Deprecating old stderr stream setting code in packages using RcppArmadillo

[eddelbuettel]

Issue #391 works towards no longer suppressing a deprecation warning for use of a very old scheme to instantiate variables. When we no longer suppress these warnings we also see a number of packages exhibting a deprecation warning such as

warning: ‘void arma::set_cerr_stream(std::ostream&)’ is deprecated

which is easy to address via an updated, not deprecated call. This issue regroups the stautus of pull requests (or emailed patches) for the packages involved:

• legion (at https://github.com/config-i1/legion) (PR filed 2023-01-15) (merged 2023-01-16) (version 0.1.2 on CRAN 2023-01-31)
• quadrupen (at https://github.com/jchiquet/quadrupenCRAN) (patch created, tested, and emailed 2023-01-15) (heard back 2023-01-16) (version 0.2-10 on CRAN 2023-01-16) (additional issue with deprecation, commit [already made in January] (jchiquet/quadrupenCRAN@544624d)) (new version on CRAN 2023-12-18)
• Rphylopars (at https://github.com/ericgoolsby/Rphylopars) (PR filed 2023-01-15) (no longer on CRAN as of 2024-01-12)
• rxode2 (at https://github.com/nlmixr2/rxode2) (PR filed 2023-01-15) (merged 2023-01-15) (version 2.0.12 on CRAN 2023-03-28)
• smooth (at https://github.com/config-i1) (PR filed 2023-01-15) (merged 2023-01-16) (version 3.2.0 on CRAN on 2023-01-18)
• stm (at https://github.com/bstewart/stm) (PR filed 2023-01-15) (heard back 2023-01-15) (merged 2023-11-30) (new version 1.3.7 at CRAN 2023-12-01)
• r2sundials (at https://github.com/sgsokol/r2sundials) (issue found in re-run 2023-10-29) (pr #2 made 2023-10-31) (new version on CRAN 2023-12-07)

[eddelbuettel]

Ah well it would have helped if you had documented that earlier / clearer. Feel free to alter the six PRs I wrote yesterdat to avoid the deprecation noise coming from what you once posted as documented (as users apparently found that in your logging section).

I am not planning to unmask the deprecation warning for maybe two or three months so there is time to clean this up. Contributions welcome.

[eddelbuettel]

It's not part of the public API, so it should never be placed in user code.

Conrad, please. It is in the same header file as the functions barking. If you do not highlight what function to use instead of a deprecated one, you cannot be upset that us peons take one you do not like. Document it better to help us!

Please look at the six (out of over 1000 !!) packages using RcppArmadillo having this deprecation issue, study how they use the deprecated cerr stream setter and maybe we can come up with a better way between now and April (my target date for removing the deprecation muffler). At the end of the day all six were already using a kludge to suppress a likely meaningful warning so maybe we should take this time and dig deeper.

Every CRAN package foo can be inspected online via the GitHub mirror at github.com/cran/foo i.e. each package is a mirrored repo in the cran organisation.

(Doing deprecation well is hard !! We have the same issue at work in the (core C++) library tiledb I wrap in the tiledb-r package. Pointing to promoted alternative is as best as I can tell the only viable alternative.)

[bstewart]

@conradsnicta Thanks for weighing in and for Armadillo!

I am happy to take out the internal code and replace with the macro as you recommend.

For @eddelbuettel with respect to your comment:

At the end of the day all six were already using a kludge to suppress a likely meaningful warning so maybe we should take this time and dig deeper.

I just wanted to document here that the warning definitely is meaningful, we knew the situation could arise and test for it and address ourselves—the issue was that the fastest way to test was to attempt the decomposition and have it fail which produced a warning that (10 years ago) we couldn't shut off (but now we can!). Part of the challenge is that I'm more of a statistician than a programmer (and definitely more of an R programmer than a C++ programmer). It is a testament to both your efforts to make RcppArmadillo easy to use that I was able to integrate it into stm at all.

Anyways, I wanted to thank you both so much for fundamental contributions that really made stm possible. I'll get this updated as soon as I can and up on CRAN.

[eddelbuettel]

Oh yes of course -- we also need to take care of error conditions. But sometimes it is nice to do it locally:

> message("Some loud stuff")      # messaging users is good and needed at times
Some loud stuff
> suppressMessages(message("Some loud stuff"))     # peace and quiet as desired
> 

[eddelbuettel]

As of 2023-02-04:

> suppressMessages(library(data.table))
> pkgs <- c("Rphylopars", "rxode2", "stm")
> db[Package %in% pkgs,.(Package,Version,Date,Published)][order(-Published)]
      Package Version       Date  Published
       <char>  <char>     <char>     <char>
1:     rxode2  2.0.11       <NA> 2022-11-01
2: Rphylopars   0.3.9 2022-05-08 2022-05-09
3:        stm   1.3.6       <NA> 2020-09-18
> 

[eddelbuettel]

Unchanged 2023-03-12:

> suppressMessages(library(data.table))       
> db <- data.table(tools::CRAN_package_db())
> pkgs <- c("Rphylopars", "rxode2", "stm")      
> db[Package %in% pkgs,.(Package,Version,Date,Published)][order(-Published)]         
      Package Version       Date  Published              
1:     rxode2  2.0.11       <NA> 2022-11-01             
2: Rphylopars   0.3.9 2022-05-08 2022-05-09      
3:        stm   1.3.6       <NA> 2020-09-18                 
>  

[eddelbuettel]

Still present in Rphylopars and stm as of 2023-09-17.

[eddelbuettel]

RcppArmadillo release 0.12.6.6.1 is now on CRAN, and no longer defines ARMA_IGNORE_DEPRECATED_MARKER.

[eddelbuettel]

The remaining package is now off CRAN meaning the transition is now complete as is #391. Many thanks to all maintainer who updated their packages.