Use the apps' real logos as stack icons, fetched server-side (0.52.0)

0.51.0 gave every stack an icon, but a generic one: jellyfin got a clapperboard,
not the Jellyfin logo. Glyphs make a list readable; they do not make a stack
recognisable, which was the point. This resolves stacks against the selfh.st
icon catalog (~2900 self-hosted apps, the set Homarr and Homepage draw on), so
the row shows the thing people already recognise. All 83 bundled templates
resolve to their own logo.

The whole design question was *who* talks to the CDN. If the <img> points at
jsdelivr, then every client needs internet, every page load leaks the names of
somebody's stacks to a third party, and an air-gapped box gets nothing. So the
backend does it: the catalog on startup and weekly after, each logo once on
first use, both into ${DATA_DIR}/stack-icons/. Browsers keep reading icons from
the authenticated endpoint that already existed for uploads, and after the first
fetch the feature is fully offline. Logos are cached per *app*, not per stack —
verified: two stacks resolving to jellyfin produce one download.

Nothing here can fail loudly. Every entry point returns None rather than raising
when the network is absent, the catalog refresh is a task the lifespan does not
await, and an install with no outbound internet simply keeps 0.51.0's glyphs.
That fallback is also what covers a name the catalog does not know
("Mediaserver Wohnzimmer" is still a clapperboard), and the seconds after a
fresh install before the catalog lands. The glyph is derived even for stacks
that *do* have a logo, so an image that cannot be fetched degrades to something
meaningful instead of a box.

Matching gained a second source that turned out to matter more than expected:
the compose images. A stack called "medienserver" says nothing, but it pulls
lscr.io/linuxserver/jellyfin — strip the registry, the vendor and the tag and
the app is right there. Name first, then the longest run of words inside it,
then the images. It is deliberately cautious: a single word shorter than four
characters never claims a logo, because "web", "app" and "db" are all catalog
entries and a *wrong* logo is worse than a neutral glyph. A short alias table
covers what the catalog spells differently from Docker Hub (postgres →
postgresql, pihole → pi-hole, wg-easy → wireguard).

A slug arrives from the database and from query strings and then becomes a
filename, so it is pattern-checked before it is ever joined to a path, catalog
entries that are not slug-shaped are dropped on load, and a downloaded logo is
verified to start with the PNG magic bytes before being cached.

The picker searches the catalog too — pre-seeded with the stack's own name, so
opening it on "jellyfin" offers the Jellyfin logo first — which is how a wrong
match gets corrected, and how a stack can be given any app's logo on purpose.

Verified end to end against the live catalog and real downloads: list rows carry
the resolved logo, the icon endpoint serves real PNG bytes, an unmatched stack
404s (and falls through to its glyph), a hand-picked logo round-trips, reset
clears it, and a traversal slug 404s. 30 new backend tests and 12 new frontend
ones run without any network at all.

0.52.0 rather than amending 0.51.0: those images are already in the registry,
and rebuilding a published version tag with different content is exactly what
breaks the self-update checker.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/services/icon_service.py

@@ -3,10 +3,15 @@
 A stack's ``icon`` column holds one of three things:
 
 ``None``
-    Automatic. Nothing is stored and the UI derives an icon from the stack's
-    name (see ``frontend/src/lib/stackIcons.ts``). Every stack that existed
-    before this feature lands here, so an upgraded install shows sensible icons
-    without a data migration.
+    Automatic: the app logo the stack's name resolves to, and failing that the
+    glyph the frontend derives from the name (``frontend/src/lib/stackIcons.ts``).
+    Nothing is stored, so every stack that existed before this feature lands
+    here and an upgraded install shows sensible icons without a data migration.
+
+``logo:<slug>``
+    The real logo of a known app, from the selfh.st catalog (see
+    ``services/logo_service.py``). The file is cached per *slug*, not per stack,
+    so every Postgres stack shares one download.
 
 ``lucide:<name>``
     A built-in icon the user picked explicitly. The catalog of names lives in
@@ -45,6 +50,7 @@ MAX_ICON_BYTES = 512 * 1024
 _EXTENSIONS = {"png", "jpg", "gif", "webp", "svg"}
 
 _BUILTIN_RE = re.compile(r"^lucide:[a-z0-9-]{1,48}$")
+_LOGO_RE = re.compile(r"^logo:[a-z0-9][a-z0-9-]{0,63}$")
 _CUSTOM_RE = re.compile(r"^custom:(png|jpg|gif|webp|svg):(\d{1,12})$")
 _SAFE_ID_RE = re.compile(r"^[A-Za-z0-9._-]{1,128}$")
 
@@ -116,22 +122,28 @@ def file_for(stack_id: str, value: Optional[str]) -> Optional[str]:
 def normalize_choice(value: str) -> Optional[str]:
     """Validate an icon chosen through the API.
 
