When executing in parallel, vertexEffectSize() fails when the formula in the prior vertexLm() call is stored in a variable.

[dnmacdon]

This works:

library(RMINC)

#df, formulaString defined

form <- as.formula(formulaString)
sa_model <- vertexLm(form, data=df)
sa_es <- vertexEffectSize(sa_model)

This fails:

library(RMINC)
library(parallel)

#df, formulaMatrix defined
#df is data frame
#formulaMatrix contains formula strings to loop over

cl <- makeForkCluster(nNodes)

parApply( cl, formulaMatrix, 1, function(formulaString)
{
  form <- as.formula(formulaString)
  sa_model <- vertexLm(form, data=df)
  sa_es <- vertexEffectSize(sa_model)
} )

Error in checkForRemoteErrors(val) : 
  18 nodes produced errors; first error: object 'form' not found

The error occurs in line 87 of RMINC's effectsize.R. The modification shown below seems to alleviate the problem, at least in this particular use case. However, I'm not certain why these behave differently because, if I understand the R docs correctly, these two calls should be equivalent since eval defaults to the parent environment anyways.

# effectsize.R line 87. This fails when vertexEffectSize() is executed in parallel by parLapply().
mod_mat <- eval(model_call)

# Mod to force evaluation in parent environment
mod_mat <- eval.parent(model_call,n=1)