Inconsistently dropping class/attributes of column vector

[renkun-ken]

I'm creating a package to make formattable data. It works nice with data.table in most cases, but in the following cases, data.table somehow shows inconsistent behavior:

library(data.table)
library(formattable)
p <- data.table(
  id = c(1, 2, 3, 4, 5), 
  name = c("A1", "A1", "B1", "B1", "C1"),
  balance = accounting(c(52500, 36150, 25000, 18300, 7600), format = "d"),
  growth = percent(c(0.3, 0.3, 0.1, 0.15, 0.15), format = "d"),
  ready = formattable(c(TRUE, TRUE, FALSE, FALSE, TRUE), "yes", "no"))

Then p looks like

   id name balance growth ready
1:  1   A1  52,500    30%   yes
2:  2   A1  36,150    30%   yes
3:  3   B1  25,000    10%    no
4:  4   B1  18,300    15%    no
5:  5   C1   7,600    15%   yes

Then I do some aggregation:

> p[, .(balance = mean(balance)), by = .(name)]
   name balance
1:   A1  44,325
2:   B1  21,650
3:   C1   7,600
> p[, .(balance = mean(balance), growth = min(growth)), by = .(name)]
   name balance growth
1:   A1  44,325    30%
2:   B1  21,650    10%
3:   C1   7,600    15%
> p[, .(balance = mean(balance), growth = min(growth), ready = all(ready)), by = .(name)]
   name balance growth ready
1:   A1   44325    30%   yes
2:   B1   21650    10%    no
3:   C1    7600    15%   yes
> p[, .(balance = mean(balance), growth = min(growth)), by = .(name)]
   name balance growth
1:   A1  44,325    30%
2:   B1  21,650    10%
3:   C1   7,600    15%
> p[, .(balance = mean(balance), growth = min(growth), ready = all(ready)), by = .(name)]
   name balance growth ready
1:   A1   44325    30%   yes
2:   B1   21650    10%    no
3:   C1    7600    15%   yes
> p[, .(balance = last(balance), growth = min(growth), ready = all(ready)), by = .(name)]
   name balance growth ready
1:   A1  36,150    30%   yes
2:   B1  18,300    10%    no
3:   C1   7,600    15%   yes
> class(p[, .(balance = mean(balance), growth = min(growth), ready = all(ready)), by = .(name)]$balance) ## should be c("formattable", "numeric")
[1] "numeric"

In the first 2 cases and the last case, the formattable class and attributes is preserved, but in the 3rd case the balance derived from mean(balance) drops formattable numeric class while in other cases it does not behave in this way. I'm not sure if it is for some reason intended to work in this way but it pretty much looks like a bug.

I'm using the latest commit data.table@b6ea972 under R 3.2.0 and Win8.1 x64.

[renkun-ken]

I do some experiments and find that as long as a non-numeric calculation appears in j then the class/attributes do not preserve correctly.

> p[, .(balance = mean(balance), growth = mean(growth)), by = .(name)]
   name balance growth
1:   A1  44,325    30%
2:   B1  21,650    12%
3:   C1   7,600    15%
> p[, .(balance = mean(balance), growth = mean(growth), ready = all(ready)), by = .(name)]
   name balance growth ready
1:   A1   44325  0.300   yes
2:   B1   21650  0.125    no
3:   C1    7600  0.150   yes

[renkun-ken]

Thanks!