Constant index incorrectly flagged as dynamic index when name begins with `.`

[kenkellner]

On CRAN nimble 1.4.2.

If you provide an index in constants and its name starts with a period, you will get an error or extra warning note (depending on if indexing another constant or a parameter, respectively) which incorrectly flags the index as dynamic when running nimbleModel.

I think I've narrowed down the issue to the creation of constantsNamesList, here:

nimble/packages/nimble/R/BUGS_modelDef.R

Line 302 in 812894c

constantsNamesList <<- lapply(ls(constants), as.name)

By default ls omits object names that start with a period, according to the docs. So constants starting with a . don't show up in this list, and later checks against constantsNamesList incorrectly determine that they are not constants.

You can set the argument all.names = TRUE in ls to include them. You also have to set the same argument in as.list here:

nimble/packages/nimble/R/BUGS_modelDef.R

Line 1413 in 812894c

constantsEnvCopy <- list2env(as.list(constantsEnv))

I'm not sure what kind of side effects this might have. I can make a PR with these changes if needed.

Reproducible example:

library(nimble)

# note index is named 'site'
constants <- list(n = 4, x = rnorm(2), site = c(1,1,2,2))
data <- list(y = rnorm(4))

code <- nimbleCode({
  for (i in 1:n){
    mu[i] <- beta0 + beta1 * x[site[i]]
    y[i] ~ dnorm(mu[i], sd = 1)
  }
})

# works fine
mod <- nimbleModel(code, constants=constants, data=data) 

# now rename index to .site
constants <- list(n = 4, x = rnorm(2), .site = c(1,1,2,2))
data <- list(y = rnorm(4))

code <- nimbleCode({
  for (i in 1:n){
    mu[i] <- beta0 + beta1 * x[.site[i]]
    y[i] ~ dnorm(mu[i], sd = 1)
  }
})

# errors
mod <- nimbleModel(code, constants=constants, data=data)

# Also issues when indexing a non-constant
# back to 'site'
constants <- list(n = 4, site = c(1,1,2,2))
data <- list(y = rnorm(4))

code <- nimbleCode({
  for (i in 1:n){
    y[i] ~ dnorm(mu[site[i]], sd = 1)
  }
  mu[1] ~ dnorm(0, sd = 1)
  mu[2] ~ dnorm(0, sd = 1)
})

# works
mod <- nimbleModel(code, constants=constants, data=data)
mod$modelDef$constantsNamesList # site is here

# now switch to .site
constants <- list(n = 4, .site = c(1,1,2,2))
data <- list(y = rnorm(4))


code <- nimbleCode({
  for (i in 1:n){
    y[i] ~ dnorm(mu[.site[i]], sd = 1)
  }
  mu[1] ~ dnorm(0, sd = 1)
  mu[2] ~ dnorm(0, sd = 1)
})

# gives apparently incorrect warning note about
# non-constant indexes
mod <- nimbleModel(code, constants=constants, data=data)
mod$modelDef$constantsNamesList # .site is not included
mod$modelDef$constantsList # it's here though

[paciorek]

Thanks! I have on my todo list to take a look at this, but it might not be for a bit.

[paciorek]

Fixed by PR #1629