Strange recycling of factors in dplyr::mutate - bug or feature?

Question

The function mutate from the R package 'dplyr' has a peculiar recycling feature for factors, in that it seems to return the factor as.numeric. In the following example y becomes what you would expect, whereas z is c(1,1)

library(dplyr)
df <- data_frame(x=1:2)
glimpse(df %>% mutate(y="A", z=factor("B")))
# Variables:
# $ x (int) 1, 2
# $ y (chr) "A", "A"
# $ z (int) 1, 1

Is there any rationale behind this, or is it a bug?

(I am using R 3.1.1 and dplyr 0.3.0.1.)

---

EDIT:

After posting this as an issue on github, Romain Francois fixed it within hours! So if the above is a problem use devtools::install_github to get the latest version:

library(devtools)
install_github("hadley/dplyr")

and then

library(dplyr)
df <- data_frame(x=1:2)
glimpse(df %>% mutate(y="A", z=factor("B")))
# Variables:
# $ x (int) 1, 2
# $ y (chr) "A", "A"
# $ z (fctr) B, B

Nice work Romain!

Answer 1

dplyr uses C++ to perform the actual mutate operation. Following the rabbit hole and noting this is an ungrouped mutation, we can use our trusty debugger to notice the following.

debugonce(dplyr:::mutate_impl)
# Inside of mutate_impl we do:
class(dots[[2]]$expr) # which is a "call"!

So now we know the type of our lazy expression. We eval the call and notice it is a supported type (unfortunately, R's TYPEOF macro claims factors are integers - we would need Rf_isFactor to discriminate).

So what happens next? We returned the result and we're done. If you have tried (df %>% mutate(y="A", z=factor(c("A","B"))))[[3]] already, you'll know that the issue is indeed the recycling.

Specifically, the C++ Gatherer object (which should really be checking for Rf_isFactor in addition to its current date check on INTSXPs) is using C++ templating to force a Vector<INTSXP> to be created (implicitly through constructor initialization - notice the arity 2 call in ConstantGathererImpl) without remembering to carry over the factor "label."

TLDR: In R's C++, integers and factors have the same internal type when using the TYPEOF macro, and factors are a weird edge case.

Feel free to submit a pull request to dplyr, it's in active development and hadley and Romain are nice guys. You'll have to add an if statement here.