Improve report when slaves collect different tests

[pytestbot]

Originally reported by: Bruno Oliveira (BitBucket: nicoddemus, GitHub: nicoddemus)

---

As described in #437 and #301, a test run stops when slaves collect a different number of tests.

This is the code that raises the error:

    # dsession.py:147    
    def init_distribute(self):
        assert self.collection_is_completed
        # XXX allow nodes to have different collections
        node_collection_items = list(self.node2collection.items())
        first_node, col = node_collection_items[0]
        for node, collection in node_collection_items[1:]:
            report_collection_diff(
                col,
                collection,
                first_node.gateway.id,
                node.gateway.id,
            )

If instead of raising an AssertionError in report_collection_diff we just return None if the collections differ, we obtain a better error message.

scheduling tests via LoadScheduling

=================================== ERRORS ====================================
_______ ERROR collecting source/python/ben10/_tests/pytest_fixtures.py ________
.env27\lib\site-packages\py\_path\local.py:620: in pyimport
    __import__(modname)
x:\pytest_cx_freeze\pytest\_pytest\assertion\rewrite.py:134: in find_module
    co = _read_pyc(fn_pypath, pyc)
x:\pytest_cx_freeze\pytest\_pytest\assertion\rewrite.py:312: in _read_pyc
    co = marshal.load(fp)
E   EOFError: EOF read where object expected

Which makes it easy to track down the source of the problem (btw, the above issue has a proposed fix in pull request #192).

---

• Bitbucket: https://bitbucket.org/pytest-dev/pytest/issue/556

[pytestbot]

Original comment by Bruno Oliveira (BitBucket: nicoddemus, GitHub: nicoddemus):

---

Pull request 9 in xdist is a proposal to solve this.