Allow to inject warning filters during nosetest.

Put the same arguments as warnings.filterwarnings in setup.cfg at the root of your project. Separated each argument by pipes |, one filter per line. Whitespace are stripped.

for example:

[nosetests]
warningfilters=default         |.*            |DeprecationWarning |notebook.*
               ignore          |.*metadata.*  |DeprecationWarning |notebook.*
               once            |.*schema.*    |UserWarning        |nbfor.*
               error           |.*warn.*      |DeprecationWarning |notebook.services.contents.manager*

If you prefer another name for the configuration file, you can tell nose to load the configuration using the -c flag: run the tests with nosetests -c nose.cfg.

details configuration.

Each line of warning filter is separated in maximum 4 sections, that match the first 4 sections of filterwarnings:

filterwarnings(action, message="", category=Warning, module="", lineno=0, append=False)

fields 2 to 4 can be omitted, ie to say 1 line can be of the following form:

action
action| message
action| message | category
action| message | category | module

the value of each fields is treated the same as for filterwarnigns except: - whitespace are trimmed. - if the category has dots, the corresponding class try to be imported. If it does not have dots, the name is looked up in builtins or __builtins__

test are failing

For some reasons in some systems tests are failing; it seem that this package have difficulty to self-test. That's likely due to the fact that the tested package need to be in different namespaces, and by self-testing we break this assumption.