688 lines
22 KiB
PHP
688 lines
22 KiB
PHP
<?php
|
|
/**
|
|
* FINE granularity DIFF
|
|
*
|
|
* Computes a set of instructions to convert the content of
|
|
* one string into another.
|
|
*
|
|
* Copyright (c) 2011 Raymond Hill (http://raymondhill.net/blog/?p=441)
|
|
*
|
|
* Licensed under The MIT License
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in
|
|
* all copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
* THE SOFTWARE.
|
|
*
|
|
* @copyright Copyright 2011 (c) Raymond Hill (http://raymondhill.net/blog/?p=441)
|
|
* @link http://www.raymondhill.net/finediff/
|
|
* @version 0.6
|
|
* @license MIT License (http://www.opensource.org/licenses/mit-license.php)
|
|
*/
|
|
|
|
/**
|
|
* Usage (simplest):
|
|
*
|
|
* include 'finediff.php';
|
|
*
|
|
* // for the stock stack, granularity values are:
|
|
* // FineDiff::$paragraphGranularity = paragraph/line level
|
|
* // FineDiff::$sentenceGranularity = sentence level
|
|
* // FineDiff::$wordGranularity = word level
|
|
* // FineDiff::$characterGranularity = character level [default]
|
|
*
|
|
* $opcodes = FineDiff::getDiffOpcodes($from_text, $to_text [, $granularityStack = null] );
|
|
* // store opcodes for later use...
|
|
*
|
|
* ...
|
|
*
|
|
* // restore $to_text from $from_text + $opcodes
|
|
* include 'finediff.php';
|
|
* $to_text = FineDiff::renderToTextFromOpcodes($from_text, $opcodes);
|
|
*
|
|
* ...
|
|
*/
|
|
|
|
/**
|
|
* Persisted opcodes (string) are a sequence of atomic opcode.
|
|
* A single opcode can be one of the following:
|
|
* c | c{n} | d | d{n} | i:{c} | i{length}:{s}
|
|
* 'c' = copy one character from source
|
|
* 'c{n}' = copy n characters from source
|
|
* 'd' = skip one character from source
|
|
* 'd{n}' = skip n characters from source
|
|
* 'i:{c} = insert character 'c'
|
|
* 'i{n}:{s}' = insert string s, which is of length n
|
|
*
|
|
* Do not exist as of now, under consideration:
|
|
* 'm{n}:{o} = move n characters from source o characters ahead.
|
|
* It would be essentially a shortcut for a delete->copy->insert
|
|
* command (swap) for when the inserted segment is exactly the same
|
|
* as the deleted one, and with only a copy operation in between.
|
|
* TODO: How often this case occurs? Is it worth it? Can only
|
|
* be done as a postprocessing method (->optimize()?)
|
|
*/
|
|
abstract class FineDiffOp {
|
|
abstract public function getFromLen();
|
|
abstract public function getToLen();
|
|
abstract public function getOpcode();
|
|
}
|
|
|
|
class FineDiffDeleteOp extends FineDiffOp {
|
|
public function __construct($len) {
|
|
$this->fromLen = $len;
|
|
}
|
|
public function getFromLen() {
|
|
return $this->fromLen;
|
|
}
|
|
public function getToLen() {
|
|
return 0;
