shuf: prefer fwrite over fputs and fputc

On an AMD Ryzen 7 3700X running GNU/Linux:

    $ timeout 30 taskset 1 ./src/shuf-prev \
        -r -i 1000000-1000000 | pv -r > /dev/null
    [ 302MiB/s]
    $ timeout 30 taskset 1 ./src/shuf \
        -r -i 1000000-1000000 | pv -r > /dev/null
    [ 434MiB/s]

* src/shuf.c (print_number): New function.
(write_permuted_numbers, write_random_numbers): Use it.
* tests/shuf/shuf.sh: Add a test case to run 'shuf -i' with varying
numbers of digits to check that the string conversion is correct.

diff --git a/src/shuf.c b/src/shuf.c
index 948ee88f3d8bda3d4ad6ab1766b84ff8f16eb0c0..b1c645caff29402c31dd07549af54dd446125ffb 100644 (file)
--- a/src/shuf.c
+++ b/src/shuf.c
@@ -316,6 +316,21 @@ write_permuted_lines (size_t n_lines, char *const *line,
   return 0;
 }
 
+/* Print NUMBER followed by EOLBYTE to standard output.
+   Return false on failure, true on success.  */
+static bool
+print_number (unsigned long int number, char eolbyte)
+{
+  char buf[INT_BUFSIZE_BOUND (unsigned long int)];
+  char *p = buf + INT_STRLEN_BOUND (unsigned long int);
+  *p = eolbyte;
+  do
+    *--p = '0' + number % 10;
+  while ((number /= 10) != 0);
+  idx_t len = buf + sizeof buf - p;
+  return fwrite (p, 1, len, stdout) == len;
+}
+
 /* Output N_LINES of numbers to stdout, from PERMUTATION array.
    PERMUTATION must have at least N_LINES elements.  */
 static int
@@ -325,9 +340,7 @@ write_permuted_numbers (size_t n_lines, size_t lo_input,
   for (size_t i = 0; i < n_lines; i++)
     {
       unsigned long int n = lo_input + permutation[i];
-      char buf[INT_BUFSIZE_BOUND (uintmax_t)];
-      if (fputs (umaxtostr (n, buf), stdout) < 0
-          || fputc (eolbyte, stdout) < 0)
+      if (! print_number (n, eolbyte))
         return -1;
     }
 
@@ -345,9 +358,7 @@ write_random_numbers (struct randint_source *s, size_t count,
   for (size_t i = 0; i < count; i++)
     {
       unsigned long int j = lo_input + randint_choose (s, range);
-      char buf[INT_BUFSIZE_BOUND (uintmax_t)];
-      if (fputs (umaxtostr (j, buf), stdout) < 0
-          || fputc (eolbyte, stdout) < 0)
+      if (! print_number (j, eolbyte))
         return -1;
     }
 

diff --git a/tests/shuf/shuf.sh b/tests/shuf/shuf.sh
index 346d9c95660a5a43b13fc6f6b2531a72a0e2f131..387bbb649c4534a0432f623b0213501a781d9a46 100755 (executable)
--- a/tests/shuf/shuf.sh
+++ b/tests/shuf/shuf.sh
@@ -100,6 +100,18 @@ shuf -n10 -i0-9 -n3 -n20 > exp || framework_failure_
 c=$(wc -l < exp) || framework_failure_
 test "$c" -eq 3 || { fail=1; echo "Multiple -n failed">&2 ; }
 
+# Test that the conversion from integer to string doesn't write past a buffer.
+# Note that the value is too large for shell arithmetic.
+v=$ULONG_MAX
+while :; do
+  v=$(echo $v | sed 's/^0/1/')
+  test -z "$v" && break
+  echo $v > exp
+  shuf -i $v-$v > out || fail=1
+  compare exp out || fail=1
+  v=$(echo $v | cut -b2-)
+done
+
 # Test error conditions
 
 # -i and -e must not be used together