-    An empty string means "back to automatic" and maps to ``None``. Only the
-    built-in form is accepted here: a ``custom:`` value is minted by
-    :func:`store_upload` and never taken from a client, or a stack could be
-    pointed at another stack's icon file.
+    An empty string means "back to automatic" and maps to ``None``. A built-in
+    glyph and an app logo are both fine to take from a client — they name a
+    catalog entry, not a file this stack owns. A ``custom:`` value is not: it is
+    minted by :func:`store_upload`, or a stack could be pointed at another
+    stack's uploaded image.
     """
     value = (value or "").strip()
     if not value:
         return None
-    if _BUILTIN_RE.match(value):
+    if _BUILTIN_RE.match(value) or _LOGO_RE.match(value):
         return value
     raise IconError(
-        "Icon must be empty (automatic) or 'lucide:<name>'; upload custom "
-        "images through POST /api/stacks/{id}/icon"
+        "Icon must be empty (automatic), 'lucide:<name>' or 'logo:<slug>'; "
+        "upload custom images through POST /api/stacks/{id}/icon"
     )
 
 
+def logo_slug(value: Optional[str]) -> Optional[str]:
+    """The catalog slug of a ``logo:`` icon value, or None for the rest."""
+    return value[len("logo:"):] if _LOGO_RE.match(value or "") else None
+
+
 # --------------------------------------------------------------------------- #
 # Uploads
 # --------------------------------------------------------------------------- #

