extensions/universalchardet/src/base/nsHebrewProber.cpp

changeset 0
6474c204b198
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/extensions/universalchardet/src/base/nsHebrewProber.cpp	Wed Dec 31 06:09:35 2014 +0100
     1.3 @@ -0,0 +1,162 @@
     1.4 +/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
     1.5 +/* This Source Code Form is subject to the terms of the Mozilla Public
     1.6 + * License, v. 2.0. If a copy of the MPL was not distributed with this
     1.7 + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
     1.8 +
     1.9 +#include "nsHebrewProber.h"
    1.10 +#include <stdio.h>
    1.11 +
    1.12 +// windows-1255 / ISO-8859-8 code points of interest
    1.13 +#define FINAL_KAF ('\xea')
    1.14 +#define NORMAL_KAF ('\xeb')
    1.15 +#define FINAL_MEM ('\xed')
    1.16 +#define NORMAL_MEM ('\xee')
    1.17 +#define FINAL_NUN ('\xef')
    1.18 +#define NORMAL_NUN ('\xf0')
    1.19 +#define FINAL_PE ('\xf3')
    1.20 +#define NORMAL_PE ('\xf4')
    1.21 +#define FINAL_TSADI ('\xf5')
    1.22 +#define NORMAL_TSADI ('\xf6')
    1.23 +
    1.24 +// Minimum Visual vs Logical final letter score difference.
    1.25 +// If the difference is below this, don't rely solely on the final letter score distance.
    1.26 +#define MIN_FINAL_CHAR_DISTANCE (5)
    1.27 +
    1.28 +// Minimum Visual vs Logical model score difference.
    1.29 +// If the difference is below this, don't rely at all on the model score distance.
    1.30 +#define MIN_MODEL_DISTANCE (0.01)
    1.31 +
    1.32 +#define VISUAL_HEBREW_NAME ("ISO-8859-8")
    1.33 +#define LOGICAL_HEBREW_NAME ("windows-1255")
    1.34 +
    1.35 +bool nsHebrewProber::isFinal(char c)
    1.36 +{
    1.37 +  return ((c == FINAL_KAF) || (c == FINAL_MEM) || (c == FINAL_NUN) || (c == FINAL_PE) || (c == FINAL_TSADI));
    1.38 +}
    1.39 +
    1.40 +bool nsHebrewProber::isNonFinal(char c)
    1.41 +{
    1.42 +  return ((c == NORMAL_KAF) || (c == NORMAL_MEM) || (c == NORMAL_NUN) || (c == NORMAL_PE));
    1.43 +  // The normal Tsadi is not a good Non-Final letter due to words like 
    1.44 +  // 'lechotet' (to chat) containing an apostrophe after the tsadi. This 
    1.45 +  // apostrophe is converted to a space in FilterWithoutEnglishLetters causing 
    1.46 +  // the Non-Final tsadi to appear at an end of a word even though this is not 
    1.47 +  // the case in the original text.
    1.48 +  // The letters Pe and Kaf rarely display a related behavior of not being a 
    1.49 +  // good Non-Final letter. Words like 'Pop', 'Winamp' and 'Mubarak' for 
    1.50 +  // example legally end with a Non-Final Pe or Kaf. However, the benefit of 
    1.51 +  // these letters as Non-Final letters outweighs the damage since these words 
    1.52 +  // are quite rare.
    1.53 +}
    1.54 +
    1.55 +/** HandleData
    1.56 + * Final letter analysis for logical-visual decision.
    1.57 + * Look for evidence that the received buffer is either logical Hebrew or 
    1.58 + * visual Hebrew.
    1.59 + * The following cases are checked:
    1.60 + * 1) A word longer than 1 letter, ending with a final letter. This is an 
    1.61 + *    indication that the text is laid out "naturally" since the final letter 
    1.62 + *    really appears at the end. +1 for logical score.
    1.63 + * 2) A word longer than 1 letter, ending with a Non-Final letter. In normal
    1.64 + *    Hebrew, words ending with Kaf, Mem, Nun, Pe or Tsadi, should not end with
    1.65 + *    the Non-Final form of that letter. Exceptions to this rule are mentioned
    1.66 + *    above in isNonFinal(). This is an indication that the text is laid out
    1.67 + *    backwards. +1 for visual score
    1.68 + * 3) A word longer than 1 letter, starting with a final letter. Final letters 
    1.69 + *    should not appear at the beginning of a word. This is an indication that 
    1.70 + *    the text is laid out backwards. +1 for visual score.
    1.71 + *
    1.72 + * The visual score and logical score are accumulated throughout the text and 
    1.73 + * are finally checked against each other in GetCharSetName().
    1.74 + * No checking for final letters in the middle of words is done since that case
    1.75 + * is not an indication for either Logical or Visual text.
