NetBSD/gnu/usr.bin/diff/diff.c

/* GNU DIFF main routine.
   Copyright (C) 1988, 1989, 1992, 1993, 1994 Free Software Foundation, Inc.

This file is part of GNU DIFF.

GNU DIFF is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.

GNU DIFF is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GNU DIFF; see the file COPYING.  If not, write to
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.  */

/* GNU DIFF was written by Mike Haertel, David Hayes,
   Richard Stallman, Len Tower, and Paul Eggert.  */

#define GDIFF_MAIN
#include "diff.h"
#include <signal.h>
#include "getopt.h"
#include "fnmatch.h"

#ifndef DEFAULT_WIDTH
#define DEFAULT_WIDTH 130
#endif

#ifndef GUTTER_WIDTH_MINIMUM
#define GUTTER_WIDTH_MINIMUM 3
#endif

static char const *filetype PARAMS((struct stat const *));
static char *option_list PARAMS((char **, int));
static int add_exclude_file PARAMS((char const *));
static int ck_atoi PARAMS((char const *, int *));
static int compare_files PARAMS((char const *, char const *, char const *, char const *, int));
static int specify_format PARAMS((char **, char *));
static void add_exclude PARAMS((char const *));
static void add_regexp PARAMS((struct regexp_list **, char const *));
static void specify_style PARAMS((enum output_style));
static void try_help PARAMS((char const *));
static void check_stdout PARAMS((void));
static void usage PARAMS((void));

/* Nonzero for -r: if comparing two directories,
   compare their common subdirectories recursively.  */

static int recursive;

/* For debugging: don't do discard_confusing_lines.  */

int no_discards;

#if HAVE_SETMODE
/* I/O mode: nonzero only if using binary input/output.  */
static int binary_I_O;
#endif

/* Return a string containing the command options with which diff was invoked.
   Spaces appear between what were separate ARGV-elements.
   There is a space at the beginning but none at the end.
   If there were no options, the result is an empty string.

   Arguments: OPTIONVEC, a vector containing separate ARGV-elements, and COUNT,
   the length of that vector.  */

static char *
option_list (optionvec, count)
     char **optionvec;  /* Was `vector', but that collides on Alliant.  */
     int count;
{
  int i;
  size_t length = 0;
  char *result;

  for (i = 0; i < count; i++)
    length += strlen (optionvec[i]) + 1;

  result = xmalloc (length + 1);
  result[0] = 0;

  for (i = 0; i < count; i++)
    {
      strcat (result, " ");
      strcat (result, optionvec[i]);
    }

  return result;
}

/* Convert STR to a positive integer, storing the result in *OUT.
   If STR is not a valid integer, return -1 (otherwise 0). */
static int
ck_atoi (str, out)
     char const *str;
     int *out;
{
  char const *p;
  for (p = str; *p; p++)
    if (*p < '0' || *p > '9')
      return -1;

  *out = atoi (optarg);
  return 0;
}

/* Keep track of excluded file name patterns.  */

static char const **exclude;
static int exclude_alloc, exclude_count;

int
excluded_filename (f)
     char const *f;
{
  int i;
  for (i = 0;  i < exclude_count;  i++)
    if (fnmatch (exclude[i], f, 0) == 0)
      return 1;
  return 0;
}

static void
add_exclude (pattern)
     char const *pattern;
{
  if (exclude_alloc <= exclude_count)
    exclude = (char const **)
	      (exclude_alloc == 0
	       ? xmalloc ((exclude_alloc = 64) * sizeof (*exclude))
	       : xrealloc (exclude, (exclude_alloc *= 2) * sizeof (*exclude)));

  exclude[exclude_count++] = pattern;
}

static int
add_exclude_file (name)
     char const *name;
{
  struct file_data f;
  char *p, *q, *lim;

  f.name = optarg;
  f.desc = (strcmp (optarg, "-") == 0
	    ? STDIN_FILENO
	    : open (optarg, O_RDONLY, 0));
  if (f.desc < 0 || fstat (f.desc, &f.stat) != 0)
    return -1;

  sip (&f, 1);
  slurp (&f);

  for (p = f.buffer, lim = p + f.buffered_chars;  p < lim;  p = q)
    {
      q = (char *) memchr (p, '\n', lim - p);
      if (!q)
	q = lim;
      *q++ = 0;
      add_exclude (p);
    }

  return close (f.desc);
}

/* The numbers 129- that appear in the fourth element of some entries
   tell the big switch in `main' how to process those options.  */

