New v2.17.1 release for Math and Stan?

Developers
81

If we then change master from v2.17.1 to point to the old master + develop merged in (without those extra commits), it will break everyone’s local copy of master when they try to git pull as far as I know. If we agree we want to avoid that, the question comes back to if we want to merge in these commits that do the same work onto master or not. Since there is some risk there, I’m not sure why we’re not just releasing 2.18 given that we don’t know of anyone still on gcc 4.6 (but we should totally take a user survey at some point to figure out what people actually want to use).

82

What about simply tagging some commit as v2.17.1 on Github for those who
need it (the gcc 4.6 people). They can install from source.

Everyone else gets 2.18.

83

That’s what I’d propose, yeah :) Interfaces could optionally do a release against that tag if they wanted to.

84

If releasing develop as 2.18 and tagging an orphan branch as 2.17.1 will get things moving, fine.

What I don’t like about the process is that it is being done out of what I believe is an unfounded fear of there being future problems merging develop into master, despite Git being designed to handle situations exactly like this and those that are much more complicated.

2 Likes
85

@seantalts, that sound ok to you? If so, estimate of time? This is just for math and stan.

86

That works for me. I can try to do both today and/or tomorrow.

@bgoodri I agree it’s more of a fear based on ignorance, but of something that is widely warned against in what I’ve found online. If you find someone writing to the contrary you should post it. All I was saying is that we should not use the automatic conflict-resolution feature if we’re going to be cherry-picking, and that it seems like there is some fear but also more work compared with the orphan branch approach for little to no benefit.

87

Thanks!

Could you hold off on releasing CmdStan? We should fix the user build issues before it goes out.

88

Sure! Check out my pull requests re: user build confusions. I think all the
issues are fixed with that but not sure I synthesized all the complaints
correctly.

1 Like
89

Thanks! I’ll definitely take a look.

90

@seantalts, I don’t see an open pull request. Where should I be looking?

91

CmdStan repo

92

I see the doc PR #589, but there are actual build issues that came up from when we changed the makefiles around.

93

I commented on that issue about that - my impression was that there was only documentation to fix and maybe some requested flexibility on the C++ standard library used but nothing actually broken? Anyway let’s talk about it there and save this thread for the releases.

Just to double check, do we actually want to release 2.18 now? This is the first I’m hearing of that. Not opposed, just want to make sure there’s some consensus about what should go into it, that kind of thing.

94

I just realized this whole time we’ve been talking about the math repo, which AFAIK the interfaces don’t even care about? Stan repo’s master has been updated to point to the v2.17.1 tag on Math. So v.2.17.1 is released for Stan and Math.

95

They do indirectly through Stan.

Thanks!

No need to. I think I misread an earlier comment of yours to mean that we’d do both.

Thanks!

96

Right, the interfaces care about the math library version that’s included… but only indirectly through Stan. In other words, it doesn’t matter to literally anyone what is going on in the Math repo’s master branch, if I understand the situation correctly (low confidence in that, though).

1 Like
97

I’m sure we have users of the Math library that aren’t visible. They would care what the latest hash of master is, but definitely not really care how it goes from one hash to another.

1 Like
98

Got it. So the people who might care about this particular discussion we’ve been having have to be using the Math repo and not telling us about it, working off of master and for some reason not happy with switching to a tag, needing a version that is > 2.16 but stuck on a compiler before GCC 4.9.3… seems like it’s likely not a huge number of affected users for what was a pretty lengthy discussion. I think it’d be great to start getting some user data so we can start prioritizing our time better. Maybe @breckbaldwin can help facilitate something along those lines?

2 Likes
99

@syclik Ready for 2.17.1 on CmdStan now?

1 Like
100

Yup. I just checked that it compiles and runs on my local Mac, a minimal linux installation (dockerized), and Windows.

1 Like