Dag Sverre Seljebotn wrote: > Robert wrote: >> Is there any reason to even allow this (other than to be backwards >> compatible with broken code). I think we should allow object <--> >> void*, or object <--> PyObject*, but no others. (One would then be >> forced to do <double*><void*>arr.data if one *really* wanted to.) > > Excellent idea!; and as you say should not break code. I'll do this for > 0.11.2.
Done, however there's a slight issue: There's no canonical PyObject. If somebody has defined it as an empty struct, say, they now have problems (but a very helpful error is raised when compiling so it's OK). The ones we ship in Includes/ use "ctypedef void PyObject", and they still work (as it ends up as void*). I also removed the warning that used to be emitted for the specific cases of <void*>pyobj and <object>voidp. -- Dag Sverre _______________________________________________ Cython-dev mailing list [email protected] http://codespeak.net/mailman/listinfo/cython-dev
