bug#25707: [PATCH] grep: don't forcefully strip carriage returns

[Eric Blake]

On 02/14/2017 05:08 PM, Paul Eggert wrote:
> On 02/13/2017 12:20 PM, Eric Blake wrote:
>> undossify_input causes more problems than it
>> solves.  We should trust fopen("r") to do the right thing, rather than
>> reinventing it ourselves.
> 
> Yes, that makes sense. Attached is a proposed patch to implement this.
> It assumes the patch you already submitted for Bug#25707.
> 
> This patch keeps the -U option, for MS-Windows users who want to
> override fopen "-r"'s choice of binary vs text I/O. Perhaps that's too
> conservative? It would be easy to turn -U into a no-op too.

For reference, here's the patch I applied to Cygwin when building
grep-3.0; it didn't go as far as your patch in killing -u (so I like
your patch better), but had the same net effect of keeping -U as a way
to force binary behavior, while using fcntl() to learn (after-the-fact)
whether our force-to-binary was a no-op, and only doing undossify stuff
if we turned a text-mode into binary-mode:

--- origsrc/grep-3.0/src/dosbuf.c       2017-01-01 05:34:33.000000000 -0600
+++ src/grep-3.0/src/dosbuf.c   2017-02-14 09:50:08.701835000 -0600
@@ -48,6 +48,7 @@ static struct dos_map *dos_pos_map;
 static int       dos_pos_map_size  = 0;
 static int       dos_pos_map_used  = 0;
 static int       inp_map_idx = 0, out_map_idx = 1;
+static int       mode_forced;

 /* Set default DOS file type to binary.  */
 static void
@@ -95,7 +96,7 @@ guess_type (char *buf, size_t buflen)
 static size_t
 undossify_input (char *buf, size_t buflen)
 {
-  if (! O_BINARY)
+  if (! O_BINARY || ! mode_forced)
     return buflen;

   size_t bytes_left = 0;
@@ -186,7 +187,7 @@ undossify_input (char *buf, size_t bufle
 static off_t
 dossified_pos (off_t byteno)
 {
-  if (! O_BINARY)
+  if (! O_BINARY || ! mode_forced)
     return byteno;

   off_t pos_lo;
--- origsrc/grep-3.0/src/grep.c 2017-02-08 19:37:33.000000000 -0600
+++ src/grep-3.0/src/grep.c     2017-02-14 09:58:11.134855800 -0600
@@ -1860,10 +1860,17 @@ grepdesc (int desc, bool command_line)
       goto closeout;
     }

-  /* Set input to binary mode.  Pipes are simulated with files
-     on DOS, so this includes the case of "foo | grep bar".  */
+  /* Force input to binary mode, while remembering if it was originally
+     text.  Options -u and -U only matter on text mounts.  */
+  mode_forced = 0;
   if (O_BINARY && !isatty (desc))
-    set_binary_mode (desc, O_BINARY);
+    {
+      if (fcntl (desc, F_GETFL) & O_TEXT)
+       {
+         set_binary_mode (desc, O_BINARY);
+         mode_forced = 1;
+       }
+    }

   count = grep (desc, &st, &ineof);
   if (count_matches)
@@ -2586,7 +2593,7 @@ main (int argc, char **argv)
         break;

       case 'f':
-        fp = STREQ (optarg, "-") ? stdin : fopen (optarg, O_TEXT ? "rt"
: "r");
+        fp = STREQ (optarg, "-") ? stdin : fopen (optarg, "r");
         if (!fp)
           die (EXIT_TROUBLE, errno, "%s", optarg);
         oldcc = keycc;

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature