I had to pull a diff out of the old branch, apply it, and reverse things
like the addition of logging. This needs to be the minimum change for
the fix.
Tested with a quick login, then spamming market buy/sell orders. They
were correctly queued and then sent after 30s since previous API calls.
This remove all dependencies on the keyring lib, updates the
requirements.txt to reflect that, and ensures that setup.py does not
attempt to package it.
Any use of the "old" keyring code will now return None and warn about
its deprecation.
timeout_session provides two things, TimeoutAdapter, a HTTP adapter
subclass that automatically adds timeouts to all requests, and
new_session, which automatically creates a request.Session with the
adapter in the correct place.
* find_caller_frame() to do the frame walk.
* munge_module_name() to fix up for plugins.
NB: caller_attributes() now has a noqa on CCR001 as I don't think it
can sensibly be made any less complex. Pulling out the 'if frame:'
section just results in *that* new method then being labelled as too
complex.:244
* Settled on `plugins.internal` and `<plugins>.found` as the format.
* A PyCharm recommendation was to use 'cls' instead of 'self' on class
methods, so the class detection code needs to cater for that.
Technically a developer could use any string for the "myself" member
name, but we'll assume just these two.
* Found will always have at least one folder level within plugin_dir
* Internal should always have *no* folder within internal_plugin_dir,
but cater for it just in case in future.
1. This makes setting up logging everywhere slightly more involved.
2. If I then want to change, say, %(module)s value I'll end up needing
to stack walk again.
So this might be better done in a filter. But these commits for the
record, and to come back to if needs be.
* Log messages propagate up Parent.Child chains, so we don't need a
channel on the plugin logger.
* But it still needs the filter to define qualname and class for
formatting.
