/* File I/O for GNU DIFF.

   Copyright (C) 1988-1989, 1992-1995, 1998, 2001-2002, 2004, 2006, 2009-2013,

   2015-2017 Free Software Foundation, Inc.

   This file is part of GNU DIFF.

   This program 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 3 of the License, or

   (at your option) any later version.

   This program 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 this program.  If not, see <http://www.gnu.org/licenses/>.  */

#include "diff.h"

#include <binary-io.h>

#include <cmpbuf.h>

#include <file-type.h>

#include <xalloc.h>

#include <assert.h>

/* Rotate an unsigned value to the left.  */

#define ROL(v, n) ((v) << (n) | (v) >> (sizeof (v) * CHAR_BIT - (n)))

/* Given a hash value and a new character, return a new hash value.  */

#define HASH(h, c) ((c) + ROL (h, 7))

/* The type of a hash value.  */

typedef size_t hash_value;

verify (! TYPE_SIGNED (hash_value));

/* Lines are put into equivalence classes of lines that match in lines_differ.

   Each equivalence class is represented by one of these structures,

   but only while the classes are being computed.

   Afterward, each class is represented by a number.  */

struct equivclass

{

  lin next;		/* Next item in this bucket.  */

  hash_value hash;	/* Hash of lines in this class.  */

  char const *line;	/* A line that fits this class.  */

  size_t length;	/* That line's length, not counting its newline.  */

};

/* Hash-table: array of buckets, each being a chain of equivalence classes.

   buckets[-1] is reserved for incomplete lines.  */

static lin *buckets;

/* Number of buckets in the hash table array, not counting buckets[-1].  */

static size_t nbuckets;

/* Array in which the equivalence classes are allocated.

   The bucket-chains go through the elements in this array.

   The number of an equivalence class is its index in this array.  */

static struct equivclass *equivs;

/* Index of first free element in the array 'equivs'.  */

static lin equivs_index;

/* Number of elements allocated in the array 'equivs'.  */

static lin equivs_alloc;

/* Read a block of data into a file buffer, checking for EOF and error.  */

void

file_block_read (struct file_data *current, size_t size)

{

  if (size && ! current->eof)

    {

      size_t s = block_read (current->desc,

			     FILE_BUFFER (current) + current->buffered, size);

      if (s == SIZE_MAX)

	pfatal_with_name (current->name);

      current->buffered += s;

      current->eof = s < size;

    }

}

/* Check for binary files and compare them for exact identity.  */

/* Return 1 if BUF contains a non text character.

   SIZE is the number of characters in BUF.  */

#define binary_file_p(buf, size) (memchr (buf, 0, size) != 0)

/* Get ready to read the current file.

   Return nonzero if SKIP_TEST is zero,

   and if it appears to be a binary file.  */

static bool

sip (struct file_data *current, bool skip_test)

{

  /* If we have a nonexistent file at this stage, treat it as empty.  */

  if (current->desc < 0)

    {

      /* Leave room for a sentinel.  */

      current->bufsize = sizeof (word);

      current->buffer = xmalloc (current->bufsize);

    }

  else

    {

      current->bufsize = buffer_lcm (sizeof (word),

				     STAT_BLOCKSIZE (current->stat),

				     PTRDIFF_MAX - 2 * sizeof (word));

      current->buffer = xmalloc (current->bufsize);

#ifdef __KLIBC__

      /* Skip test if seek is not possible */

      skip_test = skip_test

		  || (lseek (current->desc, 0, SEEK_CUR) < 0

		      && errno == ESPIPE);

#endif

      if (! skip_test)

	{

	  /* Check first part of file to see if it's a binary file.  */

	  int prev_mode = set_binary_mode (current->desc, O_BINARY);

	  off_t buffered;

	  file_block_read (current, current->bufsize);

	  buffered = current->buffered;

	  if (prev_mode != O_BINARY)

	    {

	      /* Revert to text mode and seek back to the start to reread

		 the file.  Use relative seek, since file descriptors

		 like stdin might not start at offset zero.  */

	      if (lseek (current->desc, - buffered, SEEK_CUR) < 0)

		pfatal_with_name (current->name);

	      set_binary_mode (current->desc, prev_mode);

	      current->buffered = 0;

	      current->eof = false;

