Hi Stefan,

How does this approach in reinit() look? We first ask for a writable buffer, and if the object obliges, immediately copy it. Otherwise if it refused, ask for a read-only buffer, and this time expect that it will never change.

This still does not catch the case of mmap.mmap. I am not sure how do deal with mmap.mmap. There is no way for it to export PROT_READ as a read-only buffer without permitted mutation, so the only options seem to either be a) remove buffer support from mmap, or b) blacklist it in bytesio(!).


Antoine, I have padded out the unit tests a little. test_memoryio.py seems the best place for them. Also modified test_sizeof(), although to the way this test is designed seems inherently brittle to begin with. Now it is also sensitive to changes in Py_buffer struct.


Various other changes:

* __new__ once again returns a valid, open, empty BytesIO, since the alternative breaks pickling.

* reinit() preserves existing BytesIO state until it knows it can succeed, which fixes another of the pickle tests.

* setstate() had CHECK_CLOSED() re-added, again for the pickle tests.


Probably the patch guts could be rearranged again, since the definition of the functions is no longer as clear as it was in cow3.patch.