fifelse followup

[jangorecki]

• names of test could be retained

fifelse(c(a=T,b=F), 1:2, 2:1)
#[1] 1 1
ifelse(c(a=T,b=F), 1:2, 2:1)
#a b 
#1 1 

• recycling here needs to be fixed, AFAIU might result into segfault - PR propose to remove this feature

fifelse(c(TRUE, FALSE), 1:2, list(0L))
#[1]         1 544979712
ifelse(c(TRUE, FALSE), 1:2, list(0L))
#[[1]]
#[1] 1
#
#[[2]]
#[1] 0

• coerce int to double when desired (but not double to int, as in base::ifelse)

str(ifelse(c(T,F),c(1,2),11:12))
# num [1:2] 1 12
str(ifelse(c(T,F),1:2,c(11,12)))
# num [1:2] 1 12
str(ifelse(c(T,F),1:2,c(11,12)))
#Error in fifelse

• remove hutils, dplyr code from examples, pass --run-dontrun
• remove console output in examples
• int64 support NAs

[2005m]

would you agree if fifelse(c(TRUE, FALSE), 1:2, list(0L)) displays an error like this :

Error in fifelse(c(TRUE, FALSE), 1:2, list(0L)) : 
  'yes' is of type integer but 'no' is of type list. Please make sure that both arguments have the same type.

and that we have the following ?

DT1 = data.table(a = 0:5, b = list(1, list(2:3), list(4:6), list(6:4), list(3:2), 1))
DT2 = data.table(a = 0:5, b = list(1, list(2:3), list(4:6), list(6:4), list(3:2), 1))
DT1 = DT1[ , b1 := fifelse(a > 2, b, 0)]
DT2 = DT2[ , b1 := fifelse(a > 2, b, list(0))]
identical(DT1,DT2)
[1] TRUE

Let me know if you want me to test something else. I don't want to commit anything until you are happy with the behaviour.

[jangorecki]

So for len1 > 1 or len2 > 1 branches raise an error.
I would personally avoid this kind of inconsistency, but lets wait for opinion of others. I also think wrapping in a list for values to meant to be a list column elements is not bad.

[2005m]

Yes that is correct. This is consistent with what @MichaelChirico asked me to implement in my pull request. @jangorecki would you avoid en-listing or how would you expect en-listing to work?

[shrektan]

Thanks for implementing this. dplyr::if_else() has a missing argument I think it's quite helpful, because the number of all the possible logic values in R are 3, in fact: TRUE, FALSE, NA.

So do you think it's a good idea to add the missing argument in fifelse()?

Thanks.