	    }

	  return binary_file_p (current->buffer, buffered);

	}

    }

  current->buffered = 0;

  current->eof = false;

  return false;

}

/* Slurp the rest of the current file completely into memory.  */

static void

slurp (struct file_data *current)

{

  size_t cc;

  if (current->desc < 0)

    {

      /* The file is nonexistent.  */

      return;

    }

  if (S_ISREG (current->stat.st_mode))

    {

      /* It's a regular file; slurp in the rest all at once.  */

      /* Get the size out of the stat block.

	 Allocate just enough room for appended newline plus word sentinel,

	 plus word-alignment since we want the buffer word-aligned.  */

      size_t file_size = current->stat.st_size;

      cc = file_size + 2 * sizeof (word) - file_size % sizeof (word);

      if (file_size != current->stat.st_size || cc < file_size

	  || PTRDIFF_MAX <= cc)

	xalloc_die ();

      if (current->bufsize < cc)

	{

	  current->bufsize = cc;

	  current->buffer = xrealloc (current->buffer, cc);

	}

      /* Try to read at least 1 more byte than the size indicates, to

	 detect whether the file is growing.  This is a nicety for

	 users who run 'diff' on files while they are changing.  */

      if (current->buffered <= file_size)

	{

	  file_block_read (current, file_size + 1 - current->buffered);

	  if (current->buffered <= file_size)

	    return;

	}

    }

  /* It's not a regular file, or it's a growing regular file; read it,

     growing the buffer as needed.  */

  file_block_read (current, current->bufsize - current->buffered);

  if (current->buffered)

    {

      while (current->buffered == current->bufsize)

	{

	  if (PTRDIFF_MAX / 2 - sizeof (word) < current->bufsize)

	    xalloc_die ();

	  current->bufsize *= 2;

	  current->buffer = xrealloc (current->buffer, current->bufsize);

	  file_block_read (current, current->bufsize - current->buffered);

	}

      /* Allocate just enough room for appended newline plus word

	 sentinel, plus word-alignment.  */

      cc = current->buffered + 2 * sizeof (word);

      current->bufsize = cc - cc % sizeof (word);

      current->buffer = xrealloc (current->buffer, current->bufsize);

    }

}

/* Split the file into lines, simultaneously computing the equivalence

   class for each line.  */

#ifdef HANDLE_MULTIBYTE

# define MBC2WC(P, END, MBLENGTH, WC, STATE, CONVFAIL)		\

do								\

  {								\

    mbstate_t state_bak = STATE;				\

								\

    CONVFAIL = 0;						\

    MBLENGTH = mbrtowc (&WC, P, END - (char const *)P, &STATE);	\

								\

    switch (MBLENGTH)						\

      {								\

      case (size_t)-2:						\

      case (size_t)-1:						\

	STATE = state_bak;					\

	++CONVFAIL;						\

	/* Fall through. */					\

      case 0:							\

	MBLENGTH = 1;						\

      }								\

  }								\

 while (0)

#endif

static void

find_and_hash_each_line (struct file_data *current)

{

  char const *p = current->prefix_end;

  lin i, *bucket;

  size_t length;

  /* Cache often-used quantities in local variables to help the compiler.  */

  char const **linbuf = current->linbuf;

  lin alloc_lines = current->alloc_lines;

  lin line = 0;

  lin linbuf_base = current->linbuf_base;

  lin *cureqs = xmalloc (alloc_lines * sizeof *cureqs);

  struct equivclass *eqs = equivs;

  lin eqs_index = equivs_index;

  lin eqs_alloc = equivs_alloc;

  char const *suffix_begin = current->suffix_begin;

  char const *bufend = FILE_BUFFER (current) + current->buffered;

  bool ig_case = ignore_case;

  enum DIFF_white_space ig_white_space = ignore_white_space;

  bool diff_length_compare_anyway =

    ig_white_space != IGNORE_NO_WHITE_SPACE;

  bool same_length_diff_contents_compare_anyway =

    diff_length_compare_anyway | ig_case;

