Planning the 2.32 release

Ahh yeah that all makes sense, thanks for clarifying!

Iā€™ve put together a summary of the downstream rstan failures that will occur once the deprecated syntax has expired here: RStan & StanHeaders Downstream Patching Status Ā· Issue #1053 Ā· stan-dev/rstan Ā· GitHub

2 Likes

@WardBrian @rok_cesnovar Looking at the amount of package failures that are going to be caused by the deprecation (109!), I think thereā€™s merit in delaying the deprecation decision until RStan 2.26 is on CRAN. Given the popularity of RStan, weā€™re at risk of fragmenting the userbase if we remove the syntax before RStan is compatible (or even if the removal happens closely after RStan becomes compatible).

Given that the current CRAN rstan isnā€™t compatible, it means that we canā€™t start updating package syntax in advance either. This would likely cause rstan to fall behind the main versions again even once 2.26 and then 2.31 is released, since weā€™d be waiting on package maintainers to update (or merge submitted patches) for the deprecations.

By waiting until 2.26 is on CRAN and giving us some time to submit patches for maintainers, we can start reducing the version gap between rstan and the other interfaces.

I donā€™t think this would be too unreasonable, or unduly delay things, given that the downstream failures for 2.26 are down to 17 packages (counting rstanarm) which all have patches merged or waiting to be merged. But it could be worth contacting Ben to get a more accurate status on the submission

2 Likes

I think delaying for one version is reasonable. We anticipated this being an issue when the deprecation policy was first being discussed, but I think what we didnā€™t anticipate was the possibility we would actually be so close to an updated RStan on CRAN (thanks again for the monumental effort on your part, @andrjohns, I know it is a large part of that). I get nervous about ā€œuntilā€¦ CRANā€ since CRAN is an outside entity we canā€™t control, and at some point it wouldnā€™t make sense to wait. Do we see any reason on our side 2.26/2.31 would take longer than the next 4 months (one release cycle)? @bgoodri

Iā€™m happy to add a delay both to make RStanā€™s life easier and because @mitzimorris pointed out to me recently itā€™s a bit of a bummer to have a release where the primary changes are removals. Since next version should have some more substantial features (e.g. almost certainly Pathfinder, tuples) it seems much less likely people would balk at updating.

If we can agree that one version is enough for us to have made our best effort to get RStan on CRAN I can update the warnings and doc for the delay.

4 Likes

Great! I think one version (four months) should be enough, but I also donā€™t know what CRANā€™s requirements are right now for the submission so would probably be best if @bgoodri could weigh in

2 Likes

Docs and Stanc have both been updated to reflect the delay until 2.33

5 Likes

I assume at this point the earliest the RC would be posted is Monday? @rok_cesnovar @serban-nicusor

Yeah, sorry. I was very busy this week. If Nic has time to do it Monday, that is fine with me.

1 Like

No worries! Just wanted to keep myself up to date

Release candidates are now out!

Stanc3: Release v2.32.0-rc1 (10 April 2023) Ā· stan-dev/stanc3 Ā· GitHub
Cmdstan: Release v2.32.0-rc1 (10 April 2023) Ā· stan-dev/cmdstan Ā· GitHub
Stan: Release v2.32.0-rc1 (10 April 2023) Ā· stan-dev/stan Ā· GitHub
Math: Release v4.6.0-rc1 (10 April 2023) Ā· stan-dev/math Ā· GitHub

4 Likes

Hello! Where is the changelog for 2.32 release?

Is there something specific in need of testing? (I might have time to test via CmdStanR)

1 Like

Changelogs are being drafted this week

1 Like

Hey @avehtari, usually we run one or two tests ( ex. cmdstanr for cmdstan bins Custom CmdStan tarball unit tests Ā· stan-dev/cmdstanr@ab30e07 Ā· GitHub ) other than that I assume the new features ( after we get a changelog ).

Thanks for the release candidates. httpstan builds successfully.

It seems ChangeLog was not published during the release candidate phase? At least not in this thread, and not it seems the 2.32 has been released. I would love to help to test new features in release candidate, e.g. Laplace which now seems to be in the release, but without the release notes I didnā€™t know what is new. Oh well, Iā€™ll test the release next week.

1 Like