diff --git a/R/as.data.table.R b/R/as.data.table.R
index 1addc58aa..3137f7532 100644
--- a/R/as.data.table.R
+++ b/R/as.data.table.R
@@ -169,7 +169,7 @@ as.data.table.list = function(x,
       #       not worse than before, and gets us in a better centralized place to port as.data.table.list to C and use MAYBE_REFERENCED
       #       again in future, for #617.
     }
-    if (identical(x, list())) vector("list", nrow) else rep_len(x, nrow)   # new objects don't need copy
+    if (identical(x, list())) vector("list", nrow) else safe_rep_len(x, nrow)   # new objects don't need copy
   }
   vnames = character(ncol)
   k = 1L
diff --git a/R/print.data.table.R b/R/print.data.table.R
index 2e3dfc69f..528138279 100644
--- a/R/print.data.table.R
+++ b/R/print.data.table.R
@@ -248,6 +248,7 @@ char.trunc = function(x, trunc.char = getOption("datatable.prettyprint.char")) {
   nchar_chars = nchar(x, 'char')
   is_full_width = nchar_width > nchar_chars
   idx = pmin(nchar_width, nchar_chars) > trunc.char
+  if (!any(idx)) return(x) # strtrim() errors for width=integer() on R 3.3.0
   x[idx] = paste0(strtrim(x[idx], trunc.char * fifelse(is_full_width[idx], 2L, 1L)), "...")
   x
 }
diff --git a/R/utils.R b/R/utils.R
index cf23609ee..960b17ad1 100644
--- a/R/utils.R
+++ b/R/utils.R
@@ -21,6 +21,17 @@ nan_is_na = function(x) {
   stopf("Argument 'nan' must be NA or NaN")
 }
 
+# TODO(R>=4.0.0): Remove this workaround. From R 4.0.0, rep_len() dispatches rep.Date(), which we need.
+#   Before that, rep_len() strips attributes --> breaks data.table()'s internal recycle() helper.
+if (inherits(rep_len(Sys.Date(), 1L), "Date")) {
+  # NB: safe_rep_len=rep_len throws an R CMD check error because it _appears_ to the AST
+  #   walker that we've used .Internal ourselves (which is not true, but codetools can't tell:
+  #   safe_rep_len = rep_len; body(safe_rep_len)[[1]] # .Internal)
+  safe_rep_len = function(x, n) rep_len(x, n)
+} else {
+  safe_rep_len = function(x, n) rep(x, length.out = n) # nolint: rep_len_linter.
+}
+
 # endsWith no longer used from #5097 so no need to backport; prevent usage to avoid dev delay until GLCI's R 3.1.0 test
 endsWith = function(...) stop("Internal error: use endsWithAny instead of base::endsWith", call.=FALSE)
 
diff --git a/inst/tests/tests.Rraw b/inst/tests/tests.Rraw
index f86ec3ee4..e948166f1 100644
--- a/inst/tests/tests.Rraw
+++ b/inst/tests/tests.Rraw
@@ -14788,48 +14788,52 @@ test(2029.3, fread(txt, quote="", fill=TRUE), data.table(A=1:3, B=4:6, C=c(7:8,N
 
