Bug833 asdt list

[rsaporta]

Addresses issue #833
"as.data.table behavior inconsistent with as.data.frame for matrix list members #833"

Modified / Added:

• as.data.table.list
• as.data.table.matrix
• as.data.table.array (added)

details:

as.data.table.list -- checks if any element has dim, then rest of function splits into two parts (yes/no dim). If an element has dimension, several other tests are performed and eventually returns do.call(cbind, lapply(x, as.data.table)) Additionally, gave user option to use rbind instead of cbind via the argument bind.using. This is useful for list of matrices. For mixed lists, a warning is thrown to inform user that rbind may coerce data types.

as.data.table.matrix -- fixed issue where if keep.rownames=TRUE but x does not have rownames, an error would result. Solution implemented is to use rn=NA_character_. Alternative is to override user and set keep.rownames=FALSE

as.data.tabel.array -- added this function. Goal was to mimic results of as.data.frame.array as closely as possible. One issue remains. Namely, for single-dimension array, as.data.frame preserves dim info. We cannot do that with as.data.table since a column having a dim attribtue will annoy print(x).

[arunsrinivasan]

@rsaporta I just had a look at this PR.

The fix should address just #833. as.data.table.array should be separate FR, and the fix for as.data.table.matrix should be a separate bug fix.

Tests should be added to tests.Rraw (from my email). And we don't need additional arguments to as.data.table.list. IIUC, isn't the additional logic for as.data.table.list just this:

if (!is.null(unlist(lapply(x, dim))))
    return(eval(as.call(c(expression(data.table), x))))

??

Now I also realise that the output is not identical here as well:

ll = list(1:5, list(6:10, 11:15))
as.data.table(ll)
as.data.frame(ll)

@mattdowle any thoughts?