Cmdstanr::cmdstan_model can compile model, but fails to make it available in R, unless already compiled

Interfaces CmdStan
1

When I call cmdstan_model, it can compile and save the model, but it gives the error Error in file(con, "r") : cannot open the connection when trying to actually make it available in R. However, if the model is already compiled, then it correctly returns a CmdStanModel.

Example

The following code fails with the error Error in file(con, "r") : cannot open the connection:

library(cmdstanr)
file <- file.path(cmdstan_path(), "examples", "bernoulli", "bernoulli.stan")
mod <- cmdstan_model(file)

When running with options("cmdstanr_verbose" = TRUE) the final part (about linking the model) is:

--- Linking model ---
g++ -std=c++1y -pthread -D_REENTRANT -Wno-sign-compare -Wno-ignored-attributes      -I stan/lib/stan_math/lib/tbb_2020.3/include    -O3 -I src -I stan/src -I stan/lib/rapidjson_1.1.0/ -I lib/CLI11-1.9.1/ -I stan/lib/stan_math/ -I stan/lib/stan_math/lib/eigen_3.4.0 -I stan/lib/stan_math/lib/boost_1.78.0 -I stan/lib/stan_math/lib/sundials_6.1.1/include -I stan/lib/stan_math/lib/sundials_6.1.1/src/sundials    -DBOOST_DISABLE_ASSERTS               -Wl,-L,"/home/adam/.local/opt/cmdstan/stan/lib/stan_math/lib/tbb" -Wl,-rpath,"/home/adam/.local/opt/cmdstan/stan/lib/stan_math/lib/tbb"        /tmp/RtmpZPELHp/model-463e2ffb1b99.o src/cmdstan/main.o       -Wl,-L,"/home/adam/.local/opt/cmdstan/stan/lib/stan_math/lib/tbb" -Wl,-rpath,"/home/adam/.local/opt/cmdstan/stan/lib/stan_math/lib/tbb"     stan/lib/stan_math/lib/sundials_6.1.1/lib/libsundials_nvecserial.a stan/lib/stan_math/lib/sundials_6.1.1/lib/libsundials_cvodes.a stan/lib/stan_math/lib/sundials_6.1.1/lib/libsundials_idas.a stan/lib/stan_math/lib/sundials_6.1.1/lib/libsundials_kinsol.a  stan/lib/stan_math/lib/tbb/libtbb.so.2 -o /tmp/RtmpZPELHp/model-463e2ffb1b99
rm /tmp/RtmpZPELHp/model-463e2ffb1b99.hpp /tmp/RtmpZPELHp/model-463e2ffb1b99.o
Error in file(con, "r") : cannot open the connection

However, if I rerun the code (so the compiled model is already saved), it works.

Setup

Operating system: Ubuntu.

> library(cmdstanr)
This is cmdstanr version 0.6.1.9000
- CmdStanR documentation and vignettes: mc-stan.org/cmdstanr
- CmdStan path: /home/adam/.local/opt/cmdstan
- CmdStan version: 2.33.1
> check_cmdstan_toolchain()
The C++ toolchain required for CmdStan is setup properly!
> cmdstan_path()
[1] "/home/adam/.local/opt/cmdstan"
> cmdstan_version()
[1] "2.33.1"
> tempdir()
[1] "/tmp/Rtmp6MIhLF"

The problem appeared after I updated cmdstan to version 2.33.1 (from version 2.32, I think) and cmdstanr to the newest version from GitHub (I am not sure what version I had before).

2

This could be a tricky one to debug. Our CI runs this exact combination of OS/cmdstan/cmdstanr without error, so it looks like there is some local config causing issues.

If you run traceback() after getting that error, does it report anything?

If you reinstall cmdstan (i.e., reinstall_cmdstan(cores=...)) does the error persist?

1 Like
3

I just tried calling rebuild_cmdstan() and the problem persists.

traceback() doesn’t seem to give anything very interesting (see below). I am considering whether my best bet is to wipe my entire R installation (incl. all packages) and hope that the problem somehow fixes itself by doing a complete reinstall/reconfig.

Result from traceback() after error:

8: file(con, "r")
7: readLines(private$hpp_file_, warn = FALSE)
6: withCallingHandlers(expr, warning = function(w) if (inherits(w, 
       classes)) tryInvokeRestart("muffleWarning"))
5: suppressWarnings(private$model_methods_env_$hpp_code_ <- readLines(private$hpp_file_, 
       warn = FALSE))
4: self$compile(...)
3: initialize(...)
2: CmdStanModel$new(stan_file = stan_file, exe_file = exe_file, 
       compile = compile, ...)
1: cmdstan_model(file)
4

That output is actually quite helpful. It’s saying that the .hpp generated by cmdstan from your Stan code is no longer in the temporary directory. Your system might be very aggressively cleaning out the tempdir for some reason.

