The GNU Mailing List manager. https://git.neuromancer.sk/mailman.git/atom?h=plugin 2017-08-07T16:18:00Z 2017-08-07T16:18:00Z J08nY 2017-07-04T23:07:37Z urn:sha1:6c621405c88671a58ef24cd84a9bd74ca324207e - This is quite a huge commit, since it changes the type of the MailingList.subscription_policy and unsubscription_policy attributes to the new names of pluggable workflows, in all occurences. - Also adds a migration to migrate the attributes to the new types. - Adds tests for the migration. 2017-07-22T03:02:05Z Barry Warsaw 2017-07-22T03:02:05Z urn:sha1:f00b94f18e1d82d1488cbcee6053f03423bc2f49 2017-01-16T05:24:41Z Barry Warsaw 2017-01-16T05:24:41Z urn:sha1:abab1cb4afd1a0c41110c2e4bcd11cf74ef05775 2017-01-09T14:07:29Z Barry Warsaw 2017-01-09T14:07:29Z urn:sha1:beebdf29317a8df4fcfabd0ee4f01cb7eb6247d3 Hopefully this will help fix intermittent failures on the Gitlab CI runners. 2 minutes appears to be too short since the MySQL tests can take longer than that to get to the lock claim attempt. 2017-01-04T15:30:17Z Barry Warsaw 2017-01-04T15:30:17Z urn:sha1:8180990a2ed053f6e4880f93a75acd9845631793 2016-12-03T16:08:18Z Barry Warsaw 2016-12-03T16:08:18Z urn:sha1:a30bbdf0d483150dc4259b8d149679d38b6030f1 This doesn't solve the underlying problem of why child processes (both runners and master) occasionally don't exit, but it prevents ugly timeouts in the CI infrastructure when that happens. 2016-10-21T15:06:46Z Barry Warsaw 2016-10-21T15:06:46Z urn:sha1:f036d465725204a294a6e2f582750aaab875eb18 2016-10-17T13:13:32Z Barry Warsaw 2016-10-17T13:13:32Z urn:sha1:d45af03c4f2a560d51631fdfa7c55cd1a98e722c This merges the SubscriptionManager and UnsubscriptionManager into a single SubscriptionManager implementation that handles both register() and unregister(). This allows us to use direct class-based adaptation instead of the more clunky getAdapter() API. We can also eliminate the funky _get_workflow() implementation detail. This has a couple of side-effects. .confirm() must lookup the token in the pendings database and pull out the pending type, dispatching to the proper class depending on the type, or raising a LookupError if the token is None or there is no pendable associated with the given token. This feels like an acceptable trade-off. However, this *also* means that IWorkflowStateManager must lose its 'name' argument in its methods. That's because we won't actually know the name until its too late. Honestly, the name wasn't providing much value anyway (it was always the subclass's name), so losing that seems fine too. The complication here is that the name was a primary key in the 'workflowstate' table, so we need to add its removal in the database migration. 2016-09-01T11:39:10Z Barry Warsaw 2016-09-01T11:39:10Z urn:sha1:9f7fa58e81c2bc639645383a12b8136e2cf2cb4b 2016-09-01T10:55:09Z Barry Warsaw 2016-08-31T01:02:06Z urn:sha1:84c0f9532ae20d0f5ee8a8ecd9a92e42925b174b