static struct option const longopts[] =
{
  {"ignore-blank-lines", 0, 0, 'B'},
  {"context", 2, 0, 'C'},
  {"ifdef", 1, 0, 'D'},
  {"show-function-line", 1, 0, 'F'},
  {"speed-large-files", 0, 0, 'H'},
  {"ignore-matching-lines", 1, 0, 'I'},
  {"label", 1, 0, 'L'},
  {"file-label", 1, 0, 'L'},	/* An alias, no longer recommended */
  {"new-file", 0, 0, 'N'},
  {"entire-new-file", 0, 0, 'N'},	/* An alias, no longer recommended */
  {"unidirectional-new-file", 0, 0, 'P'},
  {"starting-file", 1, 0, 'S'},
  {"initial-tab", 0, 0, 'T'},
  {"width", 1, 0, 'W'},
  {"text", 0, 0, 'a'},
  {"ascii", 0, 0, 'a'},		/* An alias, no longer recommended */
  {"ignore-space-change", 0, 0, 'b'},
  {"minimal", 0, 0, 'd'},
  {"ed", 0, 0, 'e'},
  {"forward-ed", 0, 0, 'f'},
  {"ignore-case", 0, 0, 'i'},
  {"paginate", 0, 0, 'l'},
  {"print", 0, 0, 'l'},		/* An alias, no longer recommended */
  {"rcs", 0, 0, 'n'},
  {"show-c-function", 0, 0, 'p'},
  {"brief", 0, 0, 'q'},
  {"recursive", 0, 0, 'r'},
  {"report-identical-files", 0, 0, 's'},
  {"expand-tabs", 0, 0, 't'},
  {"version", 0, 0, 'v'},
  {"ignore-all-space", 0, 0, 'w'},
  {"exclude", 1, 0, 'x'},
  {"exclude-from", 1, 0, 'X'},
  {"side-by-side", 0, 0, 'y'},
  {"unified", 2, 0, 'U'},
  {"left-column", 0, 0, 129},
  {"suppress-common-lines", 0, 0, 130},
  {"sdiff-merge-assist", 0, 0, 131},
  {"old-line-format", 1, 0, 132},
  {"new-line-format", 1, 0, 133},
  {"unchanged-line-format", 1, 0, 134},
  {"line-format", 1, 0, 135},
  {"old-group-format", 1, 0, 136},
  {"new-group-format", 1, 0, 137},
  {"unchanged-group-format", 1, 0, 138},
  {"changed-group-format", 1, 0, 139},
  {"horizon-lines", 1, 0, 140},
  {"help", 0, 0, 141},
  {"binary", 0, 0, 142},
  {0, 0, 0, 0}
};

int
main (argc, argv)
     int argc;
     char *argv[];
{
  int val;
  int c;
  int prev = -1;
  int width = DEFAULT_WIDTH;
  int show_c_function = 0;

  /* Do our initializations.  */
  initialize_main (&argc, &argv);
  program_name = argv[0];
  output_style = OUTPUT_NORMAL;
  context = -1;

  /* Decode the options.  */

  while ((c = getopt_long (argc, argv,
			   "0123456789abBcC:dD:efF:hHiI:lL:nNpPqrsS:tTuU:vwW:x:X:y",
			   longopts, 0)) != EOF)
    {
      switch (c)
	{
	  /* All digits combine in decimal to specify the context-size.  */
	case '1':
	case '2':
	case '3':
	case '4':
	case '5':
	case '6':
	case '7':
	case '8':
	case '9':
	case '0':
	  if (context == -1)
	    context = 0;
	  /* If a context length has already been specified,
	     more digits allowed only if they follow right after the others.
	     Reject two separate runs of digits, or digits after -C.  */
	  else if (prev < '0' || prev > '9')
	    fatal ("context length specified twice");

	  context = context * 10 + c - '0';
	  break;

	case 'a':
	  /* Treat all files as text files; never treat as binary.  */
	  always_text_flag = 1;
	  break;

	case 'b':
	  /* Ignore changes in amount of white space.  */
	  ignore_space_change_flag = 1;
	  ignore_some_changes = 1;
	  ignore_some_line_changes = 1;
	  break;

	case 'B':
	  /* Ignore changes affecting only blank lines.  */
	  ignore_blank_lines_flag = 1;
	  ignore_some_changes = 1;
	  break;

	case 'C':		/* +context[=lines] */
	case 'U':		/* +unified[=lines] */
	  if (optarg)
	    {
	      if (context >= 0)
		fatal ("context length specified twice");

	      if (ck_atoi (optarg, &context))
		fatal ("invalid context length argument");
	    }