#ifdef HANDLE_MULTIBYTE

  wchar_t wc;

  size_t mblength;

  mbstate_t state;

  int convfail;

  memset (&state, '\0', sizeof (mbstate_t));

#endif

  while (p < suffix_begin)

    {

      char const *ip = p;

      hash_value h = 0;

      unsigned char c;

#ifdef HANDLE_MULTIBYTE

      if (MB_CUR_MAX > 1)

	{

	  wchar_t lo_wc;

	  char mbc[MB_LEN_MAX];

	  mbstate_t state_wc;

	  /* Hash this line until we find a newline.  */

	  switch (ig_white_space)

	    {

	    case IGNORE_ALL_SPACE:

	      while (1)

		{

		  if (*p == '\n')

		    {

		      ++p;

		      break;

		    }

		  MBC2WC (p, suffix_begin, mblength, wc, state, convfail);

		  if (convfail)

		    mbc[0] = *p++;

		  else if (!iswspace (wc))

		    {

		      bool flag = 0;

		      if (ig_case)

			{

			  lo_wc = towlower (wc);

			  if (lo_wc != wc)

			    {

			      flag = 1;

			      p += mblength;

			      memset (&state_wc, '\0', sizeof(mbstate_t));

			      mblength = wcrtomb (mbc, lo_wc, &state_wc);

			      assert (mblength != (size_t)-1 &&

				      mblength != (size_t)-2);

			      mblength = (mblength < 1) ? 1 : mblength;

			    }

			}

		      if (!flag)

			{

			  for (i = 0; i < mblength; i++)

			    mbc[i] = *p++;

			}

		    }

		  else

		    {

		      p += mblength;

		      continue;

		    }

		  for (i = 0; i < mblength; i++)

		    h = HASH (h, mbc[i]);

		}

	      break;

	    case IGNORE_SPACE_CHANGE:

	      while (1)

		{

		  if (*p == '\n')

		    {

		      ++p;

		      break;

		    }

		  MBC2WC (p, suffix_begin, mblength, wc, state, convfail);

		  if (!convfail && iswspace (wc))

		    {

		      while (1)

			{

			  if (*p == '\n')

			    {

			      ++p;

			      goto hashing_done;

			    }

			  p += mblength;

			  MBC2WC (p, suffix_begin, mblength, wc, state, convfail);

			  if (convfail || !iswspace (wc))

			    break;

			}

		      h = HASH (h, ' ');

		    }

		  /* WC is now the first non-space.  */

		  if (convfail)

		    mbc[0] = *p++;

		  else

		    {

		      bool flag = 0;

		      if (ignore_case)

			{

			  lo_wc = towlower (wc);

			  if (lo_wc != wc)

			    {

			      flag = 1;

			      p += mblength;

			      memset (&state_wc, '\0', sizeof(mbstate_t));

			      mblength = wcrtomb (mbc, lo_wc, &state_wc);

			      assert (mblength != (size_t)-1 &&

				      mblength != (size_t)-2);

			      mblength = (mblength < 1) ? 1 : mblength;

			    }

			}

		      if (!flag)

			{

			  for (i = 0; i < mblength; i++)

			    mbc[i] = *p++;

			}

		    }

		  for (i = 0; i < mblength; i++)

		    h = HASH (h, mbc[i]);

		}

	      break;

	    case IGNORE_TAB_EXPANSION:

	    case IGNORE_TAB_EXPANSION_AND_TRAILING_SPACE:

	    case IGNORE_TRAILING_SPACE:

	      {

		size_t column = 0;

		while (1)

		  {

		    if (*p == '\n')

		      {

			++p;

			break;

		      }

		    MBC2WC (p, suffix_begin, mblength, wc, state, convfail);

		    if (!convfail

			&& ig_white_space & IGNORE_TRAILING_SPACE

			&& iswspace (wc))

		      {

			char const *p1 = p;

			while (1)

			  {

			    if (*p1 == '\n')

			      {

				p = p1 + 1;

				goto hashing_done;

