On 01/07/16 03:02, John Rouillard wrote:
>> John Kristensen added the comment:
>>
>> I am not entirely convinced that this extra checking is desirable. If
>> the database does not exist I would prefer that the tests fail loudly
>> rather than skip silently (and potentially go unnoticed).
>
> Except that simply having the backend python libraries in place on
> a system doesn't mean you have configured a database.
>
> For example, if I am running roundup using mysql on a system that also
> run some other app using postgres, and I want to test the code before
> install, it will falsely fail postgres tests possibly obscuring other
> failing tests that I care about.

My counter argument to this would be that each application should be 
running in its own virtualenv. Once two applications are relying on a 
common dependency it is only a matter of time before it comes back to 
bite you with version incompatibilities.

> Right now running tests with an explicit -k 'not (mysql or
> postgresql)' is required. But is that guaranteed to only disable tests
> that require a running copy of those two databases? Could it
> accidently trap other tests?

I'll say "maybe". You would probably need to just check which test names 
are matched and hope the naming of those, or new, tests stay consistent 
in the future.

An alternative may be to look at using marks to make this sort of thing 
easier:

   https://pytest.org/latest/example/markers.html

Though given the issues I've had so far with the skip markers, I'm a 
little hesitant to try and implement them and break everything again ;)