Hi Bern:

In message <1571039971.84.0.988568062794.issue2550905@roundup.psfhosted.org>,
Bernhard Reiter writes:
>Bernhard Reiter added the comment:
>
>Note: While it is good to have some code coverage, 100% is considered
>suboptimal (in the only research I could find, reported by Robert Glass,
>but it fits my personal experience).

I agree. I would be happy with 80-90% coverage. Some areas I would
expect not to be tested:

  exception handlers - e.g triggering random disconnects from remote services
           is probably not going to happen.

  debugging code - probably routines used by the exception handling code.

I have my demo instance running under the profiler to see what code is
actually being hit often. That would be the first code that needs to
be put under test.

We need testing for external interfaces: web, email (pop, imap) and
the way to do these tests is not obvious. We also run the risk of
mocking/developing an external test harness that is broken. So
"passing" code that uses these harnesses is actually broken in real
world use.

With the python 3 changes to the web output, tal (which has almost no
coverage), jinja and chameleon which have no coverage) we are at
greater risk of producing unusable code. It is better to find the
breakage while the original modifiers are around rather than delay
months, years. As years later we have lost the reason for the
modifications if the original people are no longer available.