	  /* Falls through.  */
	case 'c':
	  /* Make context-style output.  */
	  specify_style (c == 'U' ? OUTPUT_UNIFIED : OUTPUT_CONTEXT);
	  break;

	case 'd':
	  /* Don't discard lines.  This makes things slower (sometimes much
	     slower) but will find a guaranteed minimal set of changes.  */
	  no_discards = 1;
	  break;

	case 'D':
	  /* Make merged #ifdef output.  */
	  specify_style (OUTPUT_IFDEF);
	  {
	    int i, err = 0;
	    static char const C_ifdef_group_formats[] =
	      "#ifndef %s\n%%<#endif /* not %s */\n%c#ifdef %s\n%%>#endif /* %s */\n%c%%=%c#ifndef %s\n%%<#else /* %s */\n%%>#endif /* %s */\n";
	    char *b = xmalloc (sizeof (C_ifdef_group_formats)
			       + 7 * strlen(optarg) - 14 /* 7*"%s" */
			       - 8 /* 5*"%%" + 3*"%c" */);
	    sprintf (b, C_ifdef_group_formats,
		     optarg, optarg, 0,
		     optarg, optarg, 0, 0,
		     optarg, optarg, optarg);
	    for (i = 0; i < 4; i++)
	      {
		err |= specify_format (&group_format[i], b);
		b += strlen (b) + 1;
	      }
	    if (err)
	      error ("conflicting #ifdef formats", 0, 0);
	  }
	  break;

	case 'e':
	  /* Make output that is a valid `ed' script.  */
	  specify_style (OUTPUT_ED);
	  break;

	case 'f':
	  /* Make output that looks vaguely like an `ed' script
	     but has changes in the order they appear in the file.  */
	  specify_style (OUTPUT_FORWARD_ED);
	  break;

	case 'F':
	  /* Show, for each set of changes, the previous line that
	     matches the specified regexp.  Currently affects only
	     context-style output.  */
	  add_regexp (&function_regexp_list, optarg);
	  break;

	case 'h':
	  /* Split the files into chunks of around 1500 lines
	     for faster processing.  Usually does not change the result.

	     This currently has no effect.  */
	  break;

	case 'H':
	  /* Turn on heuristics that speed processing of large files
	     with a small density of changes.  */
	  heuristic = 1;
	  break;

	case 'i':
	  /* Ignore changes in case.  */
	  ignore_case_flag = 1;
	  ignore_some_changes = 1;
	  ignore_some_line_changes = 1;
	  break;

	case 'I':
	  /* Ignore changes affecting only lines that match the
	     specified regexp.  */
	  add_regexp (&ignore_regexp_list, optarg);
	  ignore_some_changes = 1;
	  break;

	case 'l':
	  /* Pass the output through `pr' to paginate it.  */
	  paginate_flag = 1;
#if !defined(SIGCHLD) && defined(SIGCLD)
#define SIGCHLD SIGCLD
#endif
#ifdef SIGCHLD
	  /* Pagination requires forking and waiting, and
	     System V fork+wait does not work if SIGCHLD is ignored.  */
	  signal (SIGCHLD, SIG_DFL);
#endif
	  break;

	case 'L':
	  /* Specify file labels for `-c' output headers.  */
	  if (!file_label[0])
	    file_label[0] = optarg;
	  else if (!file_label[1])
	    file_label[1] = optarg;
	  else
	    fatal ("too many file label options");
	  break;

	case 'n':
	  /* Output RCS-style diffs, like `-f' except that each command
	     specifies the number of lines affected.  */
	  specify_style (OUTPUT_RCS);
	  break;

	case 'N':
	  /* When comparing directories, if a file appears only in one
	     directory, treat it as present but empty in the other.  */
	  entire_new_file_flag = 1;
	  break;

	case 'p':
	  /* Make context-style output and show name of last C function.  */
	  show_c_function = 1;
	  add_regexp (&function_regexp_list, "^[_a-zA-Z$]");
	  break;

	case 'P':
	  /* When comparing directories, if a file appears only in
	     the second directory of the two,
	     treat it as present but empty in the other.  */
	  unidirectional_new_file_flag = 1;
	  break;

	case 'q':
	  no_details_flag = 1;
	  break;

