The "\5c" sequence needs to be escaped first because other sequences like "\20" are escaped to "\5c20" and introduce more "\5c" sequences into the string which later would be escaped a second time to "\5c5c"

This makes it a tad easier to use in the common case where more
than two clients take part in the exchange.

Using @require_identity may lock onto an identity on a different
entity (for example, the domain) instead of the user’s account.

Since PEP requires interaction with the user’s account, the
identity check needs to be done there, too, which is what
@require_pep does.

This provisioner is intended for use with servers which do not
provide the full feature set with SASL ANONYMOUS (or don’t support
it at all) and do not have a mod_auth_any equivalent.

Fixes #328.

The use case for this is in other people’s CI. The non-e2e tests
are sometimes timing based and thus inherently flakey. The
e2etests are not that, since they’re based on interactions with a
real server.

Fixes #327.

There is no way to detect availability of XEP-0049 on the server
side. The newly added quirk allows to disable tests on servers
which do not support XEP-0049 at the moment or with the given
configuration.

Fixes #329.

The Travis CI pipeline passed, but due to the .org -> .com switch,
this is not visible in the UI.

Yay for Python 3.5+!

- Drop support for Python 3.4
- Raise the default test version to 3.7

Even python3.4 has it at the new location.

Closes #278

In contrast to the classic echo_bot.py example, this auto-enables
and uses message carbons.

If the remote server is unreachable network-wise, we cannot know
whether we have been removed. In addition, a rejoin will fail
deterministically.

It is thus better to stay in the undecided pinging state (treat it
like a timeout): this will allow us to wait for a rejoin until we
know more about the current state and the server is actually
reachable again.

See-Also: https://github.com/xsf/xeps/pull/834
See-Also: #312

Previously, the code would force coercion of the value through
the *nested* type (in most cases, String) first even if the value
was already an enumeration member. This obviously breaks because
the enumeration member is an instance of the Enumeration and not of
any other type.

This bug occurs only if deprecate_coerce is *false*, because
otherwise there was an additional check for enum membership in the
code path.

Those tests are awful hacks to test that the stream is healthy
after certain edge-cases. However, since the data is injected into
the stream without going through the server, the SM counters will
be off; we must thus disable the tests if SM is enabled.