StanHeaders 2.26 - Additional Evals

[andrjohns]

@hsbadr

This PR adds some additional evals to fix compilation errors with the geostan and epidemia packages. The errors were caused by the user-defined functions only accepting Eigen plain types, but there were still some Math functions returning expressions (add for epidemia and rep_vector for geostan)

[hsbadr]

Thanks @andrjohns ! Could you also look at the failure of stanette package after the last patch? See stan-dev/rstan@ec5f4a8

[andrjohns]

Can do, looking now!

[hsbadr]

Perfect! Hopefully, we can fix it too and rerun revdepcheck.

[hsbadr]

Merging this for testing reverse dependencies. Then, we can start from the new report.

[andrjohns]

Figured out the stanette issue, will have another StanHeaders PR through in a couple minutes

[andrjohns]

Also, I've got a PR open for ProbReco to add the linker flags here

[andrjohns]

And it looks like PosteriorBootstrap passes rcmdcheck with both 2.21 & 2.26 and 2.26 & 2.26 across Mac, Win, and Linux: https://github.com/andrjohns/PosteriorBootstrap/actions/runs/2561614694

So I'm not sure how the revdep testing environment is different, and if that might be where the errors are coming from

The Ubuntu (old-rel) failures are unrelated, and something related to the MASS package

[hsbadr]

Let's wait for the new revdep report.

[andrjohns]

Sounds good!

[hsbadr]

Note that tests should be with the different combinations of RStan & StanHeaders x v2.21 & v2.26. We'll release StanHeaders 1st, which needs to be compatible with the CRAN version of RStan v2.21. Then, RStan v2.26 will follow.

[andrjohns]

Yep, I've got the workflow setup to test:

• 2.21 & 2.21
• 2.21 & 2.26
• 2.26 & 2.26

So it should cover all bases

[hsbadr]

Yes, it could be related to the system I use to run revdep checks. It ends up with a segmentation fault:

  Running ‘testthat.R’
 ERROR
Running the tests in ‘tests/testthat.R’ failed.
Last 13 lines of output:
   1. ├─base::suppressWarnings(...) at test_anpl.R:177:2
   2. │ └─base::withCallingHandlers(...)
   3. ├─rstan::vb(...)
   4. └─rstan::vb(...)
   5.   └─rstan .local(object, ...)
   6.     └─sampler$call_sampler(c(args, dotlist))
  
  [ FAIL 1 | WARN 0 | SKIP 3 | PASS 51 ]
  Error: Test failures
  Execution halted
  
   *** caught segfault ***
  address 0x18, cause 'memory not mapped'
  An irrecoverable exception occurred. R is aborting now ...
  Segmentation fault (core dumped)

[hsbadr]

So, we may ignore retest the segfaults on another machine & fix the obvious errors, related to Stan/Math or the backward compatibility patches.

[andrjohns]

Sounds like a plan! Let me know if the revdep checks find any other issues and I'll chase them down

[hsbadr]

Here's the update revdep results: https://github.com/stan-dev/rstan/tree/StanHeaders_2.26/revdep

ProbReco can be fixed by importing either RcppParallel or StanHeaders (in NAMESPACE). It compiles fine but needs to find TBB symbols at loading time. The tests of 4 (IgGeneUsage, PosteriorBootstrap, prophet, and rstanarm) other packages cause segfaults, which could be due to my system environment (system libraries, compiler version, ... etc.). Those packages can be tested on another machine. Also, IgGeneUsage is BioConductor and rstanarm is ours. So, CRAN should accept v2.26.