Drops forecasts, lightning, the home-screen widget and advertising. What remains is radar on a map, the official ARPA alert bulletin, and rain notifications. The base map is now OpenFreeMap's Positron style: real OpenStreetMap vector tiles with no API key, no registration, no request limits and commercial use permitted. Every other free tier — MapTiler, Stadia, Jawg, Thunderforest — needs a key, which is a secret to manage, a quota to outgrow and a signup to complete before anyone can build the project, and the map is the one thing the app cannot work without. Positron rather than Liberty or Bright because the radar overlay has to be the loudest thing on screen, and a desaturated grey base is built to sit under data. Its style JSON carries no `attribution` field, so MapLibre displays no credits by itself. The app renders them from the region config instead: the two mandatory credits, OpenStreetMap and OpenMapTiles, go in the always-visible bar, and OpenFreeMap's own credit — optional by their terms — is listed on the Sources screen with the rest. The bundled offline style is still reachable with MAP_STYLE_URL=offline, and still claims no base map attribution, because crediting OpenStreetMap while showing it would be a false claim. Radar-DPC stays the source. ARPA Piemonte's own radar remains a disabled stub for two reasons that belong to the project owner, not to the code: the real-time access link is only issued by email, and the open-data page states the data is "gratuiti" and nothing else. Free of charge is not a licence, and rendering those volumes into frames served from a CDN is redistribution. Both questions go in the same email. An earlier draft of the docs recorded ARPA radar as CC BY 4.0; the source page does not support that, so the claim is removed rather than carried forward. The documentation is updated throughout rather than annotated: CLAUDE.md gains an explicit scope boundary, data-sources drops MET Norway and ISTAT and gains the base map, licenses records that free of charge is not a licence, privacy loses the whole advertising section, and the roadmap is renumbered so the backend worker is next — until it exists, DpcRadarSource has nothing to read. licenses.md keeps Open-Meteo and Blitzortung listed as excluded even though the features that would have used them are gone: both are non-commercial-only, ads are a plausible future, and neither should be adopted on the grounds that there are none today. Verified: analyze clean, 130 tests passing, and on the emulator the radar overlay sits correctly over Piedmont on real OSM tiles with Turin, Milan and Genoa labelled, the age reads "Aggiornato 4 minuti fa", and the Sources screen lists all five credits with their licences. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
6.8 KiB
Stack decisions
Each entry records what was chosen, what it was chosen over, and why. Revisit an entry only with a reason that invalidates its rationale.
Flutter, single codebase
Android ships first, iOS follows on the same code. Platform-specific work stays behind
interfaces (core/platform/) so the iOS port is additive rather than a rewrite.
Toolchain pinned during setup: Flutter 3.47.3 stable / Dart 3.13.3, compileSdk/targetSdk 36, minSdk 24, JDK 21.
maplibre_gl only needs API 21; 24 is the floor firebase_messaging imposes
for the rain notifications, and it is Flutter's own default, so it is the
best-tested path.
State management — Riverpod without code generation
flutter_riverpod with hand-written Notifier / AsyncNotifier classes.
Chosen over: Bloc (more ceremony than this app's state needs), plain setState
(the radar timeline, the frame cache and the alert state all cross screen
boundaries), and Riverpod with riverpod_generator.
Dropping code generation keeps build_runner out of CI and out of every edit-run
cycle. The generator's benefit — less boilerplate on providers — is small at this
size, and its cost is paid on every build. Reconsider if provider count passes ~40.
Networking — Dio
dio with interceptors for retry with exponential backoff and timeouts.
Chosen over http for the interceptor model: cross-cutting request policy belongs in
one enforced place rather than at each call site where it can be forgotten.
The DPC origin header lives only in the Python worker. The app never talks to
DPC directly.
Models — hand-written fromJson
Chosen over freezed + json_serializable. The model set is small (radar manifest and
frames, CAP alerts, region config) and the parsers are covered by
tests against real fixtures. Same rationale as Riverpod: no build_runner.
The tests, not the generator, are what guarantee the parsing is right — fixtures captured from the real endpoints catch schema drift that codegen would not.
Map — MapLibre GL
maplibre_gl 0.27.0 (Flutter 3.29+, Android API 21+, iOS 13+).