|
|
}
|
|
public function getOpcode() {
|
|
if ( $this->fromLen === 1 ) {
|
|
return 'd';
|
|
}
|
|
return "d{$this->fromLen}";
|
|
}
|
|
}
|
|
|
|
class FineDiffInsertOp extends FineDiffOp {
|
|
public function __construct($text) {
|
|
$this->text = $text;
|
|
}
|
|
public function getFromLen() {
|
|
return 0;
|
|
}
|
|
public function getToLen() {
|
|
return strlen($this->text);
|
|
}
|
|
public function getText() {
|
|
return $this->text;
|
|
}
|
|
public function getOpcode() {
|
|
$to_len = strlen($this->text);
|
|
if ( $to_len === 1 ) {
|
|
return "i:{$this->text}";
|
|
}
|
|
return "i{$to_len}:{$this->text}";
|
|
}
|
|
}
|
|
|
|
class FineDiffReplaceOp extends FineDiffOp {
|
|
public function __construct($fromLen, $text) {
|
|
$this->fromLen = $fromLen;
|
|
$this->text = $text;
|
|
}
|
|
public function getFromLen() {
|
|
return $this->fromLen;
|
|
}
|
|
public function getToLen() {
|
|
return strlen($this->text);
|
|
}
|
|
public function getText() {
|
|
return $this->text;
|
|
}
|
|
public function getOpcode() {
|
|
if ( $this->fromLen === 1 ) {
|
|
$del_opcode = 'd';
|
|
}
|
|
else {
|
|
$del_opcode = "d{$this->fromLen}";
|
|
}
|
|
$to_len = strlen($this->text);
|
|
if ( $to_len === 1 ) {
|
|
return "{$del_opcode}i:{$this->text}";
|
|
}
|
|
return "{$del_opcode}i{$to_len}:{$this->text}";
|
|
}
|
|
}
|
|
|
|
class FineDiffCopyOp extends FineDiffOp {
|
|
public function __construct($len) {
|
|
$this->len = $len;
|
|
}
|
|
public function getFromLen() {
|
|
return $this->len;
|
|
}
|
|
public function getToLen() {
|
|
return $this->len;
|
|
}
|
|
public function getOpcode() {
|
|
if ( $this->len === 1 ) {
|
|
return 'c';
|
|
}
|
|
return "c{$this->len}";
|
|
}
|
|
public function increase($size) {
|
|
return $this->len += $size;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* FineDiff ops
|
|
*
|
|
* Collection of ops
|
|
*/
|
|
class FineDiffOps {
|
|
public function appendOpcode($opcode, $from, $from_offset, $from_len) {
|
|
if ( $opcode === 'c' ) {
|
|
$edits[] = new FineDiffCopyOp($from_len);
|
|
}
|
|
else if ( $opcode === 'd' ) {
|
|
$edits[] = new FineDiffDeleteOp($from_len);
|
|
}
|
|
else /* if ( $opcode === 'i' ) */ {
|
|
$edits[] = new FineDiffInsertOp(substr($from, $from_offset, $from_len));
|
|
}
|
|
}
|
|
public $edits = array();
|
|
}
|
|
|
|
/**
|
|
* FineDiff class
|
|
*
|
|
* TODO: Document
|
|
*
|
|
*/
|
|
class FineDiff {
|
|
|
|
/**------------------------------------------------------------------------
|
|
*
|
|
* Public section
|
|
*
|
|
*/
|
|
|
|
/**
|
|
* Constructor
|
|
* ...
|
|
* The $granularityStack allows FineDiff to be configurable so that
|
|
* a particular stack tailored to the specific content of a document can
|
|
* be passed.
