Skip to main content

Module memory

Module memory 

Source
Available on crate feature memory only.
Expand description

In-memory reference implementation of ChatBackend.

Gated behind feature = "memory". Not production. See memory for the full module documentation. In-memory reference implementation of ChatBackend.

§This is a reference implementation, not production

MemoryBackend is intended for three audiences:

  1. Workspace integration tests — every tests/*.rs integration test in this crate exercises method handlers against this backend.
  2. Downstream contributors — a documented, complete, source-readable implementation of the ChatBackend trait to study when writing a real (database-backed) backend.
  3. Examples and smoke tests — boot a real JMAP-for-Chat dispatcher with one line of code, without standing up a database.

It is not suitable for production: all state is held in HashMaps behind a std::sync::Mutex, persistence is not implemented, and a number of JMAP Chat draft edge cases are simplified (see source comments).

§Feature flag and API stability

This module is gated behind feature = "memory" and is not enabled by default. Its public API stability is opt-in: it may break across minor versions while the crate is pre-1.0.

§Example

ⓘ
use std::sync::Arc;
use jmap_chat_server::{memory::MemoryBackend, register_chat_handlers};
use jmap_server::Dispatcher;

let mut dispatcher: Dispatcher<()> = Dispatcher::new();
register_chat_handlers(&mut dispatcher, Arc::new(MemoryBackend::new()));

§Concurrency

std::sync::Mutex is used for simplicity. The await_holding_lock clippy lint is enabled module-wide and enforces that no lock guard is held across an .await. If a future change requires holding a guard across .await, switch to tokio::sync::Mutex rather than disabling the lint.

Every lock() in this module ends in a bare .unwrap() on purpose. Mutex::lock fails only when another thread panicked while holding the guard, so the state behind it may be half-updated; that panic is already a bug in this module, and there is no consistent state to recover. Panicking again surfaces it instead of serving corrupt data. The crate’s tests never poison the lock. A production backend that must survive a poisoned lock should implement the trait with its own locking rather than copy this pattern.

§Tracking

Promoted from tests/common/mod.rs per Beads issue JMAP-hwdv (epic) and JMAP-hwdv.3 (this crate, mirror of canonical JMAP-hwdv.1 in jmap-mail-server).

Structs§

MemoryBackend
A fully in-memory implementation of ChatBackend.
MemoryError
Opaque storage-layer error returned by MemoryBackend operations.

Enums§

LifecycleEvent
One lifecycle callback received by the MemoryBackend, as returned by MemoryBackend::drain_lifecycle_events.

Constants§

MAX_PUSH_SUBSCRIPTIONS
The most push subscriptions the single owner may hold at once.