Collect dependencies

[ftesser]

Hello @RKrahl, the purpose of this PR is to add the possibility of automatically collect dependent tests.

Is similar to what proposed in #37, but differs from these points:

• the behaviour is automatically spread on all tests (no need to add the mark parameter collect=True)
• the behaviour is enable/disable by a configuration that can be set in the ini file;
• the searching of dependent tests is recursive;

This configuration option default is set to false, so pytest-dependency does not change its default beauvoir.

Just to explain my motivation for this PR, my use case is the following:

I use PyCharm IDE, running a single test is made by clicking on the green triangle on the left of the function test definition (see image below):

Using this new feature the result for a single click is the execution of that test and all the dependent tests. I think that most IDEs have similar way to run a single test, and so this option can have its benefit also there.

[dvarrazzo]

Looks like the author of this MR and me are going after the same idea.

See this comment for my explanation about why this feature would be really useful for a pytest dependency-tracking extension.

[dvarrazzo]

Hello! Any chance for this feature to be developed? Can we provide some help to release this useful feature? Thank you 🙂

[quanthom]

This feature is very useful, what is blocking the CI failures to be fixed and the branch merged?

[marcinwrochna]

I believe pytest_collection_modifyitems should me marked with @pytest.hookimpl(trylast=True), otherwise it may happen before items are deselected with the default built-in pytest key/mark selections (it happened to me).

[ydirson]

This code does not take into account tests in a class, when referred to as follows:

class Tests:
    @pytest.mark.dependency()
    def test_b(self):
        pass

    @pytest.mark.dependency(depends=["Tests::test_b"])
    def test_d(self):
        pass

The following Q&D patch takes care of that case, but I'm not sure it would handle all situations:

                     depend_parent = Module.from_parent(item.parent, fspath=depend_path)
                     depend_nodeid = depend
                 else:
-                    depend_func = depend
+                    if "::" in depend:
+                        depend_func = depend.split("::")[-1]
+                    else:
+                        depend_func = depend
                     depend_parent = item.parent
                     depend_nodeid = '{}::{}'.format(depend_parent.nodeid, depend_func)
                     # assert depend_nodeid == depend_nodeid2

Edit: that uses the current item's parent to create the dependency item, which will only work if they are defined in the same class, e.g. this is not covered:

@pytest.mark.dependency()
class Tests:
    def test_b(self):
        pass

class TestOtherTests:
    @pytest.mark.dependency(depends=["Tests::test_b"])
    def test_d(self):
        pass

Another limitation is the item does not seem to include fixture information, and when pytest wants to run that item we added, we get a required positional argument TypeError.

pytest_collection_modifyitems is indeed documented as "May filter or re-order the items in-place", the lack of "add" here is likely a hint that this is not the right approach.