	case 'r':
	  /* When comparing directories,
	     recursively compare any subdirectories found.  */
	  recursive = 1;
	  break;

	case 's':
	  /* Print a message if the files are the same.  */
	  print_file_same_flag = 1;
	  break;

	case 'S':
	  /* When comparing directories, start with the specified
	     file name.  This is used for resuming an aborted comparison.  */
	  dir_start_file = optarg;
	  break;

	case 't':
	  /* Expand tabs to spaces in the output so that it preserves
	     the alignment of the input files.  */
	  tab_expand_flag = 1;
	  break;

	case 'T':
	  /* Use a tab in the output, rather than a space, before the
	     text of an input line, so as to keep the proper alignment
	     in the input line without changing the characters in it.  */
	  tab_align_flag = 1;
	  break;

	case 'u':
	  /* Output the context diff in unidiff format.  */
	  specify_style (OUTPUT_UNIFIED);
	  break;

	case 'v':
	  printf ("diff - GNU diffutils version %s\n", version_string);
	  exit (0);

	case 'w':
	  /* Ignore horizontal white space when comparing lines.  */
	  ignore_all_space_flag = 1;
	  ignore_some_changes = 1;
	  ignore_some_line_changes = 1;
	  break;

	case 'x':
	  add_exclude (optarg);
	  break;

	case 'X':
	  if (add_exclude_file (optarg) != 0)
	    pfatal_with_name (optarg);
	  break;

	case 'y':
	  /* Use side-by-side (sdiff-style) columnar output. */
	  specify_style (OUTPUT_SDIFF);
	  break;

	case 'W':
	  /* Set the line width for OUTPUT_SDIFF.  */
	  if (ck_atoi (optarg, &width) || width <= 0)
	    fatal ("column width must be a positive integer");
	  break;

	case 129:
	  sdiff_left_only = 1;
	  break;

	case 130:
	  sdiff_skip_common_lines = 1;
	  break;

	case 131:
	  /* sdiff-style columns output. */
	  specify_style (OUTPUT_SDIFF);
	  sdiff_help_sdiff = 1;
	  break;

	case 132:
	case 133:
	case 134:
	  specify_style (OUTPUT_IFDEF);
	  if (specify_format (&line_format[c - 132], optarg) != 0)
	    error ("conflicting line format", 0, 0);
	  break;

	case 135:
	  specify_style (OUTPUT_IFDEF);
	  {
	    int i, err = 0;
	    for (i = 0; i < sizeof (line_format) / sizeof (*line_format); i++)
	      err |= specify_format (&line_format[i], optarg);
	    if (err)
	      error ("conflicting line format", 0, 0);
	  }
	  break;

	case 136:
	case 137:
	case 138:
	case 139:
	  specify_style (OUTPUT_IFDEF);
	  if (specify_format (&group_format[c - 136], optarg) != 0)
	    error ("conflicting group format", 0, 0);
	  break;

	case 140:
	  if (ck_atoi (optarg, &horizon_lines) || horizon_lines < 0)
	    fatal ("horizon must be a nonnegative integer");
	  break;

	case 141:
	  usage ();
	  check_stdout ();
	  exit (0);

	case 142:
	  /* Use binary I/O when reading and writing data.
	     On Posix hosts, this has no effect.  */
#if HAVE_SETMODE
	  binary_I_O = 1;
	  setmode (STDOUT_FILENO, O_BINARY);
#endif
	  break;

	default:
	  try_help (0);
	}
      prev = c;
    }

  if (argc - optind != 2)
    try_help (argc - optind < 2 ? "missing operand" : "extra operand");


  {
    /*
     *	We maximize first the half line width, and then the gutter width,
     *	according to the following constraints:
     *	1.  Two half lines plus a gutter must fit in a line.
     *	2.  If the half line width is nonzero:
     *	    a.  The gutter width is at least GUTTER_WIDTH_MINIMUM.
     *	    b.  If tabs are not expanded to spaces,
     *		a half line plus a gutter is an integral number of tabs,
     *		so that tabs in the right column line up.
     */
    int t = tab_expand_flag ? 1 : TAB_WIDTH;
    int off = (width + t + GUTTER_WIDTH_MINIMUM) / (2*t)  *  t;
    sdiff_half_width = max (0, min (off - GUTTER_WIDTH_MINIMUM, width - off)),
    sdiff_column2_offset = sdiff_half_width ? off : width;
  }

