# 10 - SPA Runtime Branding Shim
> Why the static `ARRFLIX` patch wasn't enough, what the shim does,
> and how to extend it when Jellyfin starts overwriting more state on upgrade.
Last verified: 2026-05-08 against Jellyfin 10.10.3 web bundle.
---
## TL;DR
`/web/index.html` already says `ARRFLIX` and embeds our logo as the
favicon (data URL). The browser tab still showed "Jellyfin" + the Jellyfin teal
triangle because **Jellyfin's SPA overwrites `document.title` at runtime** as it
hydrates. Two compounding effects:
1. **SPA runtime overwrite (primary cause).** Bundle code does
`document.title = d.Ay.translateHtml(document.title ...)` and per-route updates
via `LibraryMenu.setTitle` / `Page.setTitle`. Whatever the static `` is,
the SPA will replace it on first hydrate and again on every navigation.
2. **Service worker churn (secondary).** `serviceworker.js` registers and calls
`self.clients.claim()`. The shipped SW does NOT actually cache `index.html`
(it's a notification-only worker), but its presence still pins old clients
to whatever they had at first paint. Killing the SW once forces a clean reload.
Fix: a tiny self-contained shim in `<head>`, INLINE, BEFORE the bundle scripts.
It enforces the title and favicon both on hydrate and on every later DOM mutation.
---
## What the shim does
Lives in `web-overrides/index.html` between `<!-- ARRFLIX-SHIM-BEGIN -->` /
`-END` markers. Insertion is idempotent via `bin/inject-shim.py` (re-running the
script REPLACES the existing block instead of stacking duplicates).
Behaviour:
| Step | When | What |
|------|------|------|
| 1 | Page parse | `<script>` runs first thing inside `<head>`, before bundle. |
| 2 | DOM ready | `lockTitle()` strips/replaces "Jellyfin"; `lockFavicon()` re-pins icon hrefs to the data URL already in the page. |
| 3 | Mutation | A `MutationObserver` on `document.head` re-runs both lockers when the SPA tries to change `<title>` text or any `<link rel*="icon">` href. |
| 4 | Interval | A 1s `setInterval` is the safety net for late-binding navigations / route changes that bypass the head observer. |
| 5 | Once | `serviceWorker.getRegistrations()` finds anything pointing at `serviceworker.js` and calls `r.unregister()` so old clients reload fresh. `caches.keys()` is also flushed. |
| 6 | Drawer | `nukeSettings()` finds any `a[href*="mypreferencesmenu"]` / `[to*="mypreferencesmenu"]` node, walks up to the closest `li` / `.MuiListItem-root` / `[role="menuitem"]` wrapper and sets `display:none`. Wired into `start()`, the head observer body-twin, and the 1s setInterval. |
The shim does NOT re-encode the logo. It reads the existing `data:image/png;base64,...`
href from the `<link rel="shortcut icon">` already in the page. So there's exactly
one copy of the logo data URL in the file.
---
## Files in play
```
/tmp/ARRFLIX/
bin/inject-shim.py # idempotent injector; the source of truth for shim content
web-overrides/index.html # bind-mounted to /jellyfin/jellyfin-web/index.html in the container
docs/10-spa-runtime-shim.md # this file
```
Server bind-mount (compose):
```yaml
volumes:
- /opt/docker/jellyfin/web-overrides/index.html:/jellyfin/jellyfin-web/index.html:ro
```
The container reads the bind-mounted file fresh on each request - no `docker
restart` needed when you re-deploy index.html.
---
## Deploying changes
```bash
# 1. Edit bin/inject-shim.py (NOT index.html directly - the script is the source of truth)
# 2. Re-run injector locally
python3 /tmp/ARRFLIX/bin/inject-shim.py
# 3. Copy to nullstone
scp /tmp/ARRFLIX/web-overrides/index.html user@192.168.0.100:/opt/docker/jellyfin/web-overrides/index.html
# 4. Verify
curl -ks https://arrflix.s8n.ru/web/index.html | grep -oE "ARRFLIX-SHIM-(BEGIN|END)"
curl -ks https://arrflix.s8n.ru/web/index.html | grep -oE "<title>[^<]*"
```
No container recreate is needed; the file is read on every HTTP request.
### Single-file bind mount inode gotcha
The compose volume binds a SINGLE file (not a directory). Docker resolves the
bind once at container start and pins to that inode. If your editor or a
`python os.replace()` writes via temp-file-and-rename, the canonical path on
the host now points to a NEW inode while the container still holds the OLD
inode (visible as `Links: 0` in `stat` inside the container, and as a
`//deleted` suffix in `/proc//mountinfo` on the host).
Symptom: you `scp` a new file, the host shows the new content, but
`curl -ks https://arrflix.s8n.ru/web/index.html` still returns the OLD content.
Two safe fixes:
1. **In-place truncate (preferred).** Edit with `cat > /opt/.../index.html`
redirection or `truncate -s 0` then `cat >>`. This reuses the existing
inode, so the bind mount keeps showing the new content. The `scp` command
in step 3 above DOES truncate-in-place (not rename), so the standard
workflow is safe.