    1.76 + *
    1.77 + * The input buffer should not contain any white spaces that are not (' ')
    1.78 + * or any low-ascii punctuation marks. 
    1.79 + */
    1.80 +nsProbingState nsHebrewProber::HandleData(const char* aBuf, uint32_t aLen)
    1.81 +{
    1.82 +  // Both model probers say it's not them. No reason to continue.
    1.83 +  if (GetState() == eNotMe)
    1.84 +    return eNotMe;
    1.85 +
    1.86 +  const char *curPtr, *endPtr = aBuf+aLen;
    1.87 +  char cur;
    1.88 +
    1.89 +  for (curPtr = (char*)aBuf; curPtr < endPtr; ++curPtr)
    1.90 +  {
    1.91 +    cur = *curPtr;
    1.92 +    if (cur == ' ') // We stand on a space - a word just ended
    1.93 +    {
    1.94 +      if (mBeforePrev != ' ') // *(curPtr-2) was not a space so prev is not a 1 letter word
    1.95 +      {
    1.96 +        if (isFinal(mPrev)) // case (1) [-2:not space][-1:final letter][cur:space]
    1.97 +          ++mFinalCharLogicalScore;
    1.98 +        else if (isNonFinal(mPrev)) // case (2) [-2:not space][-1:Non-Final letter][cur:space]
    1.99 +          ++mFinalCharVisualScore;
   1.100 +      }
   1.101 +    }
   1.102 +    else  // Not standing on a space
   1.103 +    {
   1.104 +      if ((mBeforePrev == ' ') && (isFinal(mPrev)) && (cur != ' ')) // case (3) [-2:space][-1:final letter][cur:not space]
   1.105 +        ++mFinalCharVisualScore;
   1.106 +    }
   1.107 +    mBeforePrev = mPrev;
   1.108 +    mPrev = cur;
   1.109 +  }
   1.110 +
   1.111 +  // Forever detecting, till the end or until both model probers return eNotMe (handled above).
   1.112 +  return eDetecting;
   1.113 +}
   1.114 +
   1.115 +// Make the decision: is it Logical or Visual?
   1.116 +const char* nsHebrewProber::GetCharSetName()
   1.117 +{
   1.118 +  // If the final letter score distance is dominant enough, rely on it.
   1.119 +  int32_t finalsub = mFinalCharLogicalScore - mFinalCharVisualScore;
   1.120 +  if (finalsub >= MIN_FINAL_CHAR_DISTANCE) 
   1.121 +    return LOGICAL_HEBREW_NAME;
   1.122 +  if (finalsub <= -(MIN_FINAL_CHAR_DISTANCE))
   1.123 +    return VISUAL_HEBREW_NAME;
   1.124 +
   1.125 +  // It's not dominant enough, try to rely on the model scores instead.
   1.126 +  float modelsub = mLogicalProb->GetConfidence() - mVisualProb->GetConfidence();
   1.127 +  if (modelsub > MIN_MODEL_DISTANCE)
   1.128 +    return LOGICAL_HEBREW_NAME;
   1.129 +  if (modelsub < -(MIN_MODEL_DISTANCE))
   1.130 +    return VISUAL_HEBREW_NAME;
   1.131 +
   1.132 +  // Still no good, back to final letter distance, maybe it'll save the day.
   1.133 +  if (finalsub < 0) 
   1.134 +    return VISUAL_HEBREW_NAME;
   1.135 +
   1.136 +  // (finalsub > 0 - Logical) or (don't know what to do) default to Logical.
   1.137 +  return LOGICAL_HEBREW_NAME;
   1.138 +}
   1.139 +
   1.140 +
   1.141 +void nsHebrewProber::Reset(void)
   1.142 +{
   1.143 +  mFinalCharLogicalScore = 0;
   1.144 +  mFinalCharVisualScore = 0;
   1.145 +
   1.146 +  // mPrev and mBeforePrev are initialized to space in order to simulate a word 
   1.147 +  // delimiter at the beginning of the data
   1.148 +  mPrev = ' ';
   1.149 +  mBeforePrev = ' ';
   1.150 +}
   1.151 +
   1.152 +nsProbingState nsHebrewProber::GetState(void) 
   1.153 +{
   1.154 +  // Remain active as long as any of the model probers are active.
   1.155 +  if ((mLogicalProb->GetState() == eNotMe) && (mVisualProb->GetState() == eNotMe))
   1.156 +    return eNotMe;
   1.157 +  return eDetecting;
   1.158 +}
   1.159 +
   1.160 +#ifdef DEBUG_chardet
   1.161 +void  nsHebrewProber::DumpStatus()
   1.162 +{
   1.163 +  printf("  HEB: %d - %d [Logical-Visual score]\r\n", mFinalCharLogicalScore, mFinalCharVisualScore);
   1.164 +}
   1.165 +#endif

mercurial