Permalink
Commits on Nov 25, 2009
  1. Send benchmarks

    majek
    majek committed Nov 25, 2009
Commits on Nov 18, 2009
Commits on Nov 9, 2009
Commits on Nov 6, 2009
Commits on Oct 22, 2009
  1. Use heartbeats in demo_relay.

    tonyg committed Oct 22, 2009
  2. Implement heartbeating.

    tonyg committed Oct 22, 2009
  3. Rearrange call to _codec_repr.

    tonyg committed Oct 22, 2009
  4. shutdown_event_loop becomes disconnect_transport, and we ensure that …

    tonyg committed Oct 22, 2009
    …on_disconnected is called after an explicit disconnection.
Commits on Oct 21, 2009
Commits on Oct 20, 2009
  1. According to the letter of the spec, the server is permitted to accept

    tonyg committed Oct 20, 2009
    a client protocol version higher than it supports itself, and to
    respond with a Connection.Start method. For some reason, RabbitMQ's
    0-8 branch implements this behaviour, so when we ask for 0-9-1, we get
    a 0-8 Connection.Start back from the 0-8 server. The fix is to pay
    attention to Connection.Start's contents, and is done here on the
    master branch so that it can be merged into the amqp_0_9_1 branch.
Commits on Oct 19, 2009
  1. The great renaming

    tonyg committed Oct 19, 2009
  2. Change .hgignore to .gitignore

    tonyg committed Oct 19, 2009
  3. Done, with BlockingConnection.

    tonyg committed Oct 19, 2009
  4. Minimal tweaks to tests

    tonyg committed Oct 19, 2009
  5. Avoid firing the state change handler on registration if no

    tonyg committed Oct 19, 2009
    transitions have happened yet.
  6. Correct missing import.

    tonyg committed Oct 19, 2009
  7. Occasionally, recv gives EAGAIN, and immediate retrying doesn't help;

    tonyg committed Oct 19, 2009
    so this patch returns to the main loop, where we run through another
    select() before trying recv again.
  8. Automatic reconnection.

    tonyg committed Oct 19, 2009
  9. Add connection.ensure_closed(); avoid potential race in channel._rpc;

    tonyg committed Oct 19, 2009
    report event handler errors by default; reset pending lists on
    disconnection too, as there's no point in attempting relaying of
    pending deliveries when the source isn't connected, since when the
    source reconnects it's just going to send them through again.