Using ifelse with NAs in mutate sometimes results in strange coercion behavior

[joranE]

In 0.1.2,

d <- data.frame(x = rep(1:5, each = 3))
mutate(group_by(d,x),val = ifelse(x < 3, NA, 2)) #Coerces to logical
mutate(group_by(d,x),val = ifelse(x < 3, NA_real_, 2)) #Works

It seems to be very subtle, because this does not have the same behavior:

d1 <- data.frame(x = rep(1:3,each = 3),y = rep(1:3,times = 3))
mutate(group_by(d1,x),val = ifelse(y < 3, NA, 5))

[hadley]

@romainfrancois do we have a custom version of ifelse in the hybrid evaluator?

[romainfrancois]

Not yet. I'll look into it. Might be hard.

Envoyé de mon iPhone

Le 5 mars 2014 à 16:47, Hadley Wickham notifications@github.com a écrit :

@romainfrancois do we have a custom version of ifelse in the hybrid evaluator?

—
Reply to this email directly or view it on GitHub.

[romainfrancois]

What happens here is that the first group has the special responsability of finding the eventual type of the result column. Maybe I can check that subsequent groups produce compatible results and at least throw an error in that case as a first measure.

[romainfrancois]

I've put this in place:

> mutate(group_by(d,x),val = ifelse(x < 3, NA, 2)) 
Error : incompatible types, expecting a logical vector

Maybe we can instead do promotions, etc ... but that's more work

[hadley]

That seems reasonable. At some point we'll need some general coercion mechanism that can work for this, rbind, joins, etc.