backend/services/logo_service.py

@@ -0,0 +1,428 @@
+"""Real app logos for stacks, fetched once and then served from disk.
+
+A stack called ``jellyfin`` should show *the Jellyfin logo*, not a generic
+clapperboard. The logos come from the selfh.st icon set (~2900 self-hosted
+apps), which is the same catalog Homarr, Homepage and Dashy draw on.
+
+Everything crosses the network exactly once and on the server:
+
+* the **catalog** (a JSON index of slugs and display names) is downloaded on
+  startup and refreshed weekly into ``${DATA_DIR}/stack-icons/catalog.json``,
+* a **logo** is downloaded the first time something asks for it and cached at
+  ``${DATA_DIR}/stack-icons/logos/<slug>.png``, keyed by slug rather than by
+  stack so ten Postgres stacks share one file.
+
+Browsers therefore never talk to the CDN: they fetch logos from StackPilot's
+own authenticated icon endpoint, like an uploaded image. After the first fetch
+the whole feature works offline, and an installation with no outbound internet
+degrades to the built-in glyphs rather than breaking — every entry point here
+returns None instead of raising when the network is not there.
+
+Matching a stack to a slug is deliberately conservative: an exact name, then the
+name with punctuation rearranged, then the longest run of words inside it, then
+the images its compose file pulls (``lscr.io/linuxserver/jellyfin:latest`` →
+``jellyfin``). A stack whose name means nothing to the catalog gets no logo and
+falls back to the keyword-derived glyph in the frontend.
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+import re
+import time
+from typing import Iterable, Optional
+
+import httpx
+import yaml
+
+from config import settings
+from services import compose_service
+
+logger = logging.getLogger("stackpilot.logos")
+
+CATALOG_URL = "https://cdn.jsdelivr.net/gh/selfhst/icons/index.json"
+LOGO_URL = "https://cdn.jsdelivr.net/gh/selfhst/icons/png/{slug}.png"
+
+#: The catalog gains a handful of apps a week; there is nothing to gain from
+#: checking more often, and a failed refresh simply keeps the previous copy.
+CATALOG_TTL = 7 * 24 * 3600
+CATALOG_MAX_BYTES = 8 * 1024 * 1024
+LOGO_MAX_BYTES = 2 * 1024 * 1024
+TIMEOUT = httpx.Timeout(15.0)
+
+_SLUG_RE = re.compile(r"^[a-z0-9][a-z0-9-]{0,63}$")
+
+#: Docker image names that are not what the catalog calls the app. Kept short on
+#: purpose — this is for the cases the matcher genuinely cannot reach, not a
+#: second catalog.
+ALIASES = {
+    "postgres": "postgresql",
+    "pgsql": "postgresql",
+    "mongo": "mongodb",
+    "trilium": "trilium-notes",
+    "wg-easy": "wireguard",
+    "wg": "wireguard",
+    "homeassistant": "home-assistant",
+    "hass": "home-assistant",
+    "pihole": "pi-hole",
+    "openwebui": "open-webui",
+    "nextcloud-aio": "nextcloud",
+    "paperless": "paperless-ngx",
+    "paperless-ng": "paperless-ngx",
+    "code-server": "coder",
+    "filebrowser": "file-browser",
+    "qbit": "qbittorrent",
+    "sab": "sabnzbd",
+}
+
+#: Image name components that say nothing about the app.
+_IMAGE_NOISE = {
+    "latest", "linuxserver", "lscr", "ghcr", "docker", "io", "com", "library",
+    "hotio", "alpine", "amd64", "arm64v8", "bitnami", "official",
+}
+
+
+# --------------------------------------------------------------------------- #
+# Paths
+# --------------------------------------------------------------------------- #
+
+
+def _root() -> str:
+    return os.path.join(settings.DATA_DIR, "stack-icons")
+
+
+def catalog_path() -> str:
+    return os.path.join(_root(), "catalog.json")
+
+
+def logo_dir() -> str:
+    return os.path.join(_root(), "logos")
+
+
+def logo_path(slug: str) -> Optional[str]:
+    """Where a slug's logo is cached, or None if the slug is malformed.
+
+    The slug reaches this from the database and from query strings, and it is
+    about to become a filename.
+    """
+    if not _SLUG_RE.match(slug or ""):
+        return None
+    return os.path.join(logo_dir(), f"{slug}.png")
+
+
+# --------------------------------------------------------------------------- #
+# The catalog
+# --------------------------------------------------------------------------- #
+
+
+class Catalog:
+    """Slug lookups built once per catalog file."""
+
+    def __init__(self, entries: list[dict]):
+        self.entries = entries
+        self.slugs: set[str] = set()
+        self.by_name: dict[str, str] = {}
+        self.compact: dict[str, str] = {}
+        for entry in entries:
+            slug = (entry.get("Reference") or "").strip().lower()
+            if not _SLUG_RE.match(slug):
+                continue
+            self.slugs.add(slug)
+            # "AdGuard Home" → "adguard home", so a stack named that matches
+            # even though the slug is hyphenated.
+            name = _normalize(entry.get("Name") or "")
+            self.by_name.setdefault(name, slug)
+            # "pihole" → "pi-hole": people drop the punctuation the catalog keeps.
+            self.compact.setdefault(slug.replace("-", ""), slug)
+            self.compact.setdefault(name.replace(" ", ""), slug)
+
+    def display_name(self, slug: str) -> str:
+        for entry in self.entries:
+            if (entry.get("Reference") or "").lower() == slug:
