diff -urNp coreutils-8.17-orig/src/sort.c coreutils-8.17/src/sort.c
--- coreutils-8.17-orig/src/sort.c	2012-08-20 13:15:39.703470009 +0200
+++ coreutils-8.17/src/sort.c	2012-08-20 13:29:47.177468563 +0200
@@ -292,6 +292,9 @@ struct merge_node_queue
                                    when popping. */
 };
 
+/* Used to implement --unique (-u). */
+static struct line saved_line;
+
 /* FIXME: None of these tables work with multibyte character sets.
    Also, there are many other bugs when handling multibyte characters.
    One way to fix this is to rewrite `sort' to use wide characters
@@ -3943,13 +3946,12 @@ queue_pop (struct merge_node_queue *queu
 static void
 write_unique (struct line const *line, FILE *tfp, char const *temp_output)
 {
-  static struct line saved;
 
   if (unique)
     {
-      if (saved.text && ! compare (line, &saved))
+      if (saved_line.text && ! compare (line, &saved_line))
         return;
-      saved = *line;
+      saved_line = *line;
     }
 
   write_line (line, tfp, temp_output);
@@ -4451,6 +4453,7 @@ sort (char *const *files, size_t nfiles,
               break;
             }
 
+          saved_line.text = NULL;
           line = buffer_linelim (&buf);
           if (buf.eof && !nfiles && !ntemps && !buf.left)
             {
diff -urNp coreutils-8.17-orig/tests/misc/sort coreutils-8.17/tests/misc/sort
--- coreutils-8.17-orig/tests/misc/sort	2012-02-03 10:22:06.000000000 +0100
+++ coreutils-8.17/tests/misc/sort	2012-08-20 13:31:38.685565488 +0200
@@ -226,6 +226,10 @@ my @Tests =
 ["15c", '-i -u', {IN=>"a\1\na\n"}, {OUT=>"a\1\n"}],
 ["15d", '-i -u', {IN=>"\1a\na\n"}, {OUT=>"\1a\n"}],
 ["15e", '-i -u', {IN=>"a\n\1\1\1\1\1a\1\1\1\1\n"}, {OUT=>"a\n"}],
+# Before 8.19, this would trigger a free-memory read.
+["unique-free-mem-read", '-u --p=1 -S32b',
+ {IN=>"a\n"."b"x900 ."\n"},
+ {OUT=>"a\n"."b"x900 ."\n"}],
 
 # From Erick Branderhorst -- fixed around 1.19e
 ["16a", '-f',