|
|
*/
|
|
public function __construct($from_text = '', $to_text = '', $granularityStack = null) {
|
|
// setup stack for generic text documents by default
|
|
$this->granularityStack = $granularityStack ? $granularityStack : FineDiff::$characterGranularity;
|
|
$this->edits = array();
|
|
$this->from_text = $from_text;
|
|
$this->doDiff($from_text, $to_text);
|
|
}
|
|
|
|
public function getOps() {
|
|
return $this->edits;
|
|
}
|
|
|
|
public function getOpcodes() {
|
|
$opcodes = array();
|
|
foreach ( $this->edits as $edit ) {
|
|
$opcodes[] = $edit->getOpcode();
|
|
}
|
|
return implode('', $opcodes);
|
|
}
|
|
|
|
public function renderDiffToHTML() {
|
|
$in_offset = 0;
|
|
ob_start();
|
|
foreach ( $this->edits as $edit ) {
|
|
$n = $edit->getFromLen();
|
|
if ( $edit instanceof FineDiffCopyOp ) {
|
|
FineDiff::renderDiffToHTMLFromOpcode('c', $this->from_text, $in_offset, $n);
|
|
}
|
|
else if ( $edit instanceof FineDiffDeleteOp ) {
|
|
FineDiff::renderDiffToHTMLFromOpcode('d', $this->from_text, $in_offset, $n);
|
|
}
|
|
else if ( $edit instanceof FineDiffInsertOp ) {
|
|
FineDiff::renderDiffToHTMLFromOpcode('i', $edit->getText(), 0, $edit->getToLen());
|
|
}
|
|
else /* if ( $edit instanceof FineDiffReplaceOp ) */ {
|
|
FineDiff::renderDiffToHTMLFromOpcode('d', $this->from_text, $in_offset, $n);
|
|
FineDiff::renderDiffToHTMLFromOpcode('i', $edit->getText(), 0, $edit->getToLen());
|
|
}
|
|
$in_offset += $n;
|
|
}
|
|
return ob_get_clean();
|
|
}
|
|
|
|
/**------------------------------------------------------------------------
|
|
* Return an opcodes string describing the diff between a "From" and a
|
|
* "To" string
|
|
*/
|
|
public static function getDiffOpcodes($from, $to, $granularities = null) {
|
|
$diff = new FineDiff($from, $to, $granularities);
|
|
return $diff->getOpcodes();
|
|
}
|
|
|
|
/**------------------------------------------------------------------------
|
|
* Return an iterable collection of diff ops from an opcodes string
|
|
*/
|
|
public static function getDiffOpsFromOpcodes($opcodes) {
|
|
$diffops = new FineDiffOps();
|
|
FineDiff::renderFromOpcodes(null, $opcodes, array($diffops,'appendOpcode'));
|
|
return $diffops->edits;
|
|
}
|
|
|
|
/**------------------------------------------------------------------------
|
|
* Re-create the "To" string from the "From" string and an "Opcodes" string
|
|
*/
|
|
public static function renderToTextFromOpcodes($from, $opcodes) {
|
|
ob_start();
|
|
FineDiff::renderFromOpcodes($from, $opcodes, array('FineDiff','renderToTextFromOpcode'));
|
|
return ob_get_clean();
|
|
}
|
|
|
|
/**------------------------------------------------------------------------
|
|
* Render the diff to an HTML string
|
|
*/
|
|
public static function renderDiffToHTMLFromOpcodes($from, $opcodes) {
|
|
ob_start();
|
|
FineDiff::renderFromOpcodes($from, $opcodes, array('FineDiff','renderDiffToHTMLFromOpcode'));
|
|
return ob_get_clean();
|
|
}
|
|
|
|
/**------------------------------------------------------------------------
|
|
* Generic opcodes parser, user must supply callback for handling
|
|
* single opcode
|
|
*/
|
|
public static function renderFromOpcodes($from, $opcodes, $callback) {
|
|
if ( !is_callable($callback) ) {
|
|
return;
|
|
}
|
|
$opcodes_len = strlen($opcodes);
|
|
$from_offset = $opcodes_offset = 0;
|
|
while ( $opcodes_offset < $opcodes_len ) {
|
|
$opcode = substr($opcodes, $opcodes_offset, 1);
|
|
$opcodes_offset++;
|
|
$n = intval(substr($opcodes, $opcodes_offset));
|
|
if ( $n ) {
|
|
$opcodes_offset += strlen(strval($n));
|
|
}
|
|
else {
|
|
$n = 1;
|
|
}
|
|
if ( $opcode === 'c' ) { // copy n characters from source
|
|
call_user_func($callback, 'c', $from, $from_offset, $n, '');
|
|
