split: avoid unnecessary input buffering

Input buffering is best avoided because it introduces
delayed processing of output for intermittent input,
especially when the output size is less than that of
the input buffer.  This is significant when output
is being further processed which could happen if split
is writing to precreated fifos, or through --filter.

If input is arriving quickly from a pipe then this will
already be buffered before we read it, so fast arriving
input shouldn't be a performance issue.

* src/split.c (lines_split, lines_bytes_split, bytes_split,
lines_chunk_split, bytes_chunk_extract): s/full_read/safe_read/.
* THANKS.in: Mention the reporter.
* NEWS: Mention the improvement.

diff --git a/NEWS b/NEWS
index c2caa427edb3a3f4bbae5a25d2d5496dac5d024a..f68ab4f7c1d52a5aadb0069bf7ca0e403d3c1e70 100644 (file)
--- a/NEWS
+++ b/NEWS
@@ -45,6 +45,9 @@ GNU coreutils NEWS                                    -*- outline -*-
   causing name look-up errors.  Also look-ups are first done outside the chroot,
   in case the look-up within the chroot fails due to library conflicts etc.
 
+  split avoids unnecessary input buffering, immediately writing input to output
+  which is significant with --filter or when writing to fifos or stdout etc.
+
   stat and tail work better with HFS+ and HFSX.  stat -f --format=%T now reports
   the file system type, and tail -f now uses inotify for files, rather than the
   default of issuing a warning and reverting to polling.

diff --git a/THANKS.in b/THANKS.in
index 20654fe5f2ba9e60cd58ab926b454d7c78adc18a..7f14a6993ad3b236826f8457418fd61da55573d3 100644 (file)
--- a/THANKS.in
+++ b/THANKS.in
@@ -184,6 +184,7 @@ Egmont Koblinger                    egmont@uhulinux.hu
 Eirik Fuller                        eirik@hackrat.com
 Eivind                              eivindt@multinet.no
 Elbert Pol                          elbert.pol@gmail.com
+Eldon Stegall                       eldon@eldondev.com
 Eli Zaretskii                       eliz@is.elta.co.il
 Emile LeBlanc                       leblanc@math.toronto.edu
 Emmanuel Lacour                     elacour@home-dn.net

diff --git a/src/split.c b/src/split.c
index 29d3dbfe42259d354aa60e60fc0a81999ced129f..dacacaa881457d7fdeebe0d6f596e519f1802f3b 100644 (file)
--- a/src/split.c
+++ b/src/split.c
@@ -33,7 +33,6 @@
 #include "error.h"
 #include "fd-reopen.h"
 #include "fcntl--.h"
-#include "full-read.h"
 #include "full-write.h"
 #include "ioblksize.h"
 #include "quote.h"
@@ -526,8 +525,8 @@ bytes_split (uintmax_t n_bytes, char *buf, size_t bufsize, uintmax_t max_files)
 
   do
     {
-      n_read = full_read (STDIN_FILENO, buf, bufsize);
-      if (n_read < bufsize && errno)
+      n_read = safe_read (STDIN_FILENO, buf, bufsize);
+      if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);
       bp_out = buf;
       to_read = n_read;
@@ -562,7 +561,7 @@ bytes_split (uintmax_t n_bytes, char *buf, size_t bufsize, uintmax_t max_files)
             }
         }
     }
-  while (n_read == bufsize);
+  while (n_read);
 
   /* Ensure NUMBER files are created, which truncates
      any existing files or notifies any consumers on fifos.
@@ -584,8 +583,8 @@ lines_split (uintmax_t n_lines, char *buf, size_t bufsize)
 
   do
     {
-      n_read = full_read (STDIN_FILENO, buf, bufsize);
-      if (n_read < bufsize && errno)
+      n_read = safe_read (STDIN_FILENO, buf, bufsize);
+      if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);
       bp = bp_out = buf;
       eob = bp + n_read;
@@ -614,7 +613,7 @@ lines_split (uintmax_t n_lines, char *buf, size_t bufsize)
             }
         }
     }
-  while (n_read == bufsize);
+  while (n_read);
 }
 
 /* Split into pieces that are as large as possible while still not more
@@ -633,8 +632,8 @@ line_bytes_split (uintmax_t n_bytes, char *buf, size_t bufsize)
 
   do
     {
-      n_read = full_read (STDIN_FILENO, buf, bufsize);
-      if (n_read < bufsize && errno)
+      n_read = safe_read (STDIN_FILENO, buf, bufsize);
+      if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);
       size_t n_left = n_read;
       char *sob = buf;
@@ -718,7 +717,7 @@ line_bytes_split (uintmax_t n_bytes, char *buf, size_t bufsize)
             }
         }
     }
-  while (n_read == bufsize);
+  while (n_read);
 
   /* Handle no eol at end of file.  */
   if (n_hold)
@@ -762,8 +761,8 @@ lines_chunk_split (uintmax_t k, uintmax_t n, char *buf, size_t bufsize,
   while (n_written < file_size)
     {
       char *bp = buf, *eob;
-      size_t n_read = full_read (STDIN_FILENO, buf, bufsize);
-      if (n_read < bufsize && errno)
+      size_t n_read = safe_read (STDIN_FILENO, buf, bufsize);
+      if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);
       else if (n_read == 0)
         break; /* eof.  */
@@ -857,8 +856,8 @@ bytes_chunk_extract (uintmax_t k, uintmax_t n, char *buf, size_t bufsize,
 
   while (start < end)
     {
-      size_t n_read = full_read (STDIN_FILENO, buf, bufsize);
-      if (n_read < bufsize && errno)
+      size_t n_read = safe_read (STDIN_FILENO, buf, bufsize);
+      if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);
       else if (n_read == 0)
         break; /* eof.  */
@@ -998,8 +997,6 @@ lines_rr (uintmax_t k, uintmax_t n, char *buf, size_t bufsize)
   while (true)
     {
       char *bp = buf, *eob;
-      /* Use safe_read() rather than full_read() here
-         so that we process available data immediately.  */
       size_t n_read = safe_read (STDIN_FILENO, buf, bufsize);
       if (n_read == SAFE_READ_ERROR)
         error (EXIT_FAILURE, errno, "%s", infile);