The lean-build job proved the delete-a-plugin guarantee with two plugins and four hand-picked marker components. It now builds a real twelve-plugin set, and the assertions follow a rule instead of a hunch: one component from EVERY omitted plugin must be absent, one from EVERY chosen plugin must be present, and the backend tree is checked both ways. An omitted set that grows without its check growing is how a leak ships. CHANGELOG had not been touched across the whole range, so the largest security-behaviour change in the product - per-PC endpoint credentials, the shared-key refusal for enrolled PCs, server pinning, agent 1.20.1 and tray 1.7.1 - had no release record. An [Unreleased] section now carries it, written from the commits rather than from memory. CLAUDE.md and the agent quick-reference stated that ADR-016 bans storing credentials in the registry. It does not, and this fleet depends on HKLM for its collector key and server pins: the next person following that text removes them and silently breaks reporting and pinning on every un-enrolled PC. Corrected, along with two other statements that were false on this branch - the database is MySQL, not PostgreSQL, and column names are lowercase concatenated, not snake_case, which is the opposite of what the gate enforces. The wiki drafts are updated against the code as it stands, each old finding given a verdict with a file and line rather than left as an assertion from a week ago, and a new page records the ADR conformance result - including that the ADR set itself is not on this branch, so nothing here enforces it except the naming and style script. They are also scrubbed of the things that must not reach a published wiki: the site name, the internal git server, developer home paths and the assistant tooling. They now pass the repository's own publishability patterns, which is what the scrub is for.
17 KiB
ADR-019: Plugin Dynamic Settings Routes
Status: Accepted (2026-09-11)
Context: Plugin UI extensions, lean builds
Supersedes: Hardcoded plugin routes in core router
Last checked against code: 2026-09-18, branch gh-wj at 35d4fcd41
This draft lives only in docs/wiki-drafts/. There is no docs/adr/ directory
on this branch, so this is the sole ADR-shaped record here and nothing enforces
it except the tests named below. See ADR-Conformance-2026-09-18.md.
Problem
Plugins that extend core settings pages (like Operating Systems in Model Catalog) required hardcoded imports in frontend/src/router/routes/core.js:
{
path: 'operatingsystems',
name: 'model-catalog-os',
component: () => import('../../.plugins-staged/computers/views/OperatingSystemsList.vue')
}
This broke lean builds (ci-lean profile with only machines,printers plugins) because the import referenced a component that didn't exist in the staged plugins.
The deeper issue: Core code shouldn't know about plugin UI structure. Every plugin that wants a settings tab requires a core code change, violating the plugin architecture principle that plugins are self-contained extensions.
Decision
Plugins declare their settings routes via a get_settings_routes() hook that returns route metadata. The frontend fetches these at runtime from /api/plugins/settings-routes and dynamically registers them via router.addRoute().
Backend: BasePlugin Hook
Added to shopdb/plugins/base.py:
def get_settings_routes(self) -> List[Dict]:
"""Frontend routes this plugin adds to settings pages.
Returns a list of route configurations for the plugin's settings pages.
Each route is a dict with:
hub - settings hub to attach to ('model-catalog', 'asset-config', etc.)
path - route path relative to hub (e.g. 'operatingsystems')
name - unique route name (e.g. 'model-catalog-os')
tab - tab label shown in hub UI (e.g. 'Operating Systems')
component - path to Vue component relative to frontend/src/
(e.g. '.plugins-staged/computers/views/OperatingSystemsList.vue')
Return [] (default) if the plugin adds no settings pages.
"""
return []
Example: ComputersPlugin
class ComputersPlugin(BasePlugin):
def get_settings_routes(self) -> List[Dict]:
"""Register Operating Systems management page in Model Catalog."""
return [{
'hub': 'model-catalog',
'path': 'operatingsystems',
'name': 'model-catalog-os',
'tab': 'Operating Systems',
'component': '.plugins-staged/computers/views/OperatingSystemsList.vue'
}]
Backend: API Endpoint
In shopdb/core/api/plugins.py:46-95, as it stands on gh-wj at 35d4fcd41:
@plugins_bp.route('/settings-routes', methods=['GET'])
@require_permission('settings.view')
def get_plugin_settings_routes():
"""Return settings routes contributed by all enabled plugins."""
