Problems configuring Makevars.win since R 3.6.0

General
1

I had been happily using Rstan on several Windows 10 machines with R 3.5.3. However, after updating each to R 3.6.0, I have been unable to compile the C++ code for Stan. The error messages suggest that there are problems finding the C++14 standard.

To troubleshoot, I uninstalled R and Rtools and then reinstalled R 3.6.0 and Rtools 3.5 in the default locations. I then followed the instructions on https://github.com/stan-dev/rstan/wiki/RStan-Getting-Started and got the following error when I ran the stan() command:

Error in compileCode(f, code, language = language, verbose = verbose) : 
  Compilation ERROR, function(s)/method(s) not created! Error in .shlib_internal(commandArgs(TRUE)) : 
  C++14 standard requested but CXX14 is not defined
Calls: <Anonymous> -> .shlib_internal
Execution halted
In addition: Warning message:
In system(cmd, intern = !verbose) :
  running command 'C:/PROGRA~1/R/R-3.6.0/bin/x64/R CMD SHLIB file31c4b6d5db0.cpp 2> file31c4b6d5db0.cpp.err.txt' had status 1
Error in sink(type = "output") : invalid connection

My Makevars.win file includes the following:

CXX14FLAGS=-O3 -march=native -mtune=native
CXX11FLAGS=-O3 -march=native -mtune=native

I assume that it is looking for a definition for CXX14 but that is not there by default. So even though Rtools is installed, R does not seem to finding g++ correctly.

This is all a bit over my head, so any advice would be really appreciated.

2

I believe this has been solved here: Running rstan with latest R (3.6.0) fails with Error in compileCode

Though of course you may want to try to see if it works without -O2 since that will reduce runtime performance. And may not be a necessary part of the solution. I can’t confirm.

3

Thanks for the quick response. Following the instructions on the page you linked, I updated the Makevars to:

CXX14=$(BINPREF)g++ -O2 -march=native -mtune=native
CXX14FLAGS=-O3 -march=native -mtune=native
CXX11FLAGS=-O3 -march=native -mtune=native

This allowed me to run a model that I know worked previously. It completed in around 580 seconds.
Then I tried again after removing -O2 and it completed again in around 580 seconds.

1 Like
4

Curious.

5

In a (perhaps) related issue, I am now getting the following error when I try to use cores > 1:

Error in unserialize(socklist[[n]]) : error reading from connection
Error in serialize(data, node$con, xdr = FALSE) : error writing to connection

Any ideas on how to troubleshoot this?

6

If it’s in RStudio then I assume you are hitting a bug related to mc.cores and its ilk. Does it work in console R? If so then it’s a known RStudio issue.

7

Here is some discussion: https://stat.ethz.ch/pipermail/r-devel/2019-April/077632.html. I don’t actually know what error it would cause as I’ve never encountered it before.

8

@bgoodri, @jonah is the $(BINPREF) the canonical answer for R 3.6 on Windows moving forwards? Can the RStan quick start for Windows be updated if so? This caused a bunch of issues in my course last week that were eventually solved by prefixing BINPREF by hand.

9

It might be canonical for today. It shouldn’t be necessary going forward now that we use pkgbuild, but I haven’t yet been able to get a rstan 2.19 onto CRAN without breaking a bunch of other packages.

10

Note that it was also helpful for me to move from RTools 3.5 to 4.0 following the instructions on this page: