test_metaclass and `set_meta_data`: it seems to be operating on the wrong legacy class?

[Matti Picus]

In test_metaclass there is this code:

hpy/test/test_hpytype.py

Line 1462 in f611473

 assert mod.set_meta_data(mod.Dummy) is None 

It calls into this function:

hpy/test/test_hpytype.py

Lines 244 to 254 in f611473

             HPyDef_METH(set_meta_data, "set_meta_data", HPyFunc_O) 
             static HPy set_meta_data_impl(HPyContext *ctx, HPy self, HPy arg) 
             { 
                 DummyMeta *data = DummyMeta_AsStruct(ctx, arg); 
                 data->meta_magic = 42; 
                 data->meta_member = 11; 
                 for (size_t i = 0; i < 64; ++i) 
                     data->some_more[i] = (char) i; 
                 return HPy_Dup(ctx, ctx->h_None); 
             } 
  

In legacy mode the macro DummyMeta_AsStruct is converted into

__attribute__((unused)) 
static inline DummyMeta * DummyMeta_AsStruct(HPyContext *ctx, HPy h) { 
    return (DummyMeta *) _HPy_AsStruct_Legacy(ctx, h); 
}

where _HPy_AsStruct_Legacy for CPython is _h2py(h). But the h in the test is the Dummy type object, not the DummyMeta type object, so doesn't _h2py(h) return the Dummy data struct (and not the DummyMeta struct the test is expecting? In universal mode, DummyMeta_AsStruct calls _HPy_AsStruct_Type, which I wrote as obj = handles.deref(h); type = type(obj); return dataptr(type) so that it will return the DummyMeta struct, but I don't know what to do for _HPy_AsStruct_Legacy.

[Matti Picus]

I had a bug in the PyPy implementation, and now things work. I am still not sure how.