tp_print renamed to tp_vectorcall_offset in cython master

[tacaswell]

This was changed in:

python/cpython#13185 / python/cpython@aacc77f to support https://www.python.org/dev/peps/pep-0590/

A brute-force tp_print -> tp_vectorcall_offset find-and-replace fixes it on cpython master, but is very not back-compatible 😈 . Will look into seeing if I can sort out how to gate this correctly.

[tacaswell]

https://mail.python.org/pipermail/python-dev/2018-June/153927.html Seems cython is already in the loop on this at the high-level

[scoder]

I would be happy about a PR that

• replaces the direct setting of tp_print with a C macro __Pyx_SET_TP_PRINT()
• renames tp_print in the extension type struct comments with the new name, since the old name is a legacy Py2 name anyway.

[scoder]

That should be against the 0.29.x branch, to keep it compatible with Py3.8. (And since that's a point release, I'd keep the tp_print name in the comments as they are and change them only in master, just in case…)

[scoder]

I'll take care of this change, and will also prepare a new release once it's clear that nothing else will change in CPython 3.8b1.

[tacaswell]

Thanks @scoder ! Sorry I did not get to doing something about this fast enough.

[jdemeyer]

Sorry for that, I didn't expect that Cython would assign anything to tp_print explicitly.

Apart from this, there is no breakage due to PEP 590?

[scoder]

I got everything fixed in my local tests, at least. The latest 0.29.10 release should be ok with CPython 3.8 beta 1. Let's see what our users uncover. :)

[vstinner]

I opened https://bugs.python.org/issue37250 to maybe fix Python 3.8 before Python 3.8.0 final release.

[tacaswell]

@vstinner a few times a week I build cpython + the pydata stack from all of the master branches (and test at least Matplotlib). When I find issues with down-stream projects should I also report them upstream to cpython?