2. **Re-bind via nsenter (if you accidentally orphaned the inode).** Requires
privileged container + nsenter into the jellyfin mount namespace:
```bash
# Copy current host file into the container's writable /tmp first
docker cp /opt/docker/jellyfin/web-overrides/index.html \
jellyfin:/tmp/arrflix-index.html
CPID=$(docker inspect jellyfin --format "{{.State.Pid}}")
docker run --rm --privileged --userns=host --pid=host -v /:/host \
-e CPID=$CPID debian:stable-slim sh -c '
nsenter -t $CPID -m -p -- umount /jellyfin/jellyfin-web/index.html
nsenter -t $CPID -m -p -- mount --bind -o ro \
/tmp/arrflix-index.html /jellyfin/jellyfin-web/index.html
'
```
This relies on the docker-group sudo-bypass trick. The new bind survives
until the next container restart, after which Docker re-resolves from the
host path natively.
---
## First-deploy: forcing existing clients to reload
The shim's `r.unregister()` block runs unconditionally. The first time a user with
an old SW + old `index.html` cached visits the site after this deploy:
1. They get the new `index.html` (bind-mount serves it).
2. The shim runs, unregisters the SW, flushes `caches`.
3. The current page still shows the old behaviour (because the SW already
intercepted this navigation). They need ONE hard reload (Ctrl+Shift+R or
Cmd+Shift+R) to get a fully clean session.
4. Subsequent visits are clean.
If you want to STOP unregistering the SW after the wide deploy (e.g. to let
Jellyfin reinstate notification-click handling), edit the script section in
`bin/inject-shim.py`:
```js
if ('serviceWorker' in navigator) {
// disabled after 2026-05-08 wide rollout
// navigator.serviceWorker.getRegistrations().then(...);
}
```
Then re-run the injector and redeploy. There is no harm in leaving it on - the
shipped Jellyfin SW only handles `notificationclick`, which we don't use on a
private invite-only service.
---
## Why the JS shim handles drawer Settings (not CustomCss)
The CustomCss block in `/Branding/Configuration` already includes belt-and-braces
selectors for hiding the drawer Settings entry:
```css
a[href*="mypreferencesmenu"],
li:has(> a[href*="mypreferencesmenu"]),
.MuiListItem-root:has(a[href*="mypreferencesmenu"]),
.MuiListItemButton-root:has(a[href*="mypreferencesmenu"]),
[to="/mypreferencesmenu.html"],
.MuiListItem-root:has([to="/mypreferencesmenu.html"]) { display: none !important; }
```
In testing as a non-admin user (`marco`), the entry STILL rendered in the drawer
despite all of those rules. Root cause investigation:
1. The bundle source (`main.jellyfin.bundle.js`) renders the Settings link as
`(0,a.jsxs)(T.A,{component:C.N_,to:"/mypreferencesmenu.html",...})` where
`T.A` is MUI `ListItemButton` and `C.N_` is React Router `NavLink`.
2. NavLink in hash-router mode renders ``.
3. The `a[href*="mypreferencesmenu"]` selector SHOULD match. It does at the link
level, but MUI's `ListItemButton` puts the click target on the wrapping `li`
which is what the user actually sees as the drawer row.
4. The `:has()` selectors targeting that wrapper depend on browser support for
CSS `:has()` (Chromium 105+, Firefox 121+, Safari 15.4+) AND on the bundle's
class names matching `.MuiListItem-root` — which the production build may
minify or restructure across MUI versions.
5. CustomCss is fetched AFTER the SPA hydrates (see "Pre-bundle critical-path
styles" above). Even when matching, there's a brief flash where the entry
is visible.
The JS shim's `nukeSettings()` sidesteps all of that. It runs in `` before
the bundle, on every body mutation, and once per second. It walks up via
`closest()` so it finds the wrapper regardless of MUI class minification, and
it doesn't rely on `:has()` browser support.
The CustomCss rules are kept in place as belt; the JS shim is the suspenders.
Both the server-side `EnableUserPreferenceAccess=false` policy AND the drawer
hide are needed: the policy stops `/mypreferencesmenu.html` from rendering its
controls, the shim stops the link from showing in the first place.
---
## Caveats
- **Owner's first browser session** must hard-reload once after deploy to evict
the previously-registered service worker. Subsequent reloads, and all
first-time visitors, get the clean experience.
- **CustomCss** (logo replacement on the in-app drawer) is owned by a sibling
agent / `/Branding/Configuration`. The shim does NOT touch CustomCss. If the
in-app drawer logo regresses, that's the CSS branch, not this one.
- **The shim must remain self-contained.** No external `src=`. If it ever needs
more code, add it to the IIFE in `bin/inject-shim.py` and re-run.
- **First-paint flash.** Because the SPA still loads its own bundle, you may
briefly see "Jellyfin" in the tab title before the observer kicks in. Sub-100ms
on a fast connection - acceptable. For the colour flash (dark blue / grey
before Cineplex CSS arrives), see "Pre-bundle critical-path styles" below.
---
## Pre-bundle critical-path styles
A second inline block — a `