_ (underscore) is not supported by apache as domain name
(https://bugzilla.redhat.com/show_bug.cgi?id=1410130), so avoid using it.

The build of nodejs uses python, and python leaking from the system can lead
to unexpected results, thus use provided from SlapOS.

/reviewed-on nexedi/slapos!319

The correct is use require and not required.

/reviewed-by: TrustMe

@alain, @rafael 

Test is here: https://nexedi.erp5.net/test_result_module/20180412-3688F368/2/view

Its failures which are different from  master one (https://nexedi.erp5.net/test_result_module/20180412-3501D1A/1) seems unrelated to my changes thus I propose to emrge it in and include in next webrunner release.

/reviewed-on nexedi/slapos!317

This reverts commit 1bdac5da.

This commit breaks webrunner resilience importer. This is not the proper fix

Because of this: https://lab.nexedi.com/nexedi/slapos.core/blob/master/slapos/grid/promise/generic.py#L44

When slapgrid run promises, it now save promise logs. Thus we need to logrotate them.
Thanks @Alain.

This recipe will be used to generate promise in etc/plugin dir. It solve the problem of promise eggs
dependencies and allow to set custom parameter to use in promise.

The generated script will looks like:

    import sys
    sys.path[0:0] = [
      ...
    ]

    extra_config_dict = {
      'KEY': 'VALUE'
    }

    CONTENT

CONTENT is a python code, the expected content looks like:

    from namespace.module import RunPromise

then the promise section in buildout will be something like:

    [my-promise]
    recipe = slapos.cookbook:promise.plugin
    eggs =
      NAME
      ...
    output = OUTPUT
    content =
      from namespace.module import RunPromise

    config-KEY = VALUE

Instead of having running test installing eggs with setuptools during `python setup.py test`, which fails for `caucase` because one of its dependency (namely `cryptography`) cannot be installed so easily, we want to install the dependencies via buildout, using the same installation methods than in the actual software profiles using our eggs.

What I initially believed would be a small change turned out to be a long journey, especially because of using a develop version of `slapos.recipe.cmmi` caused signature changes in the parts installed by this recipe ( `git`, `openssl` etc - some parts that takes a bit of time to install ) so I had to fight with the software being reinstalled each time. Even though the cases which leads to reinstallation are understood (this even involved fixing a bug in buildout nexedi/slapos.buildout!14 ), this is still the case. Some solutions are proposed in the commit message of f4b6eedafd52b0d64f8b2bd1b57c752e02c88a65 , but this reached a state where we can consider first merging with this known problem or discuss ways of improving that before merging if it's considered as blocking. Current state is that it takes one hour to re-install what has to be reinstalled and run the test suite on test runner.

Despite this issue, this approach already improve things, because:
 * `caucase` tests are running (and passing)
 * some `slapos.cookbook` are improved so that they reuse eggs from the buildout and do not install again eggs when the are run.
 *  ... and some other small cleanups

This depends on nexedi/erp5!619 on the `erp5.util` side.

/reviewed-on nexedi/slapos!309

This function factorize the code to instanciate a recipe in a fake
buildout.
The new feature is that if running in a buildout directory, the recipe
will reuse the eggs from this buildout instead of trying to install eggs
again.

we now use slapos.cookbook:wrapper instead

Instead of letting `python setup.py test` install the depencies, use
buildout way of installing the eggs.

This software use `interpreter` recipe of `zc.recipe.egg` to install a
python with all eggs pre-installed. This is a way to get all the
dependencies at install time instead of getting them at run time from
pypi when running `python setup.py test`.
`erp5.util.testsuite` has been extended to support a parameter to
specify which python interpreter to use.

One issue is that this way of installing eggs by buildout cause chicken
and egg problem: cloning repository containing `slapos.recipe.cmmi`
needs git, and to compiling git needs `slapos.recipe.cmmi`.
The consequence of this is that re-running software will install too
many parts again.
One solution for this would be to clone `slapos.recipe.cmmi` with a
`git` command provided by testnode or system package.
Another solution would be to not install `slapos.recipe.cmmi` develop
egg, simply install the egg from it's current pypi version while
installing the software (running tests will be from the git checkout
anyway).
For now this is open issue.

Another point of attention is that `python setup.py test` install the
requirements listed in `test_requires`, but `zc.recipe.egg` does not
provide a way of installing these. Some of our packages have `[test]`
entrypoints, in this case, the software installs the test entrypoints.
For others, we install the eggs.

Other improvements:
 * use a simple `slapos.recipe:wrapper` instead of `slapos.cookbook:egg_test`
 * fix the typo in repository name erp5-util-repository ->
   erp5.util-repository ( this mean we will have to fix the test suites in
   nexedi ERP5 )
 * document "what is this software" and a scenario of how this software
   can be used to develop slapos eggs.
 * switch to buildout-hash.cfg for easier template hash management.

/reviewed-on nexedi/slapos!308