pairs.stanfit delays arguments evaluation in a way that breaks lexical scoping.

[iago-lito]

The following script errors in a very confusing way:

library(rstan)

#================================================================================
# Produce two dummy `stanfit` objects

model <- rstan::stan_model(model_code = r"(
  data {
    int<lower=0> N;
    array[N] int<lower=0, upper=1> y;
  }
  parameters {
    real<lower=0, upper=1> theta;
  }
  model {
    // uniform prior on interval 0,1
    theta ~ beta(1, 1);
    y ~ bernoulli(theta);
  }
)")

data <- list(
  a = round(runif(10, 0, 1)),
  b = round(runif(15, 0, 1))
)

fits <- lapply(data, function(y) {
    rstan::sampling(model, data = list(y = y, N = length(y)))
})

#================================================================================
# Plot from within a closure:

plot_function <- function() {
  for (name in names(fits)) {
    print(name) # Correctly prints "a".
    pairs(fits[[name]], main = name)
  }
}
pdf("test.pdf")
plot_function() # Error: object 'name' not found
dev.off()

From what I understand, pairs.stanfit, for some reason, does not evaluate its main = name argument within the context of the enclosing closure, but within the global scope instead. This was the source of a very nasty bug difficult to hunt down.

As far as I can check, this is not the expected/common behaviour of pair, and the problem does not happen, with, say:

pairs(data.frame(x = runif(10, 1, 10), y = runif(10, 1, 10)), main = name)

instead.