			      }

			    p1 += mblength;

			    MBC2WC (p1, suffix_begin, mblength, wc, state, convfail);

			    if (convfail || !iswspace (wc))

			      break;

			  }

		      }

		    size_t repetitions = 1;

		    bool no_convert = 0;

		    if (ig_white_space & IGNORE_TAB_EXPANSION)

		      {

			if (convfail)

			  column++;

			else

			  switch (wc)

			    {

			    case L'\b':

			      column -= 0 < column;

			      break;

			    case L'\t':

			      mbc[0] = ' ';

			      mblength = 1;

			      no_convert = 1;

			      p++;

			      assert(mblength == 1);

			      repetitions = tabsize - column % tabsize;

			      column = (column + repetitions < column

					? 0

					: column + repetitions);

			      break;

			    case L'\r':

			      column = 0;

			      break;

			    default:

			      column += wcwidth (wc);

			      break;

			    }

		      }

		    if (ig_case)

		      {

			lo_wc = towlower (wc);

			if (lo_wc != wc)

			  {

			    no_convert = 1;

			    p += mblength;

			    memset (&state_wc, '\0', sizeof(mbstate_t));

			    mblength = wcrtomb (mbc, lo_wc, &state_wc);

			    assert (mblength != (size_t)-1 &&

				    mblength != (size_t)-2);

			    mblength = (mblength < 1) ? 1 : mblength;

			  }

		      }

		    if (!no_convert)

		      for (i = 0; i < mblength; i++)

			mbc[i] = *p++;

		    do

		      {

			for (i = 0; i < mblength; i++)

			  h = HASH (h, mbc[i]);

		      }

		    while (--repetitions != 0);

		  }

	      }

	      break;

	    default:

	      while (1)

		{

		  if (*p == '\n')

		    {

		      ++p;

		      break;

		    }

		  MBC2WC (p, suffix_begin, mblength, wc, state, convfail);

		  if (convfail)

		    mbc[0] = *p++;

		  else

		    {

		      int flag = 0;

		      if (ig_case)

			{

			  lo_wc = towlower (wc);

			  if (lo_wc != wc)

			    {

			      flag = 1;

			      p += mblength;

			      memset (&state_wc, '\0', sizeof(mbstate_t));

			      mblength = wcrtomb (mbc, lo_wc, &state_wc);

			      assert (mblength != (size_t)-1 &&

				      mblength != (size_t)-2);

			      mblength = (mblength < 1) ? 1 : mblength;

			    }

			}

		      if (!flag)

			{

			  for (i = 0; i < mblength; i++)

			    mbc[i] = *p++;

			}

		    }

		  for (i = 0; i < mblength; i++)

		    h = HASH (h, mbc[i]);

		}

	    }

	}

      else

#endif

      /* Hash this line until we find a newline.  */

      switch (ig_white_space)

	{

	case IGNORE_ALL_SPACE:

	  while ((c = *p++) != '\n')

	    if (! isspace (c))

	      h = HASH (h, ig_case ? tolower (c) : c);

	  break;

	case IGNORE_SPACE_CHANGE:

	  while ((c = *p++) != '\n')

	    {

	      if (isspace (c))

		{

		  do

		    if ((c = *p++) == '\n')

		      goto hashing_done;

		  while (isspace (c));

		  h = HASH (h, ' ');

		}

	      /* C is now the first non-space.  */

	      h = HASH (h, ig_case ? tolower (c) : c);

	    }

	  break;

	case IGNORE_TAB_EXPANSION:

	case IGNORE_TAB_EXPANSION_AND_TRAILING_SPACE:

	case IGNORE_TRAILING_SPACE:

	  {

	    size_t column = 0;

	    while ((c = *p++) != '\n')

	      {

		if (ig_white_space & IGNORE_TRAILING_SPACE

		    && isspace (c))

		  {

		    char const *p1 = p;

		    unsigned char c1;

		    do

		      if ((c1 = *p1++) == '\n')

			{

			  p = p1;

			  goto hashing_done;

			}

		    while (isspace (c1));