$from_offset += $n;
|
|
}
|
|
else if ( $opcode === 'd' ) { // delete n characters from source
|
|
call_user_func($callback, 'd', $from, $from_offset, $n, '');
|
|
$from_offset += $n;
|
|
}
|
|
else /* if ( $opcode === 'i' ) */ { // insert n characters from opcodes
|
|
call_user_func($callback, 'i', $opcodes, $opcodes_offset + 1, $n);
|
|
$opcodes_offset += 1 + $n;
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Stock granularity stacks and delimiters
|
|
*/
|
|
|
|
const paragraphDelimiters = "\n\r";
|
|
public static $paragraphGranularity = array(
|
|
FineDiff::paragraphDelimiters
|
|
);
|
|
const sentenceDelimiters = ".\n\r";
|
|
public static $sentenceGranularity = array(
|
|
FineDiff::paragraphDelimiters,
|
|
FineDiff::sentenceDelimiters
|
|
);
|
|
const wordDelimiters = " \t.\n\r";
|
|
public static $wordGranularity = array(
|
|
FineDiff::paragraphDelimiters,
|
|
FineDiff::sentenceDelimiters,
|
|
FineDiff::wordDelimiters
|
|
);
|
|
const characterDelimiters = "";
|
|
public static $characterGranularity = array(
|
|
FineDiff::paragraphDelimiters,
|
|
FineDiff::sentenceDelimiters,
|
|
FineDiff::wordDelimiters,
|
|
FineDiff::characterDelimiters
|
|
);
|
|
|
|
public static $textStack = array(
|
|
".",
|
|
" \t.\n\r",
|
|
""
|
|
);
|
|
|
|
/**------------------------------------------------------------------------
|
|
*
|
|
* Private section
|
|
*
|
|
*/
|
|
|
|
/**
|
|
* Entry point to compute the diff.
|
|
*/
|
|
private function doDiff($from_text, $to_text) {
|
|
$this->last_edit = false;
|
|
$this->stackpointer = 0;
|
|
$this->from_text = $from_text;
|
|
$this->from_offset = 0;
|
|
// can't diff without at least one granularity specifier
|
|
if ( empty($this->granularityStack) ) {
|
|
return;
|
|
}
|
|
$this->_processGranularity($from_text, $to_text);
|
|
}
|
|
|
|
/**
|
|
* This is the recursive function which is responsible for
|
|
* handling/increasing granularity.
|
|
*
|
|
* Incrementally increasing the granularity is key to compute the
|
|
* overall diff in a very efficient way.
|
|
*/
|
|
private function _processGranularity($from_segment, $to_segment) {
|
|
$delimiters = $this->granularityStack[$this->stackpointer++];
|
|
$has_next_stage = $this->stackpointer < count($this->granularityStack);
|
|
foreach ( FineDiff::doFragmentDiff($from_segment, $to_segment, $delimiters) as $fragment_edit ) {
|
|
// increase granularity
|
|
if ( $fragment_edit instanceof FineDiffReplaceOp && $has_next_stage ) {
|
|
$this->_processGranularity(
|
|
substr($this->from_text, $this->from_offset, $fragment_edit->getFromLen()),
|
|
$fragment_edit->getText()
|
|
);
|
|
}
|
|
// fuse copy ops whenever possible
|
|
else if ( $fragment_edit instanceof FineDiffCopyOp && $this->last_edit instanceof FineDiffCopyOp ) {
|
|
$this->edits[count($this->edits)-1]->increase($fragment_edit->getFromLen());
|
|
$this->from_offset += $fragment_edit->getFromLen();
|
|
}
|
|
else {
|
|
/* $fragment_edit instanceof FineDiffCopyOp */
|
|
/* $fragment_edit instanceof FineDiffDeleteOp */
|
|
/* $fragment_edit instanceof FineDiffInsertOp */
|
|
$this->edits[] = $this->last_edit = $fragment_edit;
|
|
$this->from_offset += $fragment_edit->getFromLen();
|
|
}
|
|
}
|
|
$this->stackpointer--;
|
|
}
|
|
|
|
/**
|
|
* This is the core algorithm which actually perform the diff itself,
|
|
* fragmenting the strings as per specified delimiters.
|
|
*
|
|
* This function is naturally recursive, however for performance purpose
|
|
* a local job queue is used instead of outright recursivity.
|
|
*/
|
|
private static function doFragmentDiff($from_text, $to_text, $delimiters) {
|
|
// Empty delimiter means character-level diffing.
|
|
// In such case, use code path optimized for character-level
|
|
// diffing.