+                return entry.get("Name") or slug
+        return slug
+
+    def __len__(self) -> int:
+        return len(self.slugs)
+
+
+_catalog: Optional[Catalog] = None
+_catalog_mtime: float = 0.0
+
+
+def load_catalog() -> Optional[Catalog]:
+    """The cached catalog, re-read only when the file on disk changed."""
+    global _catalog, _catalog_mtime
+    path = catalog_path()
+    try:
+        mtime = os.path.getmtime(path)
+    except OSError:
+        return _catalog  # never downloaded, or removed under us
+    if _catalog is not None and mtime == _catalog_mtime:
+        return _catalog
+    try:
+        with open(path, "r", encoding="utf-8") as fh:
+            entries = json.load(fh)
+    except (OSError, json.JSONDecodeError) as exc:
+        logger.warning("Icon catalog is unreadable (%s); ignoring it", exc)
+        return _catalog
+    if not isinstance(entries, list):
+        return _catalog
+    _catalog = Catalog(entries)
+    _catalog_mtime = mtime
+    logger.info("Loaded %d app logos from the icon catalog", len(_catalog))
+    return _catalog
+
+
+def catalog_age() -> Optional[float]:
+    try:
+        return time.time() - os.path.getmtime(catalog_path())
+    except OSError:
+        return None
+
+
+async def refresh_catalog(force: bool = False) -> bool:
+    """Download the catalog unless the copy on disk is still fresh."""
+    age = catalog_age()
+    if not force and age is not None and age < CATALOG_TTL:
+        return False
+    try:
+        async with httpx.AsyncClient(follow_redirects=True, timeout=TIMEOUT) as client:
+            response = await client.get(CATALOG_URL)
+            response.raise_for_status()
+            if len(response.content) > CATALOG_MAX_BYTES:
+                raise ValueError("catalog is implausibly large")
+            entries = response.json()
+        if not isinstance(entries, list) or not entries:
+            raise ValueError("catalog is not a non-empty list")
+    except (httpx.HTTPError, ValueError, json.JSONDecodeError) as exc:
+        # No internet is a normal state for a self-hosted box. Say so once and
+        # carry on with the built-in glyphs.
+        logger.info("Could not refresh the app icon catalog: %s", exc)
+        return False
+    os.makedirs(_root(), exist_ok=True)
+    tmp = catalog_path() + ".tmp"
+    with open(tmp, "w", encoding="utf-8") as fh:
+        json.dump(entries, fh)
+    os.replace(tmp, catalog_path())
+    load_catalog()
+    return True
+
+
+async def catalog_loop() -> None:
+    """Keep the catalog fresh for as long as the app runs."""
+    while True:
+        try:
+            await refresh_catalog()
+        except Exception as exc:  # noqa: BLE001 - a background loop may not die
+            logger.warning("Icon catalog refresh failed: %s", exc)
+        await asyncio.sleep(CATALOG_TTL)
+
+
+# --------------------------------------------------------------------------- #
+# Matching a stack to a slug
+# --------------------------------------------------------------------------- #
+
+
+def _normalize(text: str) -> str:
+    return re.sub(r"[^a-z0-9]+", " ", text.lower()).strip()
+
+
+def _lookup(catalog: Catalog, phrase: str) -> Optional[str]:
+    """A slug for one normalized phrase, trying every spelling of it."""
+    if not phrase:
+        return None
+    hyphenated = phrase.replace(" ", "-")
+    squashed = phrase.replace(" ", "")
+    if (alias := ALIASES.get(hyphenated)) and alias in catalog.slugs:
+        return alias
+    if hyphenated in catalog.slugs:
+        return hyphenated
+    if phrase in catalog.by_name:
+        return catalog.by_name[phrase]
+    if squashed in catalog.compact:
+        return catalog.compact[squashed]
+    return None
+
+
+def match_slug(name: str, images: Iterable[str] = ()) -> Optional[str]:
+    """The catalog slug a stack's name (then its images) points at."""
+    catalog = load_catalog()
+    if not catalog:
+        return None
+
+    phrase = _normalize(name)
+    if (slug := _lookup(catalog, phrase)):
+        return slug
+
+    # The app's name inside a longer one ("my jellyfin stack", "medien plex").
+    # Longest run of words first, so "home assistant" beats "home".
+    tokens = phrase.split()
+    for size in range(len(tokens), 0, -1):
+        for start in range(len(tokens) - size + 1):
+            gram = tokens[start : start + size]
+            # A single short word is far more likely to be a coincidence than
+            # an app ("app", "web", "db" are all slugs somewhere).
+            if size == 1 and len(gram[0]) < 4:
+                continue
+            if (slug := _lookup(catalog, " ".join(gram))):
+                return slug
+
+    # Nothing in the name: ask what the stack actually runs.
+    for image in images:
+        if (slug := _lookup(catalog, _normalize(image))):
+            return slug
+    return None
+
+
+def images_for(stack_id: str) -> list[str]:
+    """Image names a stack's compose file pulls, most specific part first.
+
+    ``lscr.io/linuxserver/jellyfin:latest`` contributes ``jellyfin``: the tag,
+    the registry and the vendor namespace say nothing about which app it is.
+    """
+    directory = compose_service.stack_dir(stack_id)
+    compose_file = compose_service.find_compose_file(directory)