pm = current_app.extensions.get('plugin_manager')
if not pm:
return success_response({'routes': []})
routes = []
for plugin_name, plugin in pm.get_all_plugins().items():
# Only include routes from enabled plugins
if not pm.registry.is_enabled(plugin_name):
continue
if not hasattr(plugin, 'get_settings_routes'):
continue
# One broken plugin must not take every settings hub down with it.
try:
plugin_routes = plugin.get_settings_routes()
except Exception:
current_app.logger.exception(
'get_settings_routes failed for plugin %s; skipping it',
plugin_name)
continue
routes.extend(renderable_settings_routes(plugin_name, plugin_routes,
current_app.logger))
return success_response({'routes': routes})
Earlier revisions of this draft showed a @jwt_required(optional=True) handler
iterating pm.get_enabled_plugin_instances(). Neither is in the code:
grep -rn get_enabled_plugin_instances shopdb/ plugins/ returns nothing, so
that block would not have run. It is replaced above rather than kept, because a
copy-paste block that reopens the anonymous read is worse than no block.
Auth: the endpoint is gated on settings.view. It was written anonymous, on the reasoning that unauthenticated pages (kiosks, TV displays) register plugin tabs before login. Nothing does: its only caller is ModelCatalogHub.vue, which sits behind requiresAuth and requiresAdmin. Open, it handed an anonymous caller the enabled plugin list, the settings hub layout and the internal Vue component paths, so it was closed.
Frontend: Dynamic Route Registration
Settings hub pages (e.g., ModelCatalogHub.vue) fetch plugin routes at mount and register them dynamically:
<script setup>
import { ref, onMounted } from 'vue'
import { useRoute, useRouter } from 'vue-router'
import axios from 'axios'
const route = useRoute()
const router = useRouter()
const pluginTabs = ref([])
onMounted(async () => {
// Fetch plugin-contributed settings routes
try {
const response = await axios.get('/api/plugins/settings-routes')
if (response.data && response.data.data) {
// Filter for this hub's routes
pluginTabs.value = response.data.data.routes.filter(r => r.hub === 'model-catalog')
// Dynamically register routes
pluginTabs.value.forEach(tab => {
if (!router.hasRoute(tab.name)) {
router.addRoute('model-catalog', {
path: tab.path,
name: tab.name,
component: () => import(`../../${tab.component}`)
})
}
})
}
} catch (error) {
console.error('Failed to load plugin settings routes:', error)
}
// Default redirect
if (route.name === 'model-catalog') {
router.replace({ name: 'model-catalog-vendors' })
}
})
</script>
<template>
<div class="tabs">
<!-- Core tabs -->
<router-link :to="{ name: 'model-catalog-vendors' }" class="tab">
Vendors
</router-link>
<router-link :to="{ name: 'model-catalog-types' }" class="tab">
Model Types
</router-link>
<router-link :to="{ name: 'model-catalog-models' }" class="tab">
Models
</router-link>
<!-- Plugin tabs loaded dynamically -->
<router-link
v-for="tab in pluginTabs"
:key="tab.name"
:to="{ name: tab.name }"
class="tab"
>
{{ tab.tab }}
</router-link>
</div>
<router-view />
</template>
Key behavior:
- Routes registered via
router.addRoute()persist for the session - Components lazy-load when the tab is clicked (code-splitting preserved)
- Tabs only appear if the plugin is enabled and staged
Constraints as implemented
What the sections above describe is the mechanism as designed. What ships is narrower, and a plugin author reading only the hook signature would not find this out until a tab failed to appear in front of a user. The three constraints below are the ones the contract does not express.
One hub draws plugin tabs: Model Catalog. router.addRoute appears exactly
once in the frontend, in frontend/src/views/settings/ModelCatalogHub.vue.
