feat: Add an example app and misc fixes (#81)

This commit is contained in:
Marcelo Mendonça Soares 2026-06-23 22:53:47 -03:00 • committed by GitHub
commit de7e22d91a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194

View file

@ -0,0 +1,98 @@
# Offline Sync Demo
Desktop Flutter demo for `serverpod_offline_sync`. It runs two independent local
replicas of one user's data side by side, plus a view of the server's merged
truth, so you can drive `A → server` and `B → server` independently and watch
them converge.
## How the package is wired
If you are here to learn the package, read **`lib/src/offline_replica.dart`**
first — it is the entire integration in ~60 lines, deliberately free of demo and
UI code. Every replica follows the same four steps:
1. **Open a local database.** `client.createSession(path)` opens a Serverpod
client database (a local SQLite file) and runs the client migrations.
2. **Wrap it for CRDT sync.** `CrdtDatabaseSession.wraps(raw, syncTables: …,
persistentUserId: …)`, then `crdtSession.db.initialize()` to establish this
device's CRDT node so it can take part in sync.
3. **Read and write generated models** against that session, e.g.
`Person.db.insertRow(session, person)` or `session.db.find<Person>()`. (The
demo also routes these through a generic `TableOps` registry so its
metadata-driven UI can touch any table without a per-type switch — that part
is demo-only; the `seed*` methods in `lib/src/demo_controller.dart` show the
plain form.)
4. **Synchronize through a Serverpod client.**
`client.crdt.syncOnce(session, onMergeSuccess: …)` for a one-shot push/pull,
or `client.crdt.syncContinuously(session, …)` to stream until cancelled.
Connectivity is just *which client you sync through*: the demo swaps
`Client.httpClientOverride` between a real transport and a failing one to
simulate going offline — the local database is untouched either way.
The rest of `lib/src` layers a UI on top: `demo_controller.dart` (a
`ChangeNotifier` orchestrating users, replicas, and the server panel),
`snapshot.dart` / `relationships.dart` / `table_ops.dart` (the metadata-driven
view of the synced schema), `demo_view.dart` / `sheets.dart` (widgets), and
`scenarios.dart` (guided scripts).
## What the UI shows
Each demo user owns two replicas (`<user>-a.db`, `<user>-b.db`) shown side by
side. A replica is a local SQLite store with its own CRDT node id but the same
sync scope; replicas share nothing locally and only converge by syncing through
the server — like two phones on one account (see the "i" button in the toolbar).
The layout is a **scenario rail**, **Replica A**, **Replica B**, and a **Server**
panel, with a global toolbar on top and a status line at the bottom.
**Toolbar**
- **Connectivity** toggles every client between a real and a failing HTTP
transport (the offline switch).
- **Show hidden rows** reveals locally hidden / soft-deleted (tombstoned) rows,
shown struck-through and muted alongside the visible "user view".
- **Refresh** reloads all three panels; **Reset all** wipes both replicas and the
server scope at once.
- The **i** button explains replica isolation.
- The app bar carries a light/dark toggle and an avatar menu for switching
between or creating demo users.
**Per replica**
- **Sync** runs one push/pull; **Stream** keeps syncing continuously until
toggled off — drive A and B independently and watch them (and the Server
panel) converge.
- **Add data** creates a root row or attaches a child through a foreign key; the
same menu also offers ready-made seeds (a city/person/address/town graph, a
typed row, an FK chain). Each row carries inline **attach child** and
**re-parent / detach** actions, and every edge shows its `onDelete` action
(restrict / cascade / set null / …).
- **Reset** wipes just that replica's local database (a fresh device); syncing
afterwards re-pulls whatever the server still holds.
- Click any row to open its full record, edit fields, or delete it.
**Server panel**
- Shows the server's merged truth for the current scope, fetched through the
`demoDebug.fetchScopeSnapshot` endpoint (reads run in the caller's scope via
`transactionForUser`). It refreshes after each sync and via its own refresh
button.
- **Reset server scope** hard-clears every row and CRDT metadata record in the
scope; the seed buttons write a graph straight into the scope via
`demoDebug.seedScope`, so you can reset a replica and sync to exercise the
fetch-from-scratch flow.
**Scenario rail**
- Guided scripts (restrict, set-null, cascade, unique-conflict, …) that run a
sequence of these same actions step by step. Every step is also doable by hand.
## Running
Simply run the serverpod server and the app will be started together:
```sh
cd test/serverpod_offline_sync_test_server
serverpod start
```