element-info incorrectly errors when a dependency is virtual and the provides comes after the dependency

[Jon-Paul Sullivan]

This is due to the dependency being added to the queue and examined before the user-input has been fully parsed to see what is provided by them.

This also shows a weakness where the dependencies of a provided element are loaded from the element of the name of the dependency before the dependencies of the element providing that are loaded, potentially leading to unwanted elements in an image.

Example:

The following is a simple change to the provides test that shows the incorrect dependency behaviour:

        _populate_element(self.element_dir,
                          'provides_virtual',
                          [],
                          ['virtual'])
        _populate_element(self.element_dir, 'requires_virtual', ['virtual'])
        _populate_element(self.element_dir, 'virtual', ['extra_dependency'])
        _populate_element(self.element_dir, 'extra_dependency', [])

    def test_provide(self):
        result = element_dependencies.expand_dependencies(
            ['provides_virtual', 'requires_virtual'],
            elements_dir=self.element_dir)
        self.assertEqual(set(['requires_virtual', 'provides_virtual']), result)

The following is a simple test to show the incorrect provides behaviour:

        _populate_element(self.element_dir,
                          'provides_new_virtual',
                          [],
                          ['new_virtual'])
        _populate_element(self.element_dir, 'requires_new_virtual', ['new_virtual'])

    def test_provide_virtual_ordering(self):
        result = element_dependencies.expand_dependencies(
            ['requires_new_virtual', 'provides_new_virtual'],
            elements_dir=self.element_dir)
        self.assertEqual(set(['requires_new_virtual', 'provides_new_virtual']), result)

[Brent Eagles (beagles)]

I've reset this to "nobody" and putting on list for re-triage (see https://etherpad.openstack.org/p/tripleo-bug-cleanup-2016)

[Ben Nemec (bnemec)]

This is a diskimage-builder bug. If it still exists, it should be opened against that project.