Release 2.28 checklist

[rok-cesnovar]

Here is the checklist for the next release.

Feature freeze (September 20th):

• Create a release candidate tarball. Make sure RC has the new CMDSTAN_VERSION and that RC tarballs have binaries.
• Create Math/Stan RC releases. Make sure versions are updated.
• Create Stanc3 RC binary. Make sure versions are updated.
• Check external links in docs (i.e. TBB)
• Create a release candidate feature/bugfix list (this doesn't need to be polished, just so that RC testers know what to test)
• Run CmdstanR tests with the RC tarball
• Make a Discourse RC post
• Make a Twitter RC announcement

Release (September 27th):

• Create the Math Release notes
• Create the Stan Release notes
• Create the Cmdstan Release notes
• Create the Stanc3 Release notes
• Rebuild and publish docs for the new version
• Create the Math release
• Create the Stan release
• Create the Stanc3 release
• Create the CmdStan x86 tarball (make sure the version is incremented and tarball includes binaries and the folder inside is cmdstan-version (cmdstan-2.28.0)
• Create the CmdStan non-x86 tarballs
• Run CmdStanR tests with the release tarball
• Make a Discourse release announcement post
• Cross-post the Discourse announcement on the Stan blog (dont forget thanking our sponsors)
• Make the Twitter announcement
• Update Math/Stan/CmdStan versions on develop

[rok-cesnovar]

Tagging @bob-carpenter @SteveBronder @WardBrian @nhuurre @spinkney that have been doing stuff across the Math/Stan/Cmdstan/Stanc3 repositories in the last few weeks. Does any one of you have anything they want to get in for the next release and need more eyes on it or some help for a two week push? The deadlines are generally hard but we can be flexible if there is months-long project that might miss for a day or two.

I currently do not have any open active issues/project that I would like to finish for this release.

I will mostly handle the tasks listed above with @serban-nicusor-toptal and will tag devs to double check when the time comes.

[WardBrian]

I have two small stanc3 PRs I would like included.

• The first is Alternative: Add imaginary literal syntax stanc3#963, which adds the ability to write complex numbers as 3 + 2i
• The second is Properly escape string literals with non-ascii characters stanc3#952, which just changes how we escape string literals in our generated code.

Both have docs that have been merged and are passing all tests, so all they need is review. They're pretty small (3 files/~50loc each)

Nothing else from me!

[spinkney]

• generalized inverse fix for symmetric matrices is done. Needs review check generalized inverse for full rank symmetric mat math#2577
• the multi normal derivatives just calls multi_normal_cholesky under the hood it is done except that the expression tests are failing due to different expressions for multi_normal vs multi_normal_cholesky. I need someone to help with that. Multi normal derivatives math#2572

[wds15]

@SteveBronder … varmat???

[SteveBronder]

@wds15 I think the plan is to release it immediately after 2.28 so we have a full dev cycle for testing etc.

[rok-cesnovar]

I have a few minor cleanup PRs ready if anyone has time to review:
#1039
#1040

[serban-nicusor-toptal]

Hey @rok-cesnovar should I wait until tomorrow for #1040 to be merged in ?

[rok-cesnovar]

Will fix that first thing in the morning and then we can proceed yeah. The deadline is Tuesday morning CET time anyways.

[rok-cesnovar]

@serban-nicusor-toptal we need to postpone until we address stan-dev/stan#3057 so you are free today :)

[SteveBronder]

@rok-cesnovar should we also wait till we support Eigen 3.4? I'm working on that right now and there's a weird bug with mdivide_left/right I'm trying to sort out. Seems like it would be good to make sure we are compatible so that when R etc. bumps up everything works smoothly

[SteveBronder]

Oh I'd also like to update the OpenCL hpp headers so they work with OpenCL 3.0 devices. Opened up a PR for that below

stan-dev/math#2587

[wds15]

R is not really an urgent consideration given how much rstan is behind is what I would say. So we do not need eigen 3.4 soon.

[rok-cesnovar]

should we also wait till we support Eigen 3.4?

Just doing it for R is not needed. As Sebastian notes, there is almost zero chance rstan will get to 2.28 until the next release. If there are other bigger reasons or gains we get with that for 2.28 we can do it.

Oh I'd also like to update the OpenCL hpp headers so they work with OpenCL 3.0 devices. Opened up a PR for that below

Will review, seems tests are failing with it.

[rok-cesnovar]

The main things I will highlight in the release candidate as things we wish "power" users would test and give feedback on and are user-facing:

• multi chain cmdstan with threads in a single executable
• complex numbers
• updated Power/Newton algebra solvers & faster inv_Phi
• automatic handling of precompiled headers on Windows with RTools 4.0
• improved error messaging/type error explanations in stanc3
• improve pretty printing/auto formatting
• added skew_double_exponential

There really isnt a ton of things this time around, at least not stuff that is user-facing atm.

@bob-carpenter @WardBrian do you have any sort of example, model + data for the complex stuff? No need to explain the model or have any accompanying text. Just something "power" users that typically help us test the release candidate can run and check that it works for them.

For the rest I think I have enough knowledge of what was done so I can write a sentence on it.

[rok-cesnovar]

The fix for stan-dev/stan#3057 is in so once that propagates upstream to cmdstan (10 hours or so) we are good to go in terms of the release candidate. But we dont want to make a RC on the weekend but we can do it first thing on Monday.