 # .Last.updated #1885
 d = data.table(a=1:4, b=2:5)
-d[, z:=5L]
-test(2030.01, .Last.updated, 4L) # new column
-d[, z:=6L]
-test(2030.02, .Last.updated, 4L) # update existing column
-d[2:3, z:=7L]
-test(2030.03, .Last.updated, 2L) # sub assign
-d[integer(), z:=8L]
-test(2030.04, .Last.updated, 0L) # empty sub-assign
-d[-1L, z:=9L]
-test(2030.05, .Last.updated, 3L) # inverse sub-assign
-d[-(1:4), z:=10L]
-test(2030.06, .Last.updated, 0L) # inverse empty sub-assign
+# NB: On R 3.3.0, .Last.updated gets touched by test() itself and this cascades even to re-assignments
+#   of .Last.updated like n_updated=.Last.updated, so really force a deep copy by round-tripping from string.
+#   Seen on R 3.3.0, unsure when this behavior changes.
+force_copy = function(n) as.integer(sprintf("%d", n))
+d[, z:=5L]; n_updated = force_copy(.Last.updated)
+test(2030.01, n_updated, 4L) # new column
+d[, z:=6L]; n_updated = force_copy(.Last.updated)
+test(2030.02, n_updated, 4L) # update existing column
+d[2:3, z:=7L]; n_updated = force_copy(.Last.updated)
+test(2030.03, n_updated, 2L) # sub assign
+d[integer(), z:=8L]; n_updated = force_copy(.Last.updated)
+test(2030.04, n_updated, 0L) # empty sub-assign
+d[-1L, z:=9L]; n_updated = force_copy(.Last.updated)
+test(2030.05, n_updated, 3L) # inverse sub-assign
+d[-(1:4), z:=10L]; n_updated = force_copy(.Last.updated)
+test(2030.06, n_updated, 0L) # inverse empty sub-assign
+d[, z:=NULL]; n_updated = force_copy(.Last.updated)
+test(2030.07, n_updated, 4L) # delete column
+d[2:3, z:=11L]; n_updated = force_copy(.Last.updated)
+test(2030.08, n_updated, 2L) # new column during sub-assign
 d[, z:=NULL]
-test(2030.07, .Last.updated, 4L) # delete column
-d[2:3, z:=11L]
-test(2030.08, .Last.updated, 2L) # new column during sub-assign
+d[integer(), z:=12L]; n_updated = force_copy(.Last.updated)
+test(2030.09, n_updated, 0L) # new columns from empty sub-assign
 d[, z:=NULL]
-d[integer(), z:=12L]
-test(2030.09, .Last.updated, 0L) # new columns from empty sub-assign
-d[, z:=NULL]
-d[-(1:4), z:=13L]
-test(2030.10, .Last.updated, 0L) # new columns from empty inverse sub-assign
-d[, z:=NULL][, z:=14L]
-test(2030.11, .Last.updated, 4L) # new column from chaining
-d[, z:=NULL][2:3, z:=14L]
-test(2030.12, .Last.updated, 2L) # sub-assign from chaining
-d[2:3, z:=14L][, z:=NULL]
-test(2030.13, .Last.updated, 4L) # delete column from chaining
-set(d, 1:2, "z", 15L)
-test(2030.14, .Last.updated, 2L) # set() updates .Last.updated too
+d[-(1:4), z:=13L]; n_updated = force_copy(.Last.updated)
+test(2030.10, n_updated, 0L) # new columns from empty inverse sub-assign
+d[, z:=NULL][, z:=14L]; n_updated = force_copy(.Last.updated)
+test(2030.11, n_updated, 4L) # new column from chaining
+d[, z:=NULL][2:3, z:=14L]; n_updated = force_copy(.Last.updated)
+test(2030.12, n_updated, 2L) # sub-assign from chaining
+d[2:3, z:=14L][, z:=NULL]; n_updated = force_copy(.Last.updated)
+test(2030.13, n_updated, 4L) # delete column from chaining
+set(d, 1:2, "z", 15L); n_updated = force_copy(.Last.updated)
+test(2030.14, n_updated, 2L) # set() updates .Last.updated too
 g = data.table(a=1:4, z=15L) # join
-d[g, on="a", z:=i.z]
-test(2030.15, .Last.updated, 4L) # all match of all rows
+d[g, on="a", z:=i.z]; n_updated = force_copy(.Last.updated)
+test(2030.15, n_updated, 4L) # all match of all rows
 g = data.table(a=2:4, z=16L) # join
-d[, z:=NULL][g, on="a", z:=i.z]
-test(2030.16, .Last.updated, 3L) # all match
+d[, z:=NULL][g, on="a", z:=i.z]; n_updated = force_copy(.Last.updated)
+test(2030.16, n_updated, 3L) # all match
 g = data.table(a=c(2L,4L,6L), z=17L)
-d[, z:=NULL][g, on="a", z:=i.z]
-test(2030.17, .Last.updated, 2L) # partial match
+d[, z:=NULL][g, on="a", z:=i.z]; n_updated = force_copy(.Last.updated)
+test(2030.17, n_updated, 2L) # partial match
 g = data.table(a=5:6, z=18L)
-d[, z:=NULL][g, on="a", z:=i.z]
-test(2030.18, .Last.updated, 0L) # zero match
+d[, z:=NULL][g, on="a", z:=i.z]; n_updated = force_copy(.Last.updated)
+test(2030.18, n_updated, 0L) # zero match
 
 # rbind vec with list regression dev 1.12.3; #3528
 test(2031.01, rbind(data.table(A=1:3, B=7:9), data.table(A=4:6, B=as.list(10:12))), ans<-data.table(A=1:6, B=as.list(7:12)))
@@ -18453,7 +18457,7 @@ rm(.datatable.aware)
 local({
   lc_ctype = Sys.getlocale('LC_CTYPE')
   Sys.setlocale('LC_CTYPE', "en_US.UTF-8") # Japanese multibyte characters require utf8
-  on.exit({Sys.setlocale('LC_CTYPE', lc_ctype)})
+  on.exit(Sys.setlocale('LC_CTYPE', lc_ctype))
   accented_a = "\u0061\u0301"
   ja_ichi = "\u4E00"
   ja_ni = "\u4E8C"
@@ -18718,7 +18722,7 @@ test(2264.8, print(DT, show.indices=TRUE), output=ans)
 # integer64 columns print even when bit64 isn't loaded
 if (test_bit64) local({
   DT = data.table(a = 'abc', b = as.integer64(1))
-  invisible(unloadNamespace("bit64"))
+  suppressPackageStartupMessages(unloadNamespace("bit64"))
   on.exit(suppressPackageStartupMessages(library(bit64)))
   test(2265, DT, output="abc\\s*1$")
 })
diff --git a/src/assign.c b/src/assign.c
index a78b9452c..4041bf2a9 100644
--- a/src/assign.c
+++ b/src/assign.c
@@ -220,7 +220,10 @@ SEXP setdt_nrows(SEXP x)
     }
     SEXP dim_xi = getAttrib(xi, R_DimSymbol);
     R_len_t len_xi;
-    R_len_t n_dim = LENGTH(dim_xi);
+    // NB: LENGTH() produces an undefined large number here on R 3.3.0.
+    //   There's also a note in NEWS for R 3.1.0 saying length() should always be used by packages,
+    //   but with some overhead for being a function/not macro...
+    R_len_t n_dim = length(dim_xi);
     if (n_dim) {
       if (test_matrix_cols && n_dim > 1) {
         warn_matrix_column(i+1);