05:27 <mattip> cfbolz: I guess it would be sufficient to attach extra info to subclasses of builtin types

05:28 <mattip> I see how cpyext does it in pyobject.py with add_direct_pyobj_storage(), is that what you would suggest?

06:07 <cfbolz> mattip: I'm getting some coffee, then I'll take a look

06:18 <mattip> one thing I don't understand in the use of add_direct_pyobj_storage is that it is not applied to W_AbstractIntObject

06:18 <cfbolz> mattip: no, I was thinking of something else. in cpyext the default case, w_root_as_pyobj, will perform a dict lookup

06:19 <cfbolz> mattip: add_direct_pyobj_storage gets rid of the dict lookup for some classes

06:19 <cfbolz> but we only want to do that for classes where the extra word of space is not a big memory increase

06:19 <cfbolz> making every int in pypy larger for cpyext feels like the wrong tradeoff

06:20 <cfbolz> but anyway, in hpy the problem is different: we need that extra memory only for hpy-defined subclasses of types, right?

06:21 <mattip> yes

06:22 <cfbolz> ok

06:22 <mattip> and for instances

06:22 <cfbolz> right

06:22 <cfbolz> instances = subclasses of object, right?

06:24 <cfbolz> mattip: i'll try to sketch a bit of code and see how it goes, ok?

06:24 <mattip> gr8. Then maybe we can meet up later

06:45 <arigato> cffi warning: there is a pull request on github for cffi that adds a lot of CPython-specific stuff to c/test_c.py. PyPy should probably drop the test in pypy/module/_cffi_backend/test/test_file.py and agree that it doesn't need to follow cffi that closely any more (it's only to accomodate new CPython versions, new platforms, etc. as opposed to new features)

06:51 <arigato> (it's what occurs right now anyway, because this "test_same_file" compares with the hg version of cffi anyway, which is now frozen)

12:34 <cfbolz> mattip: I did a commit to maybe fix translation, hope i didn't step on your feet with that

21:13 <mattip> arigato: thanks for the heads up