support data.table?

[ahookom]

Hey, thanks for making this!

I have an existing code base oriented around data.table, and I'd love to get the memory savings from using these instead of numeric. I'm wondering: would it be possible to add support for inserting vectors of these floats into a data.table as a column? It seems the class would probably have to inherit from an atomic such as vector or list to make that work.

Here's the very basic thing I have tried in R-3.3.1 with float==0.2-2 and data.table==1.9.6:

> library(float)
> library(data.table)

> ex <- fl(c(3.2, 1e7, -2.8e-4))
> dt <- data.table(foo=c('bar', 'baz', 'bar'))
> dt
   foo
1: bar
2: baz
3: bar
> dt[,float:=ex]
Error in X[[i]] : this S4 class is not subsettable

borrowing from your bracket.R:

> double_bracket_float32 = function(x, i, exact=TRUE)
{
  d = x@Data[[i, exact=exact]]
  float32(d)
}
> setMethod("[[", signature(x="float32"), double_bracket_float32)

but now:

> dt[,foo:=ex]
Error in `[.data.table`(dt, , `:=`(foo, ex)) :
  RHS of assignment is not NULL, not an an atomic vector (see ?is.atomic) and not a list column.

since data.table is running is.atomic, I think the class definition would have to inherit from an atomic to get past this.

Perhaps I could wrap your class with my own that inherits from vector or list in order to achieve my goal?

Please let me know your thoughts. Thanks!