Author vstinner
Recipients inada.naoki, python-dev, serhiy.storchaka, vstinner
Date 2017-01-19.21:46:34
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <CAMpsgwbcz21CGj7z5sbgpWtJFwos10ctjGZdMmxz3iz5kW+o_Q@mail.gmail.com>
In-reply-to <1484827998.47.0.746428626193.issue29296@psf.upfronthosting.co.za>
Content
Serhiy Storchaka:
> When the fastcall protocol be changed, this would require changing more handwritten code.

I don't plan to modify the fastcall protocol. I'm not sure that it's
really possible to modify it anymore. It would probably be simpler to
add a new protocol. But it may become a little bit annoying to have to
support so many calling convention :-) So someone should come up with
a serious rationale to add another one :-)

I tried to make fastcall as private as possible, but recently I
noticed that METH_FASTCALL is now public in Python 3.6 API (but
hopefully excluded from the stable ABI). I know that it's already used
by Cython since 0.25:
https://mail.python.org/pipermail/cython-devel/2016-October/004959.html

About the maintenance burden, I'm ok to handle it :-) My final goal is
to use Argument Clinic everywhere. While converting existing code to
AC is a slow process, I consider (from my recent experiences with AC)
that the new code is easier to maintain! It's a slow process beause AC
still has limitaitons, but also because it's used an opportunity to
review (and enhance!) the existing docstrings.
History
Date User Action Args
2017-01-19 21:46:34vstinnersetrecipients: + vstinner, inada.naoki, python-dev, serhiy.storchaka
2017-01-19 21:46:34vstinnerlinkissue29296 messages
2017-01-19 21:46:34vstinnercreate