Retry on ConflictError is broken

[icemac]

Retrying of requests in case of an ConflictError is broken on multiple levels:

• HTTPRequest.retry_max_count = 0 meaning no retries by default. This default is not not changed when setting max_conflict_retries in zope.conf See [1] below.
• If an exception view for Exception is reqistered it has to reraise ConflictException otherwise it will not be handled by the retry mechanism. See [2] below.
• If a request is actually retried the newly created event gets closed before processing it. See [3] below and #414.
• ConflictError is a subclass of TransientError, the code using both can be simplified. See [4] below.
• HTTPRequest.supports_retry() sleeps when called. See [5] below.

[1] The default used to be 3 in Zope 2.13 but was changed in 740eb60 to 0. In 40e5f99 the ability to change the default was added but it sets the value on ZPublisher.HTTPRequest (the module) instead of ZPublisher.HTTPRequest.HTTPRequest (the class), so it does not have the expected effect.

[2] This seems unexpected to me. Maybe ConflictError should be handled separately like it is already done for Unauthorized.

[3] The request.close() in the finally block (see WSGIPublisher.publish_module) closes the newly created request, so it is empty during the retry.

[4] Since zopefoundation/ZODB@364b2a2 a ZODB ConflictError is a subclass of transaction's TransientError. The commit is part of ZODB 3.10+.

[5] I'd not expect that a method called supports_retry does the actual sleeping. It is called three times when a ConflictError happens.