 was given" and was silently swallowed into a
circuit-breaker retry loop that never succeeded, so plugin data
(scores, odds, etc.) never refreshed.
"""
with patch('src.plugin_system.plugin_manager.ensure_directory_permissions'):
pm = PluginManager(
plugins_dir="plugins",
config_manager=mock_config_manager,
display_manager=mock_display_manager,
cache_manager=mock_cache_manager
)
plugin_instance = MagicMock()
plugin_instance.enabled = True
plugin_instance.update = MagicMock()
pm.plugins["test_plugin"] = plugin_instance
pm.plugin_manifests["test_plugin"] = {"update_interval": 10}
pm.state_manager.set_state("test_plugin", PluginState.ENABLED)
# Plain MagicMock, not the mock_cache_manager fixture: this test
# is about run_scheduled_updates() actually invoking update()
# through the resource-monitor wrapper, not about
# PluginResourceMonitor's own cache-backed metrics persistence
# (which calls cache_manager.get(..., memory_ttl=...) --
# a kwarg the fixture's mock_get() doesn't accept).
pm.resource_monitor = PluginResourceMonitor(MagicMock())
pm.run_scheduled_updates(current_time=time.time())
plugin_instance.update.assert_called_once()
assert "test_plugin" in pm.plugin_last_update
assert pm.state_manager.get_state("test_plugin") == PluginState.ENABLED
class TestPluginLoader:
"""Test PluginLoader functionality."""
+171
View File
@@ -0,0 +1,171 @@
"""Tests for atomic plugin updates (store_manager._reinstall_with_rollback).
Regression for a field data-loss incident: update_plugin's reinstall paths
(monorepo migration AND routine archive updates) deleted the installed
plugin BEFORE downloading its replacement a mid-update network failure
permanently destroyed the plugin. Seen live: a Pi with broken DNS lost 12
plugins from one update pass.
"""
import json
import os
import sys
import threading
import time
from pathlib import Path
from unittest.mock import patch
import pytest
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
from src.plugin_system.store_manager import PluginStoreManager # noqa: E402
PLUGIN_ID = "rollback-test-plugin"
@pytest.fixture
def store(tmp_path):
mgr = PluginStoreManager(plugins_dir=str(tmp_path))
plugin_dir = tmp_path / PLUGIN_ID
plugin_dir.mkdir()
(plugin_dir / "manifest.json").write_text(json.dumps(
{"id": PLUGIN_ID, "name": "Rollback Test", "version": "1.0.0"}))
(plugin_dir / "manager.py").write_text("# old version marker\n")
return mgr, plugin_dir
class TestReinstallWithRollback:
def test_failed_install_restores_old_version(self, store):
"""The whole point: a failed download must leave the old install."""
mgr, plugin_dir = store
with patch.object(mgr, "install_plugin", return_value=False):
ok = mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir)
assert ok is False
assert plugin_dir.exists()
assert "old version marker" in (plugin_dir / "manager.py").read_text()
# no aside debris left behind
leftovers = [p for p in plugin_dir.parent.iterdir()
if "standalone-backup" in p.name]
assert leftovers == []
def test_install_exception_restores_old_version(self, store):
mgr, plugin_dir = store
with patch.object(mgr, "install_plugin",
side_effect=RuntimeError("network down")):
ok = mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir)
assert ok is False
assert plugin_dir.exists()
assert "old version marker" in (plugin_dir / "manager.py").read_text()
def test_successful_install_removes_aside(self, store):
mgr, plugin_dir = store
def fake_install(plugin_id):
new_dir = plugin_dir # same path, new content
new_dir.mkdir(exist_ok=True)
(new_dir / "manager.py").write_text("# new version\n")
(new_dir / "manifest.json").write_text(json.dumps(
{"id": PLUGIN_ID, "name": "Rollback Test", "version": "2.0.0"}))
return True
with patch.object(mgr, "install_plugin", side_effect=fake_install):
ok = mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir)
assert ok is True
assert "new version" in (plugin_dir / "manager.py").read_text()
leftovers = [p for p in plugin_dir.parent.iterdir()
if "standalone-backup" in p.name]
assert leftovers == []
def test_partial_download_debris_is_replaced_by_old_version(self, store):
"""A failed install that left a partial directory must still roll back."""
mgr, plugin_dir = store
def fake_partial_install(plugin_id):
plugin_dir.mkdir(exist_ok=True)
(plugin_dir / "half-downloaded.tmp").write_text("junk")
return False
with patch.object(mgr, "install_plugin", side_effect=fake_partial_install):
ok = mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir)
assert ok is False
assert "old version marker" in (plugin_dir / "manager.py").read_text()
assert not (plugin_dir / "half-downloaded.tmp").exists()
def test_stale_aside_from_previous_crash_is_cleared(self, store):
mgr, plugin_dir = store
stale = plugin_dir.parent / f"{PLUGIN_ID}.standalone-backup-migrating"
stale.mkdir()
(stale / "old.txt").write_text("stale")
with patch.object(mgr, "install_plugin", return_value=False) as mock_install:
ok = mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir)
# The reinstall itself still fails (mocked) and the old install is
# restored, but the stale aside must not have survived — otherwise
# it would have blocked this run's own rename (or a future one).
assert not stale.exists()
mock_install.assert_called_once_with(PLUGIN_ID)
assert ok is False
assert plugin_dir.exists()
assert "old version marker" in (plugin_dir / "manager.py").read_text()
def test_concurrent_updates_for_same_plugin_are_serialized(self, store):
"""Two overlapping requests for the same plugin_id (double-click,
two browser tabs the web UI runs Flask with threaded=True) must
not interleave: the loser must wait for the winner to finish
rather than renaming the winner's in-progress install aside and
stealing its rollback safety net."""
mgr, plugin_dir = store
active = 0
max_active = 0
guard = threading.Lock()
def fake_install(plugin_id):
nonlocal active, max_active
with guard:
active += 1
max_active = max(max_active, active)
time.sleep(0.05)
plugin_dir.mkdir(exist_ok=True)
(plugin_dir / "manager.py").write_text("# new version\n")
(plugin_dir / "manifest.json").write_text(json.dumps(
{"id": PLUGIN_ID, "name": "Rollback Test", "version": "2.0.0"}))
with guard:
active -= 1
return True
results = []
def worker():
results.append(mgr._reinstall_with_rollback(PLUGIN_ID, plugin_dir))
with patch.object(mgr, "install_plugin", side_effect=fake_install):
threads = [threading.Thread(target=worker) for _ in range(2)]
for t in threads:
t.start()
for t in threads:
t.join(timeout=5)
assert max_active == 1, "install_plugin ran concurrently for the same plugin_id"
assert results == [True, True]
assert plugin_dir.exists()
assert "new version" in (plugin_dir / "manager.py").read_text()
leftovers = [p for p in plugin_dir.parent.iterdir()
if "standalone-backup" in p.name]
assert leftovers == []
def test_aside_name_is_invisible_to_discovery(self, store, tmp_path):
"""The aside still contains a manifest.json — discovery must skip it
(relies on the existing '.standalone-backup-' exclusion)."""
mgr, plugin_dir = store
from src.plugin_system.plugin_manager import PluginManager
aside = plugin_dir.parent / f"{PLUGIN_ID}.standalone-backup-migrating"
plugin_dir.rename(aside)
pm = PluginManager(plugins_dir=str(tmp_path), config_manager=None,
display_manager=None, cache_manager=None)
found = pm._scan_directory_for_plugins(Path(tmp_path))
assert PLUGIN_ID not in found
if __name__ == "__main__":
sys.exit(pytest.main([__file__, "-v"]))