AssetConfigHub, SiteConfigHub, NetworkHub, SecurityHub, LocationsHub,
IntegrationsHub and AssetTypeHub render their own tabs and nothing else, so
a route declaring any of their ids is accepted by the hook, returned by the API
and drawn by nobody. Adding a hub to the supported set means writing the
fetch-and-register code into that hub's .vue file, not editing a list. The
supported set is named once, in SETTINGS_HUBS_WITH_PLUGIN_TABS
(shopdb/plugins/base.py); /api/plugins/settings-routes drops a route outside
it and logs an error naming the plugin and the hub, the plugin manager logs the
same finding at startup, and tests/test_core/test_plugin_settings_hubs.py
fails if a bundled plugin declares an unsupported hub or if the list and the
frontend drift apart.
A plugin tab is not deep-linkable. The hub registers its plugin routes in
onMounted. Before the hub has mounted, the router has no such route, so
navigating straight to a plugin tab URL - from a link, a bookmark, or a refresh
while sitting on one - resolves against a router that does not know the path.
Only arriving through the parent hub works. This is a real limit on what a
plugin settings page can be: it cannot be somewhere you send a person.
The component does not load in a production build at all. The component
field is a server-supplied string interpolated into a dynamic import,
() => import(`../../${tab.component}`). Vite refuses that template - it has
no static file extension - and says so during every build:
[plugin vite:dynamic-import-vars] src/views/settings/ModelCatalogHub.vue:
invalid import "../../${tab.component}". A file extension must be included
in the static part of the import.
The build still succeeds, and the emitted chunk carries the expression verbatim
(import(../../${e.component})), rewritten to nothing. At runtime the browser
resolves it against the chunk's own URL under /assets/ and asks the server for
a .vue file; dist/ contains no .vue files, so the request 404s and the tab
throws when it is clicked. Under npm run dev the same import works, because
the dev server serves sources by path. So the mechanism as it stands is a
development-only path: a tab that appears in production and fails on the click.
Nothing visibly broke because the one plugin using it does not depend on it.
plugins/computers/frontend/routes.js also registers the same page statically
at /settings/operatingsystems, and that route - a normal static import, a real
URL - is what production actually uses. The dynamic tab is a duplicate way in.
Two ways out, and this ADR does not pick one: give the template a static
extension (declare component without .vue and import ${component}.vue),
which makes Vite glob every .vue under frontend/src at build time and makes
the tab real for staged plugins while staying build-time bound; or accept that
plugin settings pages are ordinary plugin routes and drop the hub-tab mechanism.
Deciding that is the open question the architectural review raised.
What a plugin author must therefore do:
- Declare
hub: 'model-catalog'or declare nothing. For a settings page that belongs anywhere else, own a full page under the plugin's own routes (plugins/<name>/frontend/routes.js), which is statically registered, deep- linkable and not subject to any of this. - Register the page in
plugins/<name>/frontend/routes.jseven when you do declare a tab, and treat that route as the real one. A tab whose component is reachable no other way is a page nobody can open in production.tests/test_core/test_plugin_settings_hubs.pyenforces this. - Point
componentat the staged path -.plugins-staged/<name>/views/X.vue, relative tofrontend/src- and keep the source atplugins/<name>/frontend/views/X.vueso staging finds it. - Expect the frontend to be rebuilt when the plugin is added to a site. Enabling the plugin alone does not put its component in the bundle.
- Do not publish, link to or bookmark the tab's URL. Send people to the hub.
Consequences
Benefits
Lean builds work: A build with SITE_PLUGINS=machines,printers doesn't stage the computers plugin, so its component doesn't exist. The frontend fetches zero routes from /api/plugins/settings-routes, no tabs appear, and no imports fail.
Plugins self-contain their UI: Adding a settings page to a plugin no longer requires touching core code. The plugin declares the route; the framework discovers it.
Runtime flexibility: Enabling/disabling a plugin (via /api/plugins/{name} PUT) changes what routes are returned. After an app restart, the tabs appear or disappear automatically - for a plugin whose frontend is in the running bundle. Enabling one that was never staged adds a tab whose component cannot be resolved (see "Constraints as implemented").
Composability: Multiple plugins can contribute to the same hub. Model Catalog could have tabs from computers (Operating Systems), printers (Printer Models), machines (Machine Types), etc., with no conflicts.
Trade-offs
Route registration is async: The tabs appear after mount, not at router initialization. In practice this is <100ms and visually seamless, but the router doesn't know about plugin routes at build time.