Can you try calling cmdstan_model() with the output_dirargument set to some local directory that you can see? We need to verify whether the error occurs while the.hpp file is still present

2 Likes
5

When running mod <- cmdstan_model(file, dir = "/home/adam/cmdstanr_test") I still get the same error and the traceback

8: file(con, "r")
7: readLines(private$hpp_file_, warn = FALSE)
6: withCallingHandlers(expr, warning = function(w) if (inherits(w, 
       classes)) tryInvokeRestart("muffleWarning"))
5: suppressWarnings(private$model_methods_env_$hpp_code_ <- readLines(private$hpp_file_, 
       warn = FALSE))
4: self$compile(...)
3: initialize(...)
2: CmdStanModel$new(stan_file = stan_file, exe_file = exe_file, 
       compile = compile, ...)
1: cmdstan_model(file, dir = "/home/adam/cmdstanr_test")

As before, the executable bernoulli does appear in the /home/adam/cmdstanr_test dir and if I rerun mod <- cmdstan_model(file, dir = "/home/adam/cmdstanr_test") it correctly loads it with the message Model executable is up to date!.

I just tried running the cmdstan_model command while keeping an eye on the dir given by tempdir() (which happens to be /tmp/RtmpncngT8).

  • In the beginning it contains releases-213c3727b5b0.json.
  • While running cmdstan_model compilation it contains model-213c4019897c.stan and model-213c4019897c.hpp.
  • After finishing the cmdstan_model (with throwing the error) the .hpp file disappears and the executable model-213c4019897c appears.
6

Ah hold on, I think I see where this is coming from. The verbose output you posted above includes a step which is deleting the .hpp file:

rm /tmp/RtmpZPELHp/model-463e2ffb1b99.hpp /tmp/RtmpZPELHp/model-463e2ffb1b99.o

This is occuring as part of the initial model compilation, so the hpp is deleted before R has a chance to get to it.

This isn’t a default step in the make command, and doesn’t occur for me (@WardBrian do you know if this has changed recently or similar?).

Do you have any custom CmdStan flags/similar set? What’s your output from cmdstan_make_local()?

7

But this does appear to be a bit fragile, so I’ll update the cmdstanr process to generate the .hpp file again if it’s not present

8

Yes, recently we un-marked the .hpp as .PRECIOUS so the makefile won’t keep it around unless you ask for it

9

Ah got it, makes sense! I’ll update the cmdstanr behaviour to address this, thanks!

10

@adamgorm Just to double-check, are you using the GitHub/develop version of cmdstan? It looks like this change isn’t present in the 2.33.1 release, only in the current develop branch

11

Yes, I installed the newest cmdstanr yesterday by running remotes::install_github("stan-dev/cmdstanr").

Edit: Ah, you asked about cmdstan. There I use the 2.33.1 release.

New edit: I am actually also using the development version of cmdstan.

12

I mean cmdstan itself (not the R package)

13

No, wait, I am actually using the development version of cmdstan as well, I think. I reinstalled yesterday by running > git clone https://github.com/stan-dev/cmdstan.git --recursive as described in the manual.

14

Yep, that’ll do it. I’ll update cmdstanr to handle the new changes, but if you want things to work in the interim then you can revert to the release version by deleting that folder calling install_cmdstan() again

1 Like
15

Yes, you’re right! I just deleted my cmdstan installation and reinstalled with install_cmdstan() and now it works flawlessly.

Thanks a lot for the help and for the very fast replies!

16

No worries, thanks for flagging this! It’s very helpful to know about it before the next release

1 Like
17

Hi all, just a note that I’m now getting this same error on the latest version of Cmdstan. My setup is:

> version
               _                                
platform       x86_64-w64-mingw32               
arch           x86_64                           
os             mingw32                          
crt            ucrt                             
system         x86_64, mingw32                  
status                                          
major          4                                
minor          3.1                              
year           2023                             
month          06                               
day            16                               
svn rev        84548                            
language       R                                
version.string R version 4.3.1 (2023-06-16 ucrt)
nickname       Beagle Scouts  

> cmdstan_version()
[1] "2.34.0"

> utils::packageVersion('cmdstanr')
[1] ‘0.7.0’

The traceback looks like this:

Error in file(con, "r") : cannot open the connection
9. file(con, "r")
8. readLines(private$hpp_file_, warn = FALSE)
7. withCallingHandlers(expr, warning = function(w) if (inherits(w, classes)) tryInvokeRestart("muffleWarning"))
6. suppressWarnings(private$model_methods_env_$hpp_code_ <- readLines(private$hpp_file_,
warn = FALSE))
5. self$compile(...)
4. initialize(...)
3. CmdStanModel$new(stan_file = stan_file, exe_file = exe_file,
compile = compile, ...)
2. cmdstanr::cmdstan_model(cmdstanr::write_stan_file(vectorised$model_file)) at mvgam.R#1908

My make file is:

> cmdstan_make_local()
[1] "CXXFLAGS += -Wno-nonnull -D_UCRT"        
[2] "TBB_CXXFLAGS= -D_UCRT"                   
[3] "CXXFLAGS += -Wno-deprecated-declarations"
[4] "CXXFLAGS += -Wno-nonnull -D_UCRT"        
[5] "TBB_CXXFLAGS= -D_UCRT"                   
[6] "CXXFLAGS += -Wno-deprecated-declarations"

Any help would be greatly appreciated

EDIT: I’ve just upgraded to the latest development version of cmdstanr (0.7.1) and everything seems to be working now. I’ll leave this post here though in case anyone else runs into similar troubles