		  }

		size_t repetitions = 1;

		if (ig_white_space & IGNORE_TAB_EXPANSION)

		  switch (c)

		    {

		    case '\b':

		      column -= 0 < column;

		      break;

		    case '\t':

		      c = ' ';

		      repetitions = tabsize - column % tabsize;

		      column = (column + repetitions < column

				? 0

				: column + repetitions);

		      break;

		    case '\r':

		      column = 0;

		      break;

		    default:

		      column++;

		      break;

		    }

		if (ig_case)

		  c = tolower (c);

		do

		  h = HASH (h, c);

		while (--repetitions != 0);

	      }

	  }

	  break;

	default:

	  if (ig_case)

	    while ((c = *p++) != '\n')

	      h = HASH (h, tolower (c));

	  else

	    while ((c = *p++) != '\n')

	      h = HASH (h, c);

	  break;

	}

   hashing_done:;

      bucket = &buckets[h % nbuckets];

      length = p - ip - 1;

      if (p == bufend

	  && current->missing_newline

	  && ROBUST_OUTPUT_STYLE (output_style))

	{

	  /* The last line is incomplete and we do not silently

	     complete lines.  If the line cannot compare equal to any

	     complete line, put it into buckets[-1] so that it can

	     compare equal only to the other file's incomplete line

	     (if one exists).  */

	  if (ig_white_space < IGNORE_TRAILING_SPACE)

	    bucket = &buckets[-1];

	}

      for (i = *bucket;  ;  i = eqs[i].next)

	if (!i)

	  {

	    /* Create a new equivalence class in this bucket.  */

	    i = eqs_index++;

	    if (i == eqs_alloc)

	      {

		if (PTRDIFF_MAX / (2 * sizeof *eqs) <= eqs_alloc)

		  xalloc_die ();

		eqs_alloc *= 2;

		eqs = xrealloc (eqs, eqs_alloc * sizeof *eqs);

	      }

	    eqs[i].next = *bucket;

	    eqs[i].hash = h;

	    eqs[i].line = ip;

	    eqs[i].length = length;

	    *bucket = i;

	    break;

	  }

	else if (eqs[i].hash == h)

	  {

	    char const *eqline = eqs[i].line;

	    /* Reuse existing class if lines_differ reports the lines

               equal.  */

	    if (eqs[i].length == length)

	      {

		/* Reuse existing equivalence class if the lines are identical.

		   This detects the common case of exact identity

		   faster than lines_differ would.  */

		if (memcmp (eqline, ip, length) == 0)

		  break;

		if (!same_length_diff_contents_compare_anyway)

		  continue;

	      }

	    else if (!diff_length_compare_anyway)

	      continue;

	    if (! lines_differ (eqline, eqs[i].length + 1, ip, length + 1))

	      break;

	  }

      /* Maybe increase the size of the line table.  */

      if (line == alloc_lines)

	{

	  /* Double (alloc_lines - linbuf_base) by adding to alloc_lines.  */

	  if (PTRDIFF_MAX / 3 <= alloc_lines

	      || PTRDIFF_MAX / sizeof *cureqs <= 2 * alloc_lines - linbuf_base

	      || PTRDIFF_MAX / sizeof *linbuf <= alloc_lines - linbuf_base)

	    xalloc_die ();

	  alloc_lines = 2 * alloc_lines - linbuf_base;

	  cureqs = xrealloc (cureqs, alloc_lines * sizeof *cureqs);

	  linbuf += linbuf_base;

	  linbuf = xrealloc (linbuf,

			     (alloc_lines - linbuf_base) * sizeof *linbuf);

	  linbuf -= linbuf_base;

	}

      linbuf[line] = ip;

      cureqs[line] = i;

      ++line;

    }

  current->buffered_lines = line;

  for (i = 0;  ;  i++)

    {

      /* Record the line start for lines in the suffix that we care about.