  if (show_c_function && output_style != OUTPUT_UNIFIED)
    specify_style (OUTPUT_CONTEXT);

  if (output_style != OUTPUT_CONTEXT && output_style != OUTPUT_UNIFIED)
    context = 0;
  else if (context == -1)
    /* Default amount of context for -c.  */
    context = 3;

  if (output_style == OUTPUT_IFDEF)
    {
      /* Format arrays are char *, not char const *,
	 because integer formats are temporarily modified.
	 But it is safe to assign a constant like "%=" to a format array,
	 since "%=" does not format any integers.  */
      int i;
      for (i = 0; i < sizeof (line_format) / sizeof (*line_format); i++)
	if (!line_format[i])
	  line_format[i] = "%l\n";
      if (!group_format[OLD])
	group_format[OLD]
	  = group_format[UNCHANGED] ? group_format[UNCHANGED] : "%<";
      if (!group_format[NEW])
	group_format[NEW]
	  = group_format[UNCHANGED] ? group_format[UNCHANGED] : "%>";
      if (!group_format[UNCHANGED])
	group_format[UNCHANGED] = "%=";
      if (!group_format[CHANGED])
	group_format[CHANGED] = concat (group_format[OLD],
					group_format[NEW], "");
    }

  no_diff_means_no_output =
    (output_style == OUTPUT_IFDEF ?
      (!*group_format[UNCHANGED]
       || (strcmp (group_format[UNCHANGED], "%=") == 0
	   && !*line_format[UNCHANGED]))
     : output_style == OUTPUT_SDIFF ? sdiff_skip_common_lines : 1);

  switch_string = option_list (argv + 1, optind - 1);

  val = compare_files (0, argv[optind], 0, argv[optind + 1], 0);

  /* Print any messages that were saved up for last.  */
  print_message_queue ();

  check_stdout ();
  exit (val);
  return val;
}

/* Add the compiled form of regexp PATTERN to REGLIST.  */

static void
add_regexp (reglist, pattern)
     struct regexp_list **reglist;
     char const *pattern;
{
  struct regexp_list *r;
  char const *m;

  r = (struct regexp_list *) xmalloc (sizeof (*r));
  bzero (r, sizeof (*r));
  r->buf.fastmap = xmalloc (256);
  m = re_compile_pattern (pattern, strlen (pattern), &r->buf);
  if (m != 0)
    error ("%s: %s", pattern, m);

  /* Add to the start of the list, since it's easier than the end.  */
  r->next = *reglist;
  *reglist = r;
}

static void
try_help (reason)
     char const *reason;
{
  if (reason)
    error ("%s", reason, 0);
  error ("Try `%s --help' for more information.", program_name, 0);
  exit (2);
}

static void
check_stdout ()
{
  if (ferror (stdout) || fclose (stdout) != 0)
    fatal ("write error");
}

