Make a v2.1 release?

[dean0x7d]

Seems like there is a decent amount of new features and bug fixes for a v2.1 release. This should also be good timing with the recent release of Visual Studio 2017 and the related fixes in pybind11.

Bugs: Are there any critical bugs that should be fixed? Looking at the issues, there don't seem to be any showstoppers.

Features: Of the feature PRs, #488 and #693 are close to ready. I'd say #693 would be best to merge right at the beginning of v2.2.dev so it gets some time in master due to the low-level core changes, rather than going straight to release. #488 doesn't affect the core library in that way, but perhaps a general feature freeze for v2.1 would be best.

Thoughts?

[jagerman]

Sounds good to me. I think current master is in pretty good shape, and I agree that #693 (or anything of a similar scope) should be in master for a while before a release. For non-core things like #488, I think we can accept them if they are ready to go in the next couple days.

[jagerman]

#488 is merged now; I also created v2.1 and v2.2 milestones to help categorize issues/PRs.

[jagerman]

I think #671 will be an easy fix, as will adding compiler version checks to close #706; I've tagged them both for v2.1, and will submit PRs tomorrow.

[wjakob]

Creating a v2.1 with the current master sounds good to me. I'll wait for @jagerman's two commits before pushing it out (and perhaps add 1-2 days of delay to be on the safe side?).

[wjakob]

https://travis-ci.org/pybind/pybind11/jobs/210795174 :( -- I sense nondeterminism.

[MathMagique]

A 2.1 release would be awesome, since I would love to use the (seemingly new) support for u16string. Thanks for the great work!

[jagerman]

@wjakob that looks like a pypy bug: they just merged a fix for the tp_bases-is-incomplete issue, but when I tried removing the workaround from the MI PR I got a segfault in the same place. I'm going to investigate and report upstream (assuming it's the same issue).

[jagerman]

The segfault issue is fixed in PyPy.

Edit: confirmed.

[dean0x7d]

@jagerman What do you think about bumping #671 to v2.2? It's not a bug fix and it seems like there's still some discussion on performance going on there.

In that case, there would be just #727 left to merge and we can call it a release.

[jagerman]

I think #730 is ready to merge, if you want to give it a quick review. There may (or may not!) be other optimizations we can make eventually for #671, but the dropping the forced copying for anything other than a c-contiguous array in #730 seems worthwhile regardless.

[jagerman]

(The diff in #730 looks bigger than it really is: it's mostly style changes to better help me understand exactly what the code was doing.)

[dean0x7d]

It might be better if @wjakob takes a look, if he has time. You'd definitely get a better review since I'm not familiar with that part of the code at all (on the "how it works" dev side, but also on the "what it does" user side as I haven't had opportunity to use py::vectorize at all).

[jagerman]

Okay. As for #727, are we decided that requiring 2015u3 at a minimum is the way to go? I didn't get any answer to my request for a stronger justification for 2015u2 support.

Edit: a possible option is to relax the compile-time check to 2015u2, but keep our official, documented requirement at 2015u3.

[dean0x7d]

As for #727, are we decided that requiring 2015u3 at a minimum is the way to go?

Yeah, I'd say it's the way to go. The arguments have been made in #727 and I don't think there is a practical way to maintain support for Update 2 and not enough justification to jump through the hoops.

Edit: Added context.