	 Record one more line start than lines,

	 so that we can compute the length of any buffered line.  */

      if (line == alloc_lines)

	{

	  /* Double (alloc_lines - linbuf_base) by adding to alloc_lines.  */

	  if (PTRDIFF_MAX / 3 <= alloc_lines

	      || PTRDIFF_MAX / sizeof *cureqs <= 2 * alloc_lines - linbuf_base

	      || PTRDIFF_MAX / sizeof *linbuf <= alloc_lines - linbuf_base)

	    xalloc_die ();

	  alloc_lines = 2 * alloc_lines - linbuf_base;

	  linbuf += linbuf_base;

	  linbuf = xrealloc (linbuf,

			     (alloc_lines - linbuf_base) * sizeof *linbuf);

	  linbuf -= linbuf_base;

	}

      linbuf[line] = p;

      if (p == bufend)

	{

	  /* If the last line is incomplete and we do not silently

	     complete lines, don't count its appended newline.  */

	  if (current->missing_newline && ROBUST_OUTPUT_STYLE (output_style))

	    linbuf[line]--;

	  break;

	}

      if (context <= i && no_diff_means_no_output)

	break;

      line++;

      while (*p++ != '\n')

	continue;

    }

  /* Done with cache in local variables.  */

  current->linbuf = linbuf;

  current->valid_lines = line;

  current->alloc_lines = alloc_lines;

  current->equivs = cureqs;

  equivs = eqs;

  equivs_alloc = eqs_alloc;

  equivs_index = eqs_index;

}

/* Prepare the text.  Make sure the text end is initialized.

   Make sure text ends in a newline,

   but remember that we had to add one.

   Strip trailing CRs, if that was requested.  */

static void

prepare_text (struct file_data *current)

{

  size_t buffered = current->buffered;

  char *p = FILE_BUFFER (current);

  if (buffered == 0 || p[buffered - 1] == '\n')

    current->missing_newline = false;

  else

    {

      p[buffered++] = '\n';

      current->missing_newline = true;

    }

  if (!p)

    return;

  /* Don't use uninitialized storage when planting or using sentinels.  */

  memset (p + buffered, 0, sizeof (word));

  if (strip_trailing_cr)

    {

      char *dst;

      char *srclim = p + buffered;

      *srclim = '\r';

      dst = rawmemchr (p, '\r');

      if (dst != srclim)

	{

	  char const *src = dst;

	  do

	    {

	      *dst = *src++;

	      dst += ! (*dst == '\r' && *src == '\n');

	    }

	  while (src < srclim);

	  buffered -= src - dst;

	}

    }

  current->buffered = buffered;

}

/* We have found N lines in a buffer of size S; guess the

   proportionate number of lines that will be found in a buffer of

   size T.  However, do not guess a number of lines so large that the

   resulting line table might cause overflow in size calculations.  */

static lin

guess_lines (lin n, size_t s, size_t t)

{

  size_t guessed_bytes_per_line = n < 10 ? 32 : s / (n - 1);

  lin guessed_lines = MAX (1, t / guessed_bytes_per_line);

  return MIN (guessed_lines, PTRDIFF_MAX / (2 * sizeof (char *) + 1) - 5) + 5;

}

/* Given a vector of two file_data objects, find the identical

   prefixes and suffixes of each object.  */

static void

find_identical_ends (struct file_data filevec[])

{

  word *w0, *w1;

  char *p0, *p1, *buffer0, *buffer1;

  char const *end0, *beg0;

  char const **linbuf0, **linbuf1;

  lin i, lines;

  size_t n0, n1;

  lin alloc_lines0, alloc_lines1;

  bool prefix_needed;

  lin buffered_prefix, prefix_count, prefix_mask;

  lin middle_guess, suffix_guess;

  slurp (&filevec[0]);

  prepare_text (&filevec[0]);

  if (filevec[0].desc != filevec[1].desc)

    {

      slurp (&filevec[1]);

      prepare_text (&filevec[1]);

    }

  else

    {

      filevec[1].buffer = filevec[0].buffer;

      filevec[1].bufsize = filevec[0].bufsize;

      filevec[1].buffered = filevec[0].buffered;

      filevec[1].missing_newline = filevec[0].missing_newline;

    }

  /* Find identical prefix.  */

  w0 = filevec[0].buffer;