static char const * const option_help[] = {
"-i  --ignore-case  Consider upper- and lower-case to be the same.",
"-w  --ignore-all-space  Ignore all white space.",
"-b  --ignore-space-change  Ignore changes in the amount of white space.",
"-B  --ignore-blank-lines  Ignore changes whose lines are all blank.",
"-I RE  --ignore-matching-lines=RE  Ignore changes whose lines all match RE.",
#if HAVE_SETMODE
"--binary  Read and write data in binary mode.",
#endif
"-a  --text  Treat all files as text.\n",
"-c  -C NUM  --context[=NUM]  Output NUM (default 2) lines of copied context.",
"-u  -U NUM  --unified[=NUM]  Output NUM (default 2) lines of unified context.",
"  -NUM  Use NUM context lines.",
"  -L LABEL  --label LABEL  Use LABEL instead of file name.",
"  -p  --show-c-function  Show which C function each change is in.",
"  -F RE  --show-function-line=RE  Show the most recent line matching RE.",
"-q  --brief  Output only whether files differ.",
"-e  --ed  Output an ed script.",
"-n  --rcs  Output an RCS format diff.",
"-y  --side-by-side  Output in two columns.",
"  -w NUM  --width=NUM  Output at most NUM (default 130) characters per line.",
"  --left-column  Output only the left column of common lines.",
"  --suppress-common-lines  Do not output common lines.",
"-DNAME  --ifdef=NAME  Output merged file to show `#ifdef NAME' diffs.",
"--GTYPE-group-format=GFMT  Similar, but format GTYPE input groups with GFMT.",
"--line-format=LFMT  Similar, but format all input lines with LFMT.",
"--LTYPE-line-format=LFMT  Similar, but format LTYPE input lines with LFMT.",
"  LTYPE is `old', `new', or `unchanged'.  GTYPE is LTYPE or `changed'.",
"  GFMT may contain:",
"    %<  lines from FILE1",
"    %>  lines from FILE2",
"    %=  lines common to FILE1 and FILE2",
"    %[-][WIDTH][.[PREC]]{doxX}LETTER  printf-style spec for LETTER",
"      LETTERs are as follows for new group, lower case for old group:",
"        F  first line number",
"        L  last line number",
"        N  number of lines = L-F+1",
"        E  F-1",
"        M  L+1",
"  LFMT may contain:",
"    %L  contents of line",
"    %l  contents of line, excluding any trailing newline",
"    %[-][WIDTH][.[PREC]]{doxX}n  printf-style spec for input line number",
"  Either GFMT or LFMT may contain:",
"    %%  %",
"    %c'C'  the single character C",
"    %c'\\OOO'  the character with octal code OOO\n",
"-l  --paginate  Pass the output through `pr' to paginate it.",
"-t  --expand-tabs  Expand tabs to spaces in output.",
"-T  --initial-tab  Make tabs line up by prepending a tab.\n",
"-r  --recursive  Recursively compare any subdirectories found.",
"-N  --new-file  Treat absent files as empty.",
"-P  --unidirectional-new-file  Treat absent first files as empty.",
"-s  --report-identical-files  Report when two files are the same.",
"-x PAT  --exclude=PAT  Exclude files that match PAT.",
"-X FILE  --exclude-from=FILE  Exclude files that match any pattern in FILE.",
"-S FILE  --starting-file=FILE  Start with FILE when comparing directories.\n",
"--horizon-lines=NUM  Keep NUM lines of the common prefix and suffix.",
"-d  --minimal  Try hard to find a smaller set of changes.",
"-H  --speed-large-files  Assume large files and many scattered small changes.\n",
"-v  --version  Output version info.",
"--help  Output this help.",
0
};

static void
usage ()
{
  char const * const *p;

  printf ("Usage: %s [OPTION]... FILE1 FILE2\n\n", program_name);
  for (p = option_help;  *p;  p++)
    printf ("  %s\n", *p);
  printf ("\nIf FILE1 or FILE2 is `-', read standard input.\n");
}

static int
specify_format (var, value)
     char **var;
     char *value;
{
  int err = *var ? strcmp (*var, value) : 0;
  *var = value;
  return err;
}

static void
specify_style (style)
     enum output_style style;
{
  if (output_style != OUTPUT_NORMAL
      && output_style != style)
    error ("conflicting specifications of output style", 0, 0);
  output_style = style;
}

static char const *
filetype (st)
     struct stat const *st;
{
  /* See Posix.2 section 4.17.6.1.1 and Table 5-1 for these formats.
     To keep diagnostics grammatical, the returned string must start
     with a consonant.  */

  if (S_ISREG (st->st_mode))
    {
      if (st->st_size == 0)
	return "regular empty file";
      /* Posix.2 section 5.14.2 seems to suggest that we must read the file
	 and guess whether it's C, Fortran, etc., but this is somewhat useless
	 and doesn't reflect historical practice.  We're allowed to guess
	 wrong, so we don't bother to read the file.  */
      return "regular file";
    }
  if (S_ISDIR (st->st_mode)) return "directory";

  /* other Posix.1 file types */
#ifdef S_ISBLK
  if (S_ISBLK (st->st_mode)) return "block special file";
#endif
#ifdef S_ISCHR
  if (S_ISCHR (st->st_mode)) return "character special file";
#endif
#ifdef S_ISFIFO
  if (S_ISFIFO (st->st_mode)) return "fifo";
#endif

  /* other Posix.1b file types */
#ifdef S_TYPEISMQ
  if (S_TYPEISMQ (st)) return "message queue";
#endif
#ifdef S_TYPEISSEM
  if (S_TYPEISSEM (st)) return "semaphore";
#endif
#ifdef S_TYPEISSHM
  if (S_TYPEISSHM (st)) return "shared memory object";
#endif

  /* other popular file types */
  /* S_ISLNK is impossible with `fstat' and `stat'.  */
#ifdef S_ISSOCK
  if (S_ISSOCK (st->st_mode)) return "socket";
#endif

  return "weird file";
}

/* Compare two files (or dirs) with specified names
   DIR0/NAME0 and DIR1/NAME1, at level DEPTH in directory recursion.
   (if DIR0 is 0, then the name is just NAME0, etc.)
   This is self-contained; it opens the files and closes them.

