internal error when running compileNimble

[ralmond]

I've set up a model as follows:

>brownianModel1 <- nimbleCode({

  #... Skipping temporal part of the model

 for (i in 1L:N) {
    for (it in 1L:NT) {
      et1[i,it] <- ilogit(1.7*(a*theta[i,it]-b1))
      et2[i,it] <- ilogit(1.7*(a*theta[i,it]-b2))
      p[i,it,1L] <- 1-et1[i,it]
      p[i,it,2L] <- max(0,et1[i,it]-et2[i,it])
      p[i,it,3L] <- et2[i,it]
      y[i,it] ~dcat(p[i,it,])
    }
  }

 #... Skipping parameter priors
})

The full model (as well as some data generation code can be found in):

hmm-bug.Rmd.txt

The model is an item response theory graded response model. et1 represents the probability of scoring a 2 or higher, and et2 the probability of scoring a 3 or higher, the categorical probabilities are then the differences among those curves. The max is to ensure that I don't accidentally get a negative probability.

I then set up initial values planning to run MCEM

>bminits <- list(
  mu_0 = 0,
  sigma_0=1,
  a=1, b1=-1,b2=0,
  growthSD=rep(.1,N),
  growthRate=rep(.25,N)
)
>brownianModel <- 
  nimbleModel(brownianModelCode1,
              constants=list(N=N,NT=NT),
              dimensions=list(p=c(N,NT,3L)),
              inits=bminits,check=FALSE,
              data=list(y=y),
              buildDerivs = TRUE)
                                        

These steps run with some warnings about unitialized variables.

The problem comes with

cBrownianModel <- compileNimble(brownianModel)

This fails with the error:

Error in if (iName == "log" && (is.numeric(iArg) | is.logical(iArg))) { :
missing value where TRUE/FALSE needed

The error happens in cppOutputNimDerivsPrependType, where iName is equal to NULL and

> NULL=="log" && TRUE
[1] NA

1. There is pretty obviously an unhandled case in cppOutputNimDerivsPrependType when iName is null.

2. It would be really useful if compileNimble could catch low-level exceptions and add context inform (e.g., what line of code or operation it was working on when the error occurred) as that would aid debugging.