|
|
if ( empty($delimiters) ) {
|
|
return FineDiff::doCharDiff($from_text, $to_text);
|
|
}
|
|
|
|
$result = array();
|
|
|
|
// fragment-level diffing
|
|
$from_text_len = strlen($from_text);
|
|
$to_text_len = strlen($to_text);
|
|
$from_fragments = FineDiff::extractFragments($from_text, $delimiters);
|
|
$to_fragments = FineDiff::extractFragments($to_text, $delimiters);
|
|
|
|
$jobs = array(array(0, $from_text_len, 0, $to_text_len));
|
|
|
|
$cached_array_keys = array();
|
|
|
|
while ( $job = array_pop($jobs) ) {
|
|
|
|
// get the segments which must be diff'ed
|
|
list($from_segment_start, $from_segment_end, $to_segment_start, $to_segment_end) = $job;
|
|
|
|
// catch easy cases first
|
|
$from_segment_length = $from_segment_end - $from_segment_start;
|
|
$to_segment_length = $to_segment_end - $to_segment_start;
|
|
if ( !$from_segment_length || !$to_segment_length ) {
|
|
if ( $from_segment_length ) {
|
|
$result[$from_segment_start * 4] = new FineDiffDeleteOp($from_segment_length);
|
|
}
|
|
else if ( $to_segment_length ) {
|
|
$result[$from_segment_start * 4 + 1] = new FineDiffInsertOp(substr($to_text, $to_segment_start, $to_segment_length));
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// find longest copy operation for the current segments
|
|
$best_copy_length = 0;
|
|
|
|
$from_base_fragment_index = $from_segment_start;
|
|
|
|
$cached_array_keys_for_current_segment = array();
|
|
|
|
while ( $from_base_fragment_index < $from_segment_end ) {
|
|
$from_base_fragment = $from_fragments[$from_base_fragment_index];
|
|
$from_base_fragment_length = strlen($from_base_fragment);
|
|
// performance boost: cache array keys
|
|
if ( !isset($cached_array_keys_for_current_segment[$from_base_fragment]) ) {
|
|
if ( !isset($cached_array_keys[$from_base_fragment]) ) {
|
|
$to_all_fragment_indices = $cached_array_keys[$from_base_fragment] = array_keys($to_fragments, $from_base_fragment, true);
|
|
}
|
|
else {
|
|
$to_all_fragment_indices = $cached_array_keys[$from_base_fragment];
|
|
}
|
|
// get only indices which falls within current segment
|
|
if ( $to_segment_start > 0 || $to_segment_end < $to_text_len ) {
|
|
$to_fragment_indices = array();
|
|
foreach ( $to_all_fragment_indices as $to_fragment_index ) {
|
|
if ( $to_fragment_index < $to_segment_start ) { continue; }
|
|
if ( $to_fragment_index >= $to_segment_end ) { break; }
|
|
$to_fragment_indices[] = $to_fragment_index;
|
|
}
|
|
$cached_array_keys_for_current_segment[$from_base_fragment] = $to_fragment_indices;
|
|
}
|
|
else {
|
|
$to_fragment_indices = $to_all_fragment_indices;
|
|
}
|
|
}
|
|
else {
|
|
$to_fragment_indices = $cached_array_keys_for_current_segment[$from_base_fragment];
|
|
}
|
|
// iterate through collected indices
|
|
foreach ( $to_fragment_indices as $to_base_fragment_index ) {
|
|
$fragment_index_offset = $from_base_fragment_length;
|
|
// iterate until no more match
|
|
for (;;) {
|
|
$fragment_from_index = $from_base_fragment_index + $fragment_index_offset;
|
|
if ( $fragment_from_index >= $from_segment_end ) {
|
|
break;
|
|
}
|
|
$fragment_to_index = $to_base_fragment_index + $fragment_index_offset;
|
|
if ( $fragment_to_index >= $to_segment_end ) {
|
|
break;
|
|
}
|
|
if ( $from_fragments[$fragment_from_index] !== $to_fragments[$fragment_to_index] ) {
|
|
break;
|
|
}
|
|
$fragment_length = strlen($from_fragments[$fragment_from_index]);
|
|
$fragment_index_offset += $fragment_length;
|
|
}
|
|
if ( $fragment_index_offset > $best_copy_length ) {
|
|
$best_copy_length = $fragment_index_offset;
|
|
$best_from_start = $from_base_fragment_index;
|
|
$best_to_start = $to_base_fragment_index;
|
|
}
|
|
}
|
|
$from_base_fragment_index += strlen($from_base_fragment);
|
|
// If match is larger than half segment size, no point trying to find better
|
|
// TODO: Really?
