String coercion inserts spaces, as.matrix.data.table

[alistaire47]

When coercing from a data.table to a matrix and types are coerced (yes this is bad and shouldn't be done, so this is definitely low-priority), a data.table inserts spaces whereas a data.frame does not:

> as.matrix(data.frame(a = LETTERS[1:2], b = c(FALSE, TRUE)))
     a   b      
[1,] "A" "FALSE"
[2,] "B" "TRUE" 
> as.matrix(data.table(a = LETTERS[1:2], b = c(FALSE, TRUE)))
     a   b      
[1,] "A" "FALSE"
[2,] "B" " TRUE"

I assume this is a result of C++ typing somehow, but I don't know enough to make a PR, sorry.

[MichaelChirico]

as.matrix.data.frame uses as.character for logical columns, whereas as.matrix.data.table uses format -- format adds the space.

@sritchie73 is this covered by your ongoing #4144? I was unable to install & test myself for the moment

[MichaelChirico]

In fact, it seems this is something that was fixed quite recently in base R itself!

wch/r-source@bd3b7ef

So I guess our issue is a vestige of the same logic being copied over from as.matrix.data.frame many years ago

[sritchie73]

This is fixed in #4144 , this is the output I see on my working test branch:

     a   b      
[1,] "A" "FALSE"
[2,] "B" "TRUE" 

The current version of data.table on CRAN simply uses base::unlist(), while #4144 is implemented in C and uses our internal memrecycle function which has the correct desired behaviour:

> dt = data.table(a = LETTERS[1:2], b = c(FALSE, TRUE))
> rbind(dt[,.(a)], dt[,.(b)], use.names=FALSE)[,a] # rbindlist uses memrecycle for type coercion
[1] "A"     "B"     "FALSE" "TRUE"