Component paths are strings: The component field is a string path that gets interpolated into import(). This prevents static analysis tools from tracing the dependency, and as written Vite will not rewrite the import at all, so it resolves in the dev server and 404s in a build (see "Constraints as implemented").
Composability is one hub wide: "Multiple plugins contribute to the same hub" is true of Model Catalog and of no other hub, because no other hub page registers plugin routes. See "Constraints as implemented".
A plugin tab has no URL worth sharing: routes register at hub mount, so the tab is not reachable by link, bookmark or refresh.
No SSR for plugin tabs: Server-side rendering can't evaluate router.addRoute() at build time. ShopDB doesn't use SSR, so this is not a concern, but it would block SSR adoption without refactoring.
Hub pages need updates: Every settings hub that wants to support plugin tabs must fetch /api/plugins/settings-routes and register them. The pattern is copy-paste (see ModelCatalogHub), but it's not automatic.
Alternatives Considered
1. Conditional imports at build time
Use Vite's conditional imports based on SITE_PLUGINS:
const routes = [
{ path: 'vendors', component: () => import('../../views/settings/VendorsList.vue') },
{ path: 'types', component: () => import('../../views/settings/ModelTypesList.vue') },
]
if (import.meta.env.VITE_SITE_PLUGINS.includes('computers')) {
routes.push({
path: 'operatingsystems',
component: () => import('../../.plugins-staged/computers/views/OperatingSystemsList.vue')
})
}
Rejected: Still requires core code to know about every plugin's routes. Not composable - adding a new plugin means editing core.
2. Plugin manifests drive build-time codegen
Each plugin has a manifest.json with routes. The build reads all manifests and generates a plugin-routes.js that imports components and exports merged routes.
Rejected: Build-time coupling is tighter than runtime. Changing a plugin's routes requires rebuilding the entire frontend. Dynamic enable/disable doesn't work without a rebuild.
3. Frontend plugin discovery via import.meta.glob
Use Vite's glob imports to discover all plugin components:
const pluginComponents = import.meta.glob('../.plugins-staged/*/views/*.vue')
Rejected: Discovery is too broad. Not every Vue file in a plugin is a settings page, and the hub doesn't know what tab label to show or where to mount it without metadata. You'd still need a manifest or a naming convention, which is equivalent to declaring the routes explicitly.
Implementation
Commits:
fe09d2745e- Plugin dynamic settings routes architecture (backend hook + frontend registration)d59d0c3194- Add/api/plugins/settings-routesto OpenAPI spec
Files changed:
shopdb/plugins/base.py- Addedget_settings_routes()hookplugins/computers/plugin.py- Implemented hook for Operating Systems tabshopdb/core/api/plugins.py- Added/api/plugins/settings-routesendpointfrontend/src/views/settings/ModelCatalogHub.vue- Fetch and register plugin routesfrontend/src/router/routes/core.js- Removed hardcoded computers plugin importdocs/openapi.json- Documented new endpoint
Tests: Gate tests pass (test_openapi_spec.py, test_public_read_surface.py).
Future Extensions
Other settings hubs: AssetConfigHub, SecurityHub, NetworkHub, etc. can adopt the same pattern. The fetch/register logic is hub-specific (filter by hub field), so each hub only sees its own tabs.
Nested routes: A plugin could return routes with children: [...] for nested settings pages. Router supports this via addRoute(parentName, childRoute).
Permissions: The tab could include a permission field (e.g., "computers.manage_os"), and the frontend would filter out tabs the user cannot reach. This would be UI gating only, and the page behind the tab must still enforce its own permission; the settings.view gate on /api/plugins/settings-routes says who may read the tab list, not who may use a given tab.
Icons: Add an icon field to the route metadata (e.g., "mdi-microsoft-windows"), and the tab renders the icon next to the label.
Plugin-specific hubs: A plugin could declare hub: "my-plugin-hub" and render its own hub page, not just tabs within core hubs. The router would need to discover and register hub pages too.
References
- ADR-001: Asset as platform contract (plugin architecture principles)
- ADR-013: Plugin catalog and lean builds (problem context)
- PR #XXX: Plugin dynamic settings routes (implementation)
- Issue #XXX: Lean builds fail with hardcoded plugin imports