|
|
if ( $best_copy_length >= $from_segment_length / 2) {
|
|
break;
|
|
}
|
|
// no point to keep looking if what is left is less than
|
|
// current best match
|
|
if ( $from_base_fragment_index + $best_copy_length >= $from_segment_end ) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
if ( $best_copy_length ) {
|
|
$jobs[] = array($from_segment_start, $best_from_start, $to_segment_start, $best_to_start);
|
|
$result[$best_from_start * 4 + 2] = new FineDiffCopyOp($best_copy_length);
|
|
$jobs[] = array($best_from_start + $best_copy_length, $from_segment_end, $best_to_start + $best_copy_length, $to_segment_end);
|
|
}
|
|
else {
|
|
$result[$from_segment_start * 4 ] = new FineDiffReplaceOp($from_segment_length, substr($to_text, $to_segment_start, $to_segment_length));
|
|
}
|
|
}
|
|
|
|
ksort($result, SORT_NUMERIC);
|
|
return array_values($result);
|
|
}
|
|
|
|
/**
|
|
* Perform a character-level diff.
|
|
*
|
|
* The algorithm is quite similar to doFragmentDiff(), except that
|
|
* the code path is optimized for character-level diff -- strpos() is
|
|
* used to find out the longest common subequence of characters.
|
|
*
|
|
* We try to find a match using the longest possible subsequence, which
|
|
* is at most the length of the shortest of the two strings, then incrementally
|
|
* reduce the size until a match is found.
|
|
*
|
|
* I still need to study more the performance of this function. It
|
|
* appears that for long strings, the generic doFragmentDiff() is more
|
|
* performant. For word-sized strings, doCharDiff() is somewhat more
|
|
* performant.
|
|
*/
|
|
private static function doCharDiff($from_text, $to_text) {
|
|
$result = array();
|
|
$jobs = array(array(0, strlen($from_text), 0, strlen($to_text)));
|
|
while ( $job = array_pop($jobs) ) {
|
|
// get the segments which must be diff'ed
|
|
list($from_segment_start, $from_segment_end, $to_segment_start, $to_segment_end) = $job;
|
|
$from_segment_len = $from_segment_end - $from_segment_start;
|
|
$to_segment_len = $to_segment_end - $to_segment_start;
|
|
|
|
// catch easy cases first
|
|
if ( !$from_segment_len || !$to_segment_len ) {
|
|
if ( $from_segment_len ) {
|
|
$result[$from_segment_start * 4 + 0] = new FineDiffDeleteOp($from_segment_len);
|
|
}
|
|
else if ( $to_segment_len ) {
|
|
$result[$from_segment_start * 4 + 1] = new FineDiffInsertOp(substr($to_text, $to_segment_start, $to_segment_len));
|
|
}
|
|
continue;
|
|
}
|
|
if ( $from_segment_len >= $to_segment_len ) {
|
|
$copy_len = $to_segment_len;
|
|
while ( $copy_len ) {
|
|
$to_copy_start = $to_segment_start;
|
|
$to_copy_start_max = $to_segment_end - $copy_len;
|
|
while ( $to_copy_start <= $to_copy_start_max ) {
|
|
$from_copy_start = strpos(substr($from_text, $from_segment_start, $from_segment_len), substr($to_text, $to_copy_start, $copy_len));
|
|
if ( $from_copy_start !== false ) {
|
|
$from_copy_start += $from_segment_start;
|
|
break 2;
|
|
}
|
|
$to_copy_start++;
|
|
}
|
|
$copy_len--;
|
|
}
|
|
}
|
|
else {
