MidnightSdkProvider

Process-wide owner of the single live MidnightSdk instance.

Why this exists. Before this, every consumer that needed a wallet built its own SDK — the wallet panel and BBoard each constructed one, so the app opened two indexer WebSockets, replayed zswap from genesis twice, and ran two dust DBs. Two instances of the same wallet on the same seed and network was never intended; it was duplication, not redundancy. This provider holds exactly one SDK and hands it to all consumers.

Ownership model — one authority, many followers.

• The config authority (the wallet panel — the only UI with the network / proving-mode / proof-server toggles) calls ensureSdk with a full WalletConfig. That is the sole place config changes enter the system, which is what makes activeConfig a single source of truth.

• Followers (BBoard's contract ops, Kicks's match manager) don't own config. They call awaitSdk for the shared handle and read activeConfig for "which chain am I on". They never pass a competing config, so the provider never thrashes between two configs.

Lifecycle. The SDK is a process-singleton, not view-model-scoped. It survives activity recreation (rotation) — a net win, since rebuilding means a fresh biometric prompt + resync. Teardown is therefore explicit via close (logout / wallet reset). No caller invokes close today; it's the seam a future session-auto-lock (idle timeout) will use.

See also