Introduce HPy_DelAttr(_s) and HPy_DelItem(_i/_s)

[fangerer]

This PR add a bunch of new API functions for deleting items and attributes (resolves issue #369).

Similar to CPython, functions HPy_DelAttr(_s) are just inline helper functions that use HPy_SetAttr(_s) and pass HPy_NULL as value.

For HPy_DelItem(_i/_s), I followed the same strategy as we did for HPy_SetItem(_i/_s).

[hodgestar]

The implementation looks good to me. The asymmetry between delitem and delattr is a bit odd, but I think this is just part of Python now. It would be nice to split the setattr and delattr tests so that each test is simpler, but perhaps there is a good reason not to do that?

[hodgestar]

I know the underlying implementation of delattr uses setattr, but could we keep the tests separate just so that each test function is a bit simpler?

[hodgestar]

And the same comment for the other setattr / delattr tests.

[fangerer]

Okay, I can do that. I just merged them into one test because setattr/-item are obviously related to delattr/-item and in this way, we don't generate a second test module which is faster.
But I don't really mind. I'll separate it.

[antocuni]

LGTM.
I think that this PR is mergeable as it, thank you!
I'm writing some random comments just for the sake of the conversation, but they shouldn't prevent you to merge:

• these tests are very comprehensive: sometimes it is hard to distinguish whether we are testing HPy (i.e. "HPy_DelAttr" should delete the attribute) or the underlying python implementation, not HPy-specific (i.e. "the logic for deleting attributes has this and that corner cases" -- you could write the same test in pure python). I think it's a common problem of our test suite, we don't have a standard way to decide which parts of the semantics are HPy-specific and which one are python-specific.

• semi-related to the previous point, the tests have a lot a duplication: e.g. test_setattr and test_setattr_s are basically the same. But we don't have a good way to merge the two, unless we start using complicate fixtures and/or templating magic (but I think we are already doing too much templating magic in our tests). Maybe one solution could be to have a "full" test for test_setattr which covers all the corner cases and a simpler one for test_setattr_s which covers just the simple case? Again, we need to decide whether we are testing HPy or the python implementation

[fangerer]

@antocuni

sometimes it is hard to distinguish whether we are testing HPy (i.e. "HPy_DelAttr" should delete the attribute) or the underlying python implementation

Agreed. One source for this problem is that we still try to map HPy API to C API if we can. We should probably define the semantics separately for HPy but then it might happen that we can no longer directly map to a C API.

the tests have a lot a duplication

We could use a template for those kinds of tests and replace the types. I'm also fine with merging them.

[fangerer]

@hodgestar please do another review round. I've addressed all your comments.

[antocuni]

Agreed. One source for this problem is that we still try to map HPy API to C API if we can. We should probably define the semantics separately for HPy but then it might happen that we can no longer directly map to a C API.

I'm not sure to understand. Even if we say that HPy_DelItem maps to PyObject_DelItem, there is no point in testing all the possible combinations of items, is there?

[hodgestar]

Looks good!