Chosen over flutter_map, which renders tiles in Dart. Radar animation redraws a
full-viewport image several times a second; a GPU-composited native renderer holds
frame rate where a Dart canvas does not.
Base map — OpenFreeMap Positron
OpenStreetMap vector tiles from OpenFreeMap: no API key, no registration, no request limits, commercial use permitted.
Chosen over MapTiler, Stadia, Jawg and Thunderforest, whose free tiers all require a key. A key is a secret to manage, a quota to outgrow and a signup to complete before anyone can build the project — and the map is the one thing the app cannot work without.
Positron rather than Liberty or Bright: the radar overlay has to be the loudest thing on screen, and Positron is a desaturated grey base built to sit under data. On a full-colour style the precipitation ramp competes with road casings and landuse fills.
The trade is a dependency on someone else's free service. The escape hatch is
self-hosting — OpenFreeMap is MIT-licensed and publishes its planet tiles — and
MAP_STYLE_URL already points the app anywhere else without a code change.
Its style JSON carries no attribution field, so MapLibre shows no credits by itself and
the app renders them from the region config. See licenses.md.
MAP_STYLE_URL=offline selects a style generated from the region bounding box, with no
network sources at all: a flat background and the extent outline. Deliberately plain so
it is never mistaken for a finished map, and it is what the widget tests run against.
Radar frame rendering — image source, double buffered
The worker publishes each frame already cropped to the region bounding box and
reprojected to EPSG:3857, so the four corners of a MapLibre LatLngQuad are exact and
no client-side warping is needed.
Animation uses two image sources, A and B: while one is visible the next frame is decoded into the other, then visibility swaps.
Chosen over: adding all ~20 frames as layers with opacity 0 (constant GPU memory matters more than the saved swap — 20 frames at 1024×1024 RGBA is ~80 MB resident), and over updating a single source in place (visible flicker during decode).
Frame cache — custom in-memory LRU
FrameCache: an in-memory LRU of PNG bytes with a byte budget, plus retainOnly so
the timeline can drop everything outside the window around the playhead.
The budget is in bytes rather than entries because frame size tracks how much precipitation is on screen — a clear sky compresses to almost nothing, a storm does not — so an entry count would let a stormy loop use several times the memory of a calm one.
Chosen over flutter_cache_manager, which does not expose the eviction control the
scrubber needs. Plain LRU keeps frames the prefetcher touched a moment ago even after
the playhead has moved to the far end of the timeline, so eviction is driven by
distance from the playhead, not by access time.
Deliberately not persistent. Mock frames already live in the asset bundle, and network frames are re-fetched from the CDN. A disk layer belongs with the network adapter, where it would actually save a request, and is worth adding once real CDN frames are flowing.
Region configuration
Everything region-specific lives in app/assets/regions/<id>.json: bounding box, map
centre and zoom limits, alert zones, active data sources, attribution strings. Adding
a region is a new JSON file plus its assets — no Dart changes.
Piedmont bounding box, padded: [6.55, 43.95, 9.30, 46.55] (W, S, E, N).
It is a render extent, not an administrative boundary: it deliberately overshoots
the region so nothing is clipped at the edges, and the tests assert that every
provincial capital falls inside it.
Backend — Python worker on a VPS
GDAL and HDF5 cannot run on Supabase Edge Functions (Deno), so the worker runs as a scheduled process on a VPS or Cloud Run.
rasterio 1.5.1 publishes Windows wheels for Python 3.14, so the worker also runs
natively on the development machine — no Docker or WSL needed to iterate.
Trigger: the DPC WebSocket push channel wss://radar-wss.protezionecivile.it, with a
5-minute cron as fallback. Polling DPC on a timer is the fallback, never the norm.
Output goes through a Publisher interface: LocalPublisher writes to backend/out/
for development, S3Publisher targets R2 or Supabase Storage once credentials exist.
Rain notifications — client-side topic subscription
The worker computes rain per geographic cell and publishes per-cell state. The app subscribes to FCM topics named after cells, from the device.
The server therefore never learns any user's position — not precisely, not even by cell. There is no user table to leak, and Data safety can honestly declare that no location is transmitted or stored.