j copies list() columns in output (#4890)

* Ensure dt[, col] copies when col is list

* Add news item

* Add unit test

* Fix unit test

* Remove now obsolete check for get in jsub

* re-use object

* errant keystroke

* Use = instead of <-

* Add test against hidden update-by-reference too

* Check is.null() outside the loop first

---------

Co-authored-by: Michael Chirico <[email protected]>

Author: tlapak

NEWS.md

@@ -59,6 +59,8 @@ rowwiseDT(
 
 7. `fread()` performance improves when specifying `Date` among `colClasses`, [#6105](https://github.com/Rdatatable/data.table/issues/6105). One implication of the change is that the column will be an `IDate` (which also inherits from `Date`), which may affect code strongly relying on the column class to be `Date` exactly; computations with `IDate` and `Date` columns should otherwise be the same. If you strongly prefer the `Date` class, run `as.Date()` explicitly following `fread()`. Thanks @scipima for the report and @MichaelChirico for the fix.
 
+8. `dt[, col]` now returns a copy of `col` also when it is a list column, as in any other case, [#4877](https://github.com/Rdatatable/data.table/issues/4877). Thanks to @tlapak for reporting and the PR.
+
 ## NOTES
 
 1. Tests run again when some Suggests packages are missing, [#6411](https://github.com/Rdatatable/data.table/issues/6411). Thanks @aadler for the note and @MichaelChirico for the fix.

R/data.table.R

@@ -1382,10 +1382,14 @@ replace_dot_alias = function(e) {
         if (jcpy) jval = copy(jval)
       } else if (address(jval) == address(SDenv$.SD)) {
         jval = copy(jval)
-      } else if ( length(jcpy <- which(vapply_1c(jval, address) %chin% vapply_1c(SDenv, address))) ) {
-        for (jidx in jcpy) { if(!is.null(jval[[jidx]])) jval[[jidx]] = copy(jval[[jidx]]) }
-      } else if (jsub %iscall% 'get') {
-        jval = copy(jval) # fix for #1212
+      } else {
+        sd_addresses = vapply_1c(SDenv, address)
+        jcpy = which(!vapply_1b(jval, is.null) & vapply_1c(jval, address) %chin% sd_addresses)
+        if (length(jcpy)) {
+          for (jidx in jcpy) jval[[jidx]] = copy(jval[[jidx]])
+        } else if (address(jval) %chin% sd_addresses) {
+          jval = copy(jval) # fix for #4877, includes fix for #1212
+        }
       }
     }
 

inst/tests/tests.Rraw

@@ -19201,3 +19201,10 @@ test(2286,
        { gc(full = TRUE); replicate(10, FALSE); x<4 },
        `attr<-`(list(3), "class", "foo")),
      structure(list(1, 2, 3), class = "foo"))
+
+# Always return a copy of columns, even when returning a single list column, #4788
+dt = data.table(A=list('a', 'b', 'c'))
+test(2287.1, address(dt[, A]) != address(dt[, A]))
+l = dt[, A]
+dt[1L, A := .(list(1L))]
+test(2287.2, !identical(DT$A[[1L]], l[[1L]]))