   Value is 0 if files are the same, 1 if different,
   2 if there is a problem opening them.  */

static int
compare_files (dir0, name0, dir1, name1, depth)
     char const *dir0, *dir1;
     char const *name0, *name1;
     int depth;
{
  struct file_data inf[2];
  register int i;
  int val;
  int same_files;
  int failed = 0;
  char *free0 = 0, *free1 = 0;

  /* If this is directory comparison, perhaps we have a file
     that exists only in one of the directories.
     If so, just print a message to that effect.  */

  if (! ((name0 != 0 && name1 != 0)
	 || (unidirectional_new_file_flag && name1 != 0)
	 || entire_new_file_flag))
    {
      char const *name = name0 == 0 ? name1 : name0;
      char const *dir = name0 == 0 ? dir1 : dir0;
      message ("Only in %s: %s\n", dir, name);
      /* Return 1 so that diff_dirs will return 1 ("some files differ").  */
      return 1;
    }

  bzero (inf, sizeof (inf));

  /* Mark any nonexistent file with -1 in the desc field.  */
  /* Mark unopened files (e.g. directories) with -2. */

  inf[0].desc = name0 == 0 ? -1 : -2;
  inf[1].desc = name1 == 0 ? -1 : -2;

  /* Now record the full name of each file, including nonexistent ones.  */

  if (name0 == 0)
    name0 = name1;
  if (name1 == 0)
    name1 = name0;

  inf[0].name = dir0 == 0 ? name0 : (free0 = dir_file_pathname (dir0, name0));
  inf[1].name = dir1 == 0 ? name1 : (free1 = dir_file_pathname (dir1, name1));

  /* Stat the files.  Record whether they are directories.  */

  for (i = 0; i <= 1; i++)
    {
      if (inf[i].desc != -1)
	{
	  int stat_result;

	  if (i && filename_cmp (inf[i].name, inf[0].name) == 0)
	    {
	      inf[i].stat = inf[0].stat;
	      stat_result = 0;
	    }
	  else if (strcmp (inf[i].name, "-") == 0)
	    {
	      inf[i].desc = STDIN_FILENO;
	      stat_result = fstat (STDIN_FILENO, &inf[i].stat);
	      if (stat_result == 0 && S_ISREG (inf[i].stat.st_mode))
		{
		  off_t pos = lseek (STDIN_FILENO, (off_t) 0, SEEK_CUR);
		  if (pos == -1)
		    stat_result = -1;
		  else
		    {
		      if (pos <= inf[i].stat.st_size)
			inf[i].stat.st_size -= pos;
		      else
			inf[i].stat.st_size = 0;
		      /* Posix.2 4.17.6.1.4 requires current time for stdin.  */
		      time (&inf[i].stat.st_mtime);
		    }
		}
	    }
	  else
	    stat_result = stat (inf[i].name, &inf[i].stat);

	  if (stat_result != 0)
	    {
	      perror_with_name (inf[i].name);
	      failed = 1;
	    }
	  else
	    {
	      inf[i].dir_p = S_ISDIR (inf[i].stat.st_mode) && inf[i].desc != 0;
	      if (inf[1 - i].desc == -1)
		{
		  inf[1 - i].dir_p = inf[i].dir_p;
		  inf[1 - i].stat.st_mode = inf[i].stat.st_mode;
		}
	    }
	}
    }

  if (! failed && depth == 0 && inf[0].dir_p != inf[1].dir_p)
    {
      /* If one is a directory, and it was specified in the command line,
	 use the file in that dir with the other file's basename.  */

      int fnm_arg = inf[0].dir_p;
      int dir_arg = 1 - fnm_arg;
      char const *fnm = inf[fnm_arg].name;
      char const *dir = inf[dir_arg].name;
      char const *p = filename_lastdirchar (fnm);
      char const *filename = inf[dir_arg].name
	= dir_file_pathname (dir, p ? p + 1 : fnm);

      if (strcmp (fnm, "-") == 0)
	fatal ("can't compare - to a directory");

      if (stat (filename, &inf[dir_arg].stat) != 0)
	{
	  perror_with_name (filename);
	  failed = 1;
	}
      else
	inf[dir_arg].dir_p = S_ISDIR (inf[dir_arg].stat.st_mode);
    }

  if (failed)
    {

      /* If either file should exist but does not, return 2.  */

      val = 2;