+    if not compose_file:
+        return []
+    try:
+        with open(compose_file, "r", encoding="utf-8", errors="replace") as fh:
+            data = yaml.safe_load(fh) or {}
+    except (OSError, yaml.YAMLError):
+        return []
+    out: list[str] = []
+    for spec in (data.get("services") or {}).values():
+        if not isinstance(spec, dict):
+            continue
+        image = spec.get("image")
+        if not isinstance(image, str) or not image:
+            continue
+        # Strip the tag/digest, then take the last path segment.
+        base = image.split("@")[0].rsplit(":", 1)[0]
+        candidate = base.rstrip("/").split("/")[-1]
+        if candidate and candidate not in _IMAGE_NOISE and candidate not in out:
+            out.append(candidate)
+    return out
+
+
+#: Per-stack results, keyed by what they were computed from. Matching is pure
+#: string work, but it reads the compose file, and the stacks list runs it for
+#: every row on every poll.
+_resolved: dict[str, tuple[tuple, Optional[str]]] = {}
+
+
+def _signature(stack_id: str, name: str) -> tuple:
+    try:
+        mtime = os.path.getmtime(
+            compose_service.find_compose_file(compose_service.stack_dir(stack_id)) or ""
+        )
+    except OSError:
+        mtime = 0.0
+    return (name, mtime, _catalog_mtime)
+
+
+def auto_slug(stack_id: str, name: str) -> Optional[str]:
+    """The logo a stack gets with nothing configured, or None for no match.
+
+    Cached against the stack's name and its compose file's mtime, so a rename or
+    an edited compose re-matches and everything else is a dictionary hit.
+    """
+    signature = _signature(stack_id, name)
+    cached = _resolved.get(stack_id)
+    if cached and cached[0] == signature:
+        return cached[1]
+    slug = match_slug(name, images_for(stack_id))
+    _resolved[stack_id] = (signature, slug)
+    return slug
+
+
+def forget(stack_id: str) -> None:
+    """Drop a stack's memoized match (it was deleted, or renamed by clone)."""
+    _resolved.pop(stack_id, None)
+
+
+# --------------------------------------------------------------------------- #
+# The logo files
+# --------------------------------------------------------------------------- #
+
+#: Slugs currently being downloaded, so N rows asking at once fetch once.
+_inflight: dict[str, asyncio.Task] = {}
+
+
+async def ensure_logo(slug: str) -> Optional[str]:
+    """Path to a slug's cached logo, downloading it the first time."""
+    path = logo_path(slug)
+    if not path:
+        return None
+    if os.path.isfile(path):
+        return path
+    catalog = load_catalog()
+    if catalog and slug not in catalog.slugs:
+        return None
+
+    if (task := _inflight.get(slug)) is None:
+        task = asyncio.create_task(_download(slug, path))
+        _inflight[slug] = task
+        task.add_done_callback(lambda _t, s=slug: _inflight.pop(s, None))
+    try:
+        return await asyncio.shield(task)
+    except asyncio.CancelledError:
+        # The *caller* went away (client disconnected); the download itself is
+        # shielded and still finishes for whoever asks next.
+        raise
+    except Exception:  # noqa: BLE001 - a missing logo is not an error
+        return None
+
+
+async def _download(slug: str, path: str) -> Optional[str]:
+    try:
+        async with httpx.AsyncClient(follow_redirects=True, timeout=TIMEOUT) as client:
+            response = await client.get(LOGO_URL.format(slug=slug))
+            response.raise_for_status()
+            data = response.content
+    except httpx.HTTPError as exc:
+        logger.info("Could not fetch the logo for '%s': %s", slug, exc)
+        return None
+    if not data.startswith(b"\x89PNG\r\n\x1a\n") or len(data) > LOGO_MAX_BYTES:
+        logger.info("Ignoring the logo for '%s': not a plausible PNG", slug)
+        return None
+    os.makedirs(logo_dir(), exist_ok=True)
+    tmp = f"{path}.{os.getpid()}.tmp"
+    try:
+        with open(tmp, "wb") as fh:
+            fh.write(data)
+        os.replace(tmp, path)
+    except OSError as exc:
+        logger.warning("Could not cache the logo for '%s': %s", slug, exc)
+        return None
+    return path
+
+
+def search(query: str, limit: int = 60) -> list[dict]:
+    """Catalog entries matching a search term, best match first."""
+    catalog = load_catalog()
+    if not catalog:
+        return []
+    needle = _normalize(query)
+    results: list[tuple[int, str, dict]] = []
+    for entry in catalog.entries:
+        slug = (entry.get("Reference") or "").lower()
+        name = entry.get("Name") or slug
+        if not _SLUG_RE.match(slug):
+            continue
+        haystack = _normalize(name)
+        if not needle:
+            rank = 2
+        elif haystack == needle or slug == needle.replace(" ", "-"):
+            rank = 0
+        elif haystack.startswith(needle) or slug.startswith(needle.replace(" ", "-")):
+            rank = 1
+        elif needle in haystack or needle.replace(" ", "-") in slug:
+            rank = 2
+        else:
+            continue
+        results.append((rank, haystack, {"slug": slug, "name": name}))
+    results.sort(key=lambda row: (row[0], row[1]))
+    return [row[2] for row in results[:limit]]