Closes #3723 and #1459 -- use proper NA for integer64 on assign

NEWS.md

@@ -198,6 +198,8 @@
 
 24. `column not found` could incorrectly occur in rare non-equi-join cases, [#3635](https:/Rdatatable/data.table/issues/3635). Thanks to @UweBlock for the report.
 
+25. `integer64` defined on a subset of a new column would leave "gibberish" on the remaining rows, [#3723](https:/Rdatatable/data.table/issues/3723). A bug in `rbindlist` with the same root cause was also fixed, [#1459](https:/Rdatatable/data.table/issues/1459). Thanks @shrektan and @jangorecki for the reports.
+
 #### NOTES
 
 1. `rbindlist`'s `use.names="check"` now emits its message for automatic column names (`"V[0-9]+"`) too, [#3484](https:/Rdatatable/data.table/pull/3484). See news item 5 of v1.12.2 below.

inst/tests/tests.Rraw

@@ -15428,6 +15428,14 @@ test(2071.10, dcast(data.table(a=1, b=1, l=list(list(1))), a ~ b, value.var='l')
 test(2071.11, dcast(data.table(a = 1, b = 2, c = 3), a ~ b, value.var = 'c', fill = '2'),
  data.table(a=1, `2`=3, key='a'))
 
+# partial instantiation of integer64 column was creating NA_REAL, not INT64_MIN
+if (test_bit64) {
+ # sub-assign from #3723
+ test(2072.1, data.table(x=1:2)[1, y := bit64::as.integer64(0L)]$y, as.integer64(c(0L, NA)))
+ # rbindlist(fill=TRUE) from #1459
+ test(2072.2, rbind(data.table(a=1:2, b=as.integer64(c(1,NA))), data.table(a=3L), fill=TRUE)$b, as.integer64(c(1, NA, NA)))
+}
+
 ###################################
 # Add new tests above this line #
 ###################################

src/assign.c

@@ -489,7 +489,7 @@ SEXP assign(SEXP dt, SEXP rows, SEXP cols, SEXP newcolnames, SEXP values)
  SEXP RHS;
 
  if (coln+1 > oldncol) { // new column
- SET_VECTOR_ELT(dt, coln, newcol=allocNAVector(TYPEOF(thisvalue), nrow));
+ SET_VECTOR_ELT(dt, coln, newcol=allocNAVectorLike(thisvalue, nrow));
  // initialize with NAs for when 'rows' is a subset and it doesn't touch
  // do not try to save the time to NA fill (contiguous branch free assign anyway) since being
  // sure all items will be written to (isNull(rows), length(rows), vlen<1, targetlen) is not worth the risk.
@@ -1040,6 +1040,14 @@ SEXP allocNAVector(SEXPTYPE type, R_len_t n)
  UNPROTECT(1);
  return(v);
 }
+// #3723 -- build NA vector like x -- mainly used to copy attributes pass this to writeNA
+SEXP allocNAVectorLike(SEXP x, R_len_t n) {
+ SEXP v = PROTECT(allocVector(TYPEOF(x), n));
+ copyMostAttrib(x, v);
+ writeNA(v, 0, n);
+ UNPROTECT(1);
+ return(v);
+}
 
 static SEXP *saveds=NULL;
 static R_len_t *savedtl=NULL, nalloc=0, nsaved=0;

src/data.table.h

@@ -106,6 +106,7 @@ SEXP SelfRefSymbol;
 
 // assign.c
 SEXP allocNAVector(SEXPTYPE type, R_len_t n);
+SEXP allocNAVectorLike(SEXP x, R_len_t n);
 void writeNA(SEXP v, const int from, const int n);
 void savetl_init(), savetl(SEXP s), savetl_end();
 int checkOverAlloc(SEXP x);