more efficient fix for PR#14263 (via PR#18503)

git-svn-id: https://svn.r-project.org/R/trunk@84182 00db46b3-68df-0310-9c12-caf00c1e9a41

doc/NEWS.Rd

@@ -272,10 +272,11 @@
       free.  Test statistics can be suppressed by giving argument
       \code{test} a false logical value.
 
-      \item The performance of \code{write.table()} has been improved
-      for data frames with a large number of unclassed columns.
-      (Thanks to Gabriel Becker's \PR{18500}, prompted by a report from
-      Toby Dylan Hocking on the R-devel mailing list.)
+      \item The performance of \code{df[j] <- value} (including for
+      missing \code{j}) and \code{write.table(df)} has been improved for
+      data frames \code{df} with a large number of columns.  (Thanks to
+      Gabriel Becker's \PR{18500}, \PR{18503} and discussants, prompted
+      by a report from Toby Dylan Hocking on the R-devel mailing list.)
 
       \item The matrix multiply operator \code{\%*\%} is now an S3 generic,
       belonging to new group generic \code{matrixOps}.  From Tomasz

src/library/base/R/dataframe.R

@@ -15,7 +15,7 @@
 #  https://www.R-project.org/Licenses/
 
 # Statlib code by John Chambers, Bell Labs, 1994
-# Changes Copyright (C) 1998-2022 The R Core Team
+# Changes Copyright (C) 1998-2023 The R Core Team
 
 
 ## As from R 2.4.0, row.names can be either character or integer.
@@ -1052,12 +1052,7 @@ data.frame <-
             }
 	}
     else if(p > 0L)
-      for(jjj in p:1L) { # we might delete columns with NULL
-        ## ... and for that reason, we'd better ensure that jseq is increasing!
-        o <- order(jseq)
-        jseq <- jseq[o]
-        jvseq <- jvseq[o]
-
+      for(jjj in order(jseq)[p:1L]) { # we might delete columns with NULL
         jj <- jseq[jjj]
         v <- value[[ jvseq[[jjj]] ]]
         ## This is consistent with the have.i case rather than with