CHANGELOG.md

v0.105:

• TBA

v0.104:

• more extensive testing (CPython3.8, nightly)
• fixed a bug wherein future_to_notify_on_dead was set_result multiple times
• switched to nose2 for tests
• removed unnecessary print()s

v0.103:

• bugfix in handling exchange in publish

v0.102:

• due to noticed behaviour on some Linuxes that changing epoll from another thread than is being waited on does not seem to alter the behaviour, EpollListener will now check manually if threads have anything to send

v0.101:

• bugfix: a race condition during connection setup itself
• moved version to coolamqp root
• split compile_definitions into a separate package
• exceptions will display their reply_text correctly if fed a memoryview
• added Docker-based tests
• far more robust wait in Cluster.start made

v0.100:

• significant docs update
• cosmetics
• fixed a bug wherein on very much loaded systems ConnectionStart would arrive before a watch could be registered for it

v0.99:

• bugfix release: extra requirements added to setup.py

v0.98:

• bugfix release: fixed multiple race conditions, added stress tests

v0.97:

• Changed copyright in connection properties to better reflect the current situation
  • also, noted that in README.

v0.96:

• Python 3.6 and 3.7 officially supported and tested against. Also same done for PyPY 3.5
• fixing #40
• added #32

v0.95:

• multiple bugs fixed

v0.94:

version skipped

v0.93:

• Large refactor of XML schema compiler

v0.92:

• Added on_fail event handler - fired upon connection loss

v0.91:

• removed annoying warnings

v0.90:

• first release with a stable API
• You can pick how your received message.body will look like - bytes, a zero-copy-for-short-messages memoryview, or maybe a totally zero-copy list of memoryviews?

v0.89.1:

• Critical bugfix: messages larger than a frame got corrupted

v0.89:

• Events are no longer timestamped by CoolAMQP, it's your job now
• You can delete queues (Cluster.delete_queue)
• Race condition Connection.start fixed
• Queue can accept bytes as name
• Consumer will set cancelled to True if Consumer Cancel Notification is received
• You can register callbacks for:
  • Consumer being cancelled for any reason
  • Consumer being cancelled with a CCN

v0.88:

* Cluster.start will RuntimeError if called more than once
* Cluster.shutdown will RuntimeError if called without .start
* Warning with content list is shorter