|
|
$copy_len = $from_segment_len;
|
|
while ( $copy_len ) {
|
|
$from_copy_start = $from_segment_start;
|
|
$from_copy_start_max = $from_segment_end - $copy_len;
|
|
while ( $from_copy_start <= $from_copy_start_max ) {
|
|
$to_copy_start = strpos(substr($to_text, $to_segment_start, $to_segment_len), substr($from_text, $from_copy_start, $copy_len));
|
|
if ( $to_copy_start !== false ) {
|
|
$to_copy_start += $to_segment_start;
|
|
break 2;
|
|
}
|
|
$from_copy_start++;
|
|
}
|
|
$copy_len--;
|
|
}
|
|
}
|
|
// match found
|
|
if ( $copy_len ) {
|
|
$jobs[] = array($from_segment_start, $from_copy_start, $to_segment_start, $to_copy_start);
|
|
$result[$from_copy_start * 4 + 2] = new FineDiffCopyOp($copy_len);
|
|
$jobs[] = array($from_copy_start + $copy_len, $from_segment_end, $to_copy_start + $copy_len, $to_segment_end);
|
|
}
|
|
// no match, so delete all, insert all
|
|
else {
|
|
$result[$from_segment_start * 4] = new FineDiffReplaceOp($from_segment_len, substr($to_text, $to_segment_start, $to_segment_len));
|
|
}
|
|
}
|
|
ksort($result, SORT_NUMERIC);
|
|
return array_values($result);
|
|
}
|
|
|
|
/**
|
|
* Efficiently fragment the text into an array according to
|
|
* specified delimiters.
|
|
* No delimiters means fragment into single character.
|
|
* The array indices are the offset of the fragments into
|
|
* the input string.
|
|
* A sentinel empty fragment is always added at the end.
|
|
* Careful: No check is performed as to the validity of the
|
|
* delimiters.
|
|
*/
|
|
private static function extractFragments($text, $delimiters) {
|
|
// special case: split into characters
|
|
if ( empty($delimiters) ) {
|
|
$chars = str_split($text, 1);
|
|
$chars[strlen($text)] = '';
|
|
return $chars;
|
|
}
|
|
$fragments = array();
|
|
$start = $end = 0;
|
|
for (;;) {
|
|
$end += strcspn($text, $delimiters, $end);
|
|
$end += strspn($text, $delimiters, $end);
|
|
if ( $end === $start ) {
|
|
break;
|
|
}
|
|
$fragments[$start] = substr($text, $start, $end - $start);
|
|
$start = $end;
|
|
}
|
|
$fragments[$start] = '';
|
|
return $fragments;
|
|
}
|
|
|
|
/**
|
|
* Stock opcode renderers
|
|
*/
|
|
private static function renderToTextFromOpcode($opcode, $from, $from_offset, $from_len) {
|
|
if ( $opcode === 'c' || $opcode === 'i' ) {
|
|
echo substr($from, $from_offset, $from_len);
|
|
}
|
|
}
|
|
|
|
private static function renderDiffToHTMLFromOpcode($opcode, $from, $from_offset, $from_len) {
|
|
if ( $opcode === 'c' ) {
|
|
echo htmlentities(htmlentities(substr($from, $from_offset, $from_len)));
|
|
}
|
|
else if ( $opcode === 'd' ) {
|
|
$deletion = substr($from, $from_offset, $from_len);
|
|
if ( strcspn($deletion, " \n\r") === 0 ) {
|
|
$deletion = str_replace(array("\n","\r"), array('\n','\r'), $deletion);
|
|
}
|
|
echo '<del>', htmlentities(htmlentities($deletion)), '</del>';
|
|
}
|
|
else /* if ( $opcode === 'i' ) */ {
|
|
echo '<ins>', htmlentities(htmlentities(substr($from, $from_offset, $from_len))), '</ins>';
|
|
}
|
|
}
|
|
}
|
|
|