Certain nlmixr fits crash the R session in current R-devel under Windows

[jranke]

When I run the first example in the nlmixr help page:

  f_ode <- function(){
         ini({
             lCl <- 1.6      #log Cl (L/hr)
             lVc <- log(80)   #log Vc (L)
             lKA <- 0.3      #log Ka (1/hr)
             prop.err <- c(0, 0.2, 1)
             eta.Cl ~ 0.3 ## BSV Cl
             eta.Vc ~ 0.2 ## BSV Vc
             eta.KA ~ 0.1 ## BSV Ka
         })
         model({
             ## First parameters are defined in terms of the initial estimates
             ## parameter names.
             Cl <- exp(lCl + eta.Cl)
             Vc = exp(lVc + eta.Vc)
             KA <- exp(lKA + eta.KA)
             ## After the differential equations are defined
             kel <- Cl / Vc;
             d/dt(depot)    = -KA*depot;
             d/dt(centr)  =  KA*depot-kel*centr;
             ## And the concentration is then calculated
             cp = centr / Vc;
             ## Last, nlmixr is told that the plasma concentration follows
             ## a proportional error (estimated by the parameter prop.err)
             cp ~ prop(prop.err)
         })
     }
     
     # Use nlme algorithm
     fit_linCmt_nlme <- try(nlmixr(f_ode, Oral_1CPT, est="nlme",
                    control=nlmeControl(maxstepsOde = 50000, pnlsTol=0.4)))

My R session consistently crashes when using R-devel (r81847 ucrt) under Windows. The crash occurs after a couple of minutes on my system. When I turn off buffered output, I can see that the crash occurs after the message "calculating Jacobian".

I came across this problem trying to find out why my test setup script crashes Winbuilder for R-devel and rhub/r-devel-windows.

> sessionInfo()
R Under development (unstable) (2022-03-03 r81847 ucrt)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19043)

Matrix products: default

Random number generation:
 RNG:     Mersenne-Twister 
 Normal:  Inversion 
 Sample:  Rounding 
 
locale:
[1] LC_COLLATE=English_United Kingdom.utf8 
[2] LC_CTYPE=English_United Kingdom.utf8   
[3] LC_MONETARY=English_United Kingdom.utf8
[4] LC_NUMERIC=C                           
[5] LC_TIME=English_United Kingdom.utf8    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] nlmixr_2.0.6

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.8          pillar_1.7.0        compiler_4.2.0     
 [4] PreciseSums_0.4     tools_4.2.0         sys_3.4            
 [7] qs_0.25.3           lattice_0.20-45     nlme_3.1-155       
[10] memoise_2.0.1       lifecycle_1.0.1     tibble_3.1.6       
[13] gtable_0.3.0        checkmate_2.0.0     pkgconfig_2.0.3    
[16] rlang_1.0.1         cli_3.2.0           parallel_4.2.0     
[19] fastmap_1.1.0       n1qn1_6.0.1-10      dplyr_1.0.8        
[22] generics_0.1.2      vctrs_0.3.8         grid_4.2.0         
[25] tidyselect_1.1.2    glue_1.6.2          RApiSerialize_0.1.0
[28] R6_2.5.1            fansi_1.0.2         ggplot2_3.3.5      
[31] purrr_0.3.4         magrittr_2.0.2      lbfgsb3c_2020-3.2  
[34] backports_1.4.1     scales_1.1.1        ellipsis_0.3.2     
[37] RxODE_1.1.4         colorspace_2.0-3    brew_1.0-7         
[40] utf8_1.2.2          stringfish_0.15.5   lotri_0.3.1        
[43] RcppParallel_5.1.5  munsell_0.5.0       cachem_1.0.6       
[46] crayon_1.5.0       

[mattfidler]

Thank you for reporting this. We found a few memory issues that we have fixed and will be pushing in the new rxode2/nlmixr2. Unfortunately, windows is the most difficult platform to track down an issue like this.

[jranke]

OK, thanks for the feedback! nlmixr2 looks promising, so I will have a look. From the github README it really seems to be ready to use, can you give me a hint on the status?

[jranke]

On second thought, this really sounds like I should abandon my nlmixr backend for mkin, as I do not see how I could be able to push it to CRAN... Would you recommend that I switch to nlmixr2 already and wait with a CRAN release of the nlmixr(2) backend for nlmixr2 to be released to CRAN?

[mattfidler]

I am hoping to release soon, as always it is up to you. I need to move vignettes around and put some assertions in as well as track down any bugs.

From a function point of view the code should be the same between nlmixr and nlmixr2 in most cases. The controls will be slightly different (ie foceiControl(), saemControl(), and nlmixControl()/nlmixrNlmeControl() have been aligned and now check to make sure that the right types of values are supplied.

[mattfidler]

Also, it seems that nlme performs a little worse for the theo_sd dataset. For me, I am unsure if it is because of a rewritten nlme engine (I could have changed residual weights for normal/additive error)

[mattfidler]

Since you are not calling the foceiControl and saemControl but using nlmixr_model it should work in both versions.

[jranke]

OK, thanks a lot for your feedback!