    }
  else if ((same_files = inf[0].desc != -1 && inf[1].desc != -1
			 && 0 < same_file (&inf[0].stat, &inf[1].stat))
	   && no_diff_means_no_output)
    {
      /* The two named files are actually the same physical file.
	 We know they are identical without actually reading them.  */

      val = 0;
    }
  else if (inf[0].dir_p & inf[1].dir_p)
    {
      if (output_style == OUTPUT_IFDEF)
	fatal ("-D option not supported with directories");

      /* If both are directories, compare the files in them.  */

      if (depth > 0 && !recursive)
	{
	  /* But don't compare dir contents one level down
	     unless -r was specified.  */
	  message ("Common subdirectories: %s and %s\n",
		   inf[0].name, inf[1].name);
	  val = 0;
	}
      else
	{
	  val = diff_dirs (inf, compare_files, depth);
	}

    }
  else if ((inf[0].dir_p | inf[1].dir_p)
	   || (depth > 0
	       && (! S_ISREG (inf[0].stat.st_mode)
		   || ! S_ISREG (inf[1].stat.st_mode))))
    {
      /* Perhaps we have a subdirectory that exists only in one directory.
	 If so, just print a message to that effect.  */

      if (inf[0].desc == -1 || inf[1].desc == -1)
	{
	  if ((inf[0].dir_p | inf[1].dir_p)
	      && recursive
	      && (entire_new_file_flag
		  || (unidirectional_new_file_flag && inf[0].desc == -1)))
	    val = diff_dirs (inf, compare_files, depth);
	  else
	    {
	      char const *dir = (inf[0].desc == -1) ? dir1 : dir0;
	      /* See Posix.2 section 4.17.6.1.1 for this format.  */
	      message ("Only in %s: %s\n", dir, name0);
	      val = 1;
	    }
	}
      else
	{
	  /* We have two files that are not to be compared.  */

	  /* See Posix.2 section 4.17.6.1.1 for this format.  */
	  message5 ("File %s is a %s while file %s is a %s\n",
		    inf[0].name, filetype (&inf[0].stat),
		    inf[1].name, filetype (&inf[1].stat));

	  /* This is a difference.  */
	  val = 1;
	}
    }
  else if ((no_details_flag & ~ignore_some_changes)
	   && inf[0].stat.st_size != inf[1].stat.st_size
	   && (inf[0].desc == -1 || S_ISREG (inf[0].stat.st_mode))
	   && (inf[1].desc == -1 || S_ISREG (inf[1].stat.st_mode)))
    {
      message ("Files %s and %s differ\n", inf[0].name, inf[1].name);
      val = 1;
    }
  else
    {
      /* Both exist and neither is a directory.  */

      /* Open the files and record their descriptors.  */

      if (inf[0].desc == -2)
	if ((inf[0].desc = open (inf[0].name, O_RDONLY, 0)) < 0)
	  {
	    perror_with_name (inf[0].name);
	    failed = 1;
	  }
      if (inf[1].desc == -2)
	if (same_files)
	  inf[1].desc = inf[0].desc;
	else if ((inf[1].desc = open (inf[1].name, O_RDONLY, 0)) < 0)
	  {
	    perror_with_name (inf[1].name);
	    failed = 1;
	  }

#if HAVE_SETMODE
      if (binary_I_O)
	for (i = 0; i <= 1; i++)
	  if (0 <= inf[i].desc)
	    setmode (inf[i].desc, O_BINARY);
#endif

      /* Compare the files, if no error was found.  */

      val = failed ? 2 : diff_2_files (inf, depth);

      /* Close the file descriptors.  */

      if (inf[0].desc >= 0 && close (inf[0].desc) != 0)
	{
	  perror_with_name (inf[0].name);
	  val = 2;
	}
      if (inf[1].desc >= 0 && inf[0].desc != inf[1].desc
	  && close (inf[1].desc) != 0)
	{
	  perror_with_name (inf[1].name);
	  val = 2;
	}
    }

  /* Now the comparison has been done, if no error prevented it,
     and VAL is the value this function will return.  */

  if (val == 0 && !inf[0].dir_p)
    {
      if (print_file_same_flag)
	message ("Files %s and %s are identical\n",
		 inf[0].name, inf[1].name);
    }
  else
    fflush (stdout);

  if (free0)
    free (free0);
  if (free1)
    free (free1);

  return val;
}