Closes #473. as.data.table.data.table checks & restores over-allocation.

R/as.data.table.R

@@ -173,5 +173,6 @@ as.data.table.data.frame <- function(x, keep.rownames=FALSE, ...) {
 as.data.table.data.table <- function(x, ...) {
     # fix for #1078 and #1128, see .resetclass() for explanation.
     setattr(x, 'class', .resetclass(x, "data.table"))
+    if (!selfrefok(x)) x = alloc.col(x) # fix for #473
     return(x)
 }

README.md

@@ -152,6 +152,8 @@
 
   45. `DT[, .BY, by=x]` and other variants of adding a column using `.BY` is handled correctly in R v3.1.0+, [#1270](https://github.com/Rdatatable/data.table/issues/1270). 
 
+  46. `as.data.table.data.table()` method checks and restores over-allocation, [#473](https://github.com/Rdatatable/data.table/issues/473). 
+
 #### NOTES
 
   1. Updated error message on invalid joins to reflect the new `on=` syntax, [#1368](https://github.com/Rdatatable/data.table/issues/1368). Thanks @MichaelChirico.

inst/tests/tests.Rraw

@@ -7646,6 +7646,11 @@ DT = data.table(x=1:2, y=5:6)
 test(1619.1, DT[, .BY, by=x]$BY, as.list(1:2))
 test(1619.2, DT[, bycol := .BY, by=x]$bycol, as.list(1:2))
 
+# fix for #473
+DT = data.frame(x=1, y=2)
+setattr(DT, 'class', c('data.table', 'data.frame')) # simulates over-allocation lost scenario
+if (!truelength(DT)) test(1620, truelength(as.data.table(DT)), 100L)
+
 ##########################
 
 # TODO: Tests involving GForce functions needs to be run with optimisation level 1 and 2, so that both functions are tested all the time.