Author nascheme
Recipients jdemeyer, lukasz.langa, nascheme, ncoghlan, pablogsal, scoder, serhiy.storchaka, vstinner
Date 2019-06-11.22:58:41
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1560293922.19.0.463517287302.issue37221@roundup.psfhosted.org>
In-reply-to
Content
I suggest we change PyCode_New() back in the next beta, despite that change breaking Cython again.  We could work with them so that they have a new release with a "PY_VERSION_HEX > 0x030800b1" test in it. Try to get that Cython version released before 3.8b2 hits.  The window of a Cython version that doesn't support CPython >= 3.8b1 can be made pretty small.  It would be nice to find these issues in alpha releases but finding them in beta is still early enough to fix it.

Having to regenerate all of the Cython emitted code embedded in different packages is painful.  I experienced having to update numpy so that I could test my software with 3.8b1.  Avoiding that is why we should change PyCode_New back.  Introducing a new API like PyCode_NewWithPosArgs() is going to work better.

If we do have to make a similar incompatible change in the future, we should try hard to make it so there is an overlapping period of compatibility.  It's bad to have an API change from one release to the next without giving time for projects like Cython time to catch up.
History
Date User Action Args
2019-06-11 22:58:42naschemesetrecipients: + nascheme, ncoghlan, scoder, vstinner, lukasz.langa, serhiy.storchaka, jdemeyer, pablogsal
2019-06-11 22:58:42naschemesetmessageid: <1560293922.19.0.463517287302.issue37221@roundup.psfhosted.org>
2019-06-11 22:58:42naschemelinkissue37221 messages
2019-06-11 22:58:41naschemecreate