GitLab

For one, it makes no sense (it won't be visible from outside the
function anyway), for another, it didn't work, because it was called
from different scope (inside the FailingStatsHttpd class) and mangling
the name wrongly.

it would only pass at the same time zone as that of the original developer.

it was a leftover from an intermediate attempt; the test should pass
without it.

see the added comments about the issue.

also prevented open sockets leaving open at the of tests (for supressing
warnings)

this case is a bit tricky.  we need to fake select.select() and delegate
test sub test cases (about command handling) to specific test cases;
other than that the sense of the test should be preserved.
for clarity the failure case was extracted into a separate test method.

some asserts were simplified at an (I hope) acceptable level.  In particular,
I didn't think we need to check the response from CfgMgr: it should be more
than enough if the module name is the expected one.

it now doesn't depend on the clumsy thread stuff (except for http server
and client).

test_command_show and test_polling cases were a bit tricky.
they are now separated into several independent sub tests, and in each case
we use faked data that would be provided as if they came from other modules.

Clarify high-level design of IPC

it's a leftover of an intermediate attempt of testing the case; on a closer
look, testing send_command turned out to be better and sufficient.

It was mostly an example and is now merged to the base version. No
longer needed.