Fix hebrew character highlight issue in a special document

core/src/fpdftext/fpdf_text_int.cpp

 // Copyright 2014 PDFium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
  
 // Original code copyright 2014 Foxit Software Inc. http://www.foxitsoftware.com
 
 #include "../../include/fpdfapi/fpdf_resource.h"
 #include "../../include/fpdfapi/fpdf_pageobj.h"
 #include "../../include/fpdftext/fpdf_text.h"
 #include "../../include/fpdfapi/fpdf_page.h"
@@ skipping 1664 matching lines @@
                     baseSpace = kerning + spacing;
                 }
                 bAllChar = FALSE;
             }
         }
         spacing = 0;
         if(baseSpace < 0.0 || (nItems == 3 && !bAllChar)) {
             baseSpace = 0.0;
         }
     }
+
+    FX_BOOL bIsBidiAndMirrosInverse = FALSE;

[jbreiden, 2014/08/18 19:10:50]

What does MirrosInverse mean?

+    IFX_BidiChar* BidiChar = IFX_BidiChar::Create();
+    FX_INT32 nR2L = 0;
+    FX_INT32 nL2R = 0;
+    FX_INT32 start = 0, count = 0;
+    CPDF_TextObjectItem item;
+    for (FX_INT32 i = 0; i < nItems; i++) {
+        pTextObj->GetItemInfo(i, &item);
+        if (item.m_CharCode == (FX_DWORD)-1) {
+            continue;
+        }
+        CFX_WideString wstrItem = pFont->UnicodeFromCharCode(item.m_CharCode);
+        FX_WCHAR wChar = wstrItem.GetAt(0);
+        if ((wstrItem.IsEmpty() || wChar == 0) && item.m_CharCode) {

[jbreiden, 2014/08/18 19:10:49]

Same question. Is the character to Unicode mapping from the PDF file used?

[Bo Xu, 2014/08/19 03:31:49]

Yes, in CFX_WideString CPDF_Font::UnicodeFromCharCode(FX_DWORD charcode) const,
it does this :

    if (!m_bToUnicodeLoaded) {
        ((CPDF_Font*)this)->LoadUnicodeMap();
    }

to load the UnicodeMap if present.

[jbreiden, 2014/08/22 16:58:44]

Okay.  If I understand correctly:

The PDF files I provided have a UnicodeMap with full coverage. Therefore
UnicodeFromCharCode(item.m_CharCode) is successful, and line 1700 never
executes.

Looks good to me.

+            wChar = (FX_WCHAR)item.m_CharCode;
+        }
+        if (!wChar) {
+            continue;
+        }
+        if (BidiChar && BidiChar->AppendChar(wChar)) {
+            FX_INT32 ret = BidiChar->GetBidiInfo(start, count);
+            if (ret == 2) {
+                nR2L++;
+            }
+            else if (ret == 1) {
+                nL2R++;
+            }
+        }
+    }
+    if (BidiChar && BidiChar->EndChar()) {
+        FX_INT32 ret = BidiChar->GetBidiInfo(start, count);
+        if (ret == 2) {
+            nR2L++;
+        }
+        else if (ret == 1) {
+            nL2R++;
+        }
+    }
+    FX_BOOL bR2L = FALSE;
+    if (nR2L > 0 && nR2L >= nL2R) {
+        bR2L = TRUE;
+    }
+    bIsBidiAndMirrosInverse = bR2L && (matrix.a * matrix.d - matrix.b * matrix.c) < 0;
+    FX_INT32 iBufStartAppend = m_TempTextBuf.GetLength();
+    FX_INT32 iCharListStartAppend = m_TempCharList.GetSize();
+
     for (int i = 0; i < nItems; i++) {
         CPDF_TextObjectItem item;
         PAGECHAR_INFO charinfo;
         charinfo.m_OriginX = 0;
         charinfo.m_OriginY = 0;
         pTextObj->GetItemInfo(i, &item);
         if (item.m_CharCode == (FX_DWORD) - 1) {
             CFX_WideString str = m_TempTextBuf.GetWideString();
             if(str.IsEmpty()) {
                 str = m_TextBuf.GetWideString();
@@ skipping 126 matching lines @@
                 }
             } else if(i == 0) {
                 CFX_WideString str = m_TempTextBuf.GetWideString();
                 if (!str.IsEmpty() && str.GetAt(str.GetLength() - 1) == TEXT_BLANK_CHAR) {
                     m_TempTextBuf.Delete(m_TempTextBuf.GetLength() - 1, 1);
                     m_TempCharList.Delete(m_TempCharList.GetSize() - 1);
                 }
             }
         }
     }
+    if (bIsBidiAndMirrosInverse) {

[jbreiden, 2014/08/18 19:10:50]

Ok, I see what is going on here. The code is reversing some
text when necessary because of BiDi. This looks like a 
heuristic.

Please, please, please let me know if there is a mistake 
with the sample PDF file. I have 100% control over the PDF 
generation and can fix them if there is a mistake. This is 
better than PDFium being "smart" and using heuristics to 
work around bad PDF files.

[Bo Xu, 2014/08/19 03:31:49]

I wonder do you generate this particular pdf file? This is indeed a very special
file, it has an image object that shows the correct text, but the actual text
objects are stored in a mirror fashion which makes the text selection goes
wrong.

[jbreiden, 2014/08/22 16:58:44]

The PDF file is generated by Tesseract, which is an open source OCR program. I
am the engineer responsible for this feature and have the ability to change it.

https://code.google.com/p/tesseract-ocr/

I have no further comments, except to encourage the best possible solution. In
the future I will be working on PDF output for several other programs, so it is
good to work trouble now.

+        FX_INT32 i, j;
+        i = iCharListStartAppend;
+        j = m_TempCharList.GetSize() - 1;
+        PAGECHAR_INFO tempCharInfo;
+        FX_INT32 tempIndex = 0;
+        for (; i < j; i++, j--) {
+            tempCharInfo = m_TempCharList[i];
+            m_TempCharList[i] = m_TempCharList[j];
+            m_TempCharList[j] = tempCharInfo;
+            tempIndex = m_TempCharList[i].m_Index;
+            m_TempCharList[i].m_Index = m_TempCharList[j].m_Index;
+            m_TempCharList[j].m_Index = tempIndex;
+        }
+        FX_WCHAR * pTempBuffer = m_TempTextBuf.GetBuffer();
+        i = iBufStartAppend;
+        j = m_TempTextBuf.GetLength() - 1;
+        FX_WCHAR wTemp;
+        for (; i < j; i++, j--) {
+            wTemp = pTempBuffer[i];
+            pTempBuffer[i] = pTempBuffer[j];
+            pTempBuffer[j] = wTemp;
+        }
+    }
 }
 FX_INT32 CPDF_TextPage::GetTextObjectWritingMode(const CPDF_TextObject* pTextObj)
 {
     FX_INT32 nChars = pTextObj->CountChars();
     if (nChars == 1) {
         return m_TextlineDir;
     }
     CPDF_TextObjectItem first, last;
     pTextObj->GetCharInfo(0, &first);
     pTextObj->GetCharInfo(nChars - 1, &last);
@@ skipping 936 matching lines @@
     if (!m_IsParserd || index < 0 || index >= m_LinkList.GetSize()) {
         return;
     }
     CPDF_LinkExt* link = NULL;
     link = m_LinkList.GetAt(index);
     if (!link) {
         return ;
     }
     m_pTextPage->GetRectArray(link->m_Start, link->m_Count, rects);
 }