| // 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 <ctype.h> | |
| #include <algorithm> | |
| #include "../../../third_party/base/nonstd_unique_ptr.h" | |
| #include "../../include/fpdfapi/fpdf_module.h" | |
| #include "../../include/fpdfapi/fpdf_page.h" | |
| #include "../../include/fpdfapi/fpdf_pageobj.h" | |
| #include "../../include/fpdfapi/fpdf_resource.h" | |
| #include "../../include/fpdftext/fpdf_text.h" | |
| #include "../../include/fxcrt/fx_bidi.h" | |
| #include "../../include/fxcrt/fx_ucd.h" | |
| #include "text_int.h" | |
| namespace { | |
| FX_BOOL _IsIgnoreSpaceCharacter(FX_WCHAR curChar) { | |
| if (curChar < 255) { | |
| return FALSE; | |
| } | |
| if ((curChar >= 0x0600 && curChar <= 0x06FF) || | |
| (curChar >= 0xFE70 && curChar <= 0xFEFF) || | |
| (curChar >= 0xFB50 && curChar <= 0xFDFF) || | |
| (curChar >= 0x0400 && curChar <= 0x04FF) || | |
| (curChar >= 0x0500 && curChar <= 0x052F) || | |
| (curChar >= 0xA640 && curChar <= 0xA69F) || | |
| (curChar >= 0x2DE0 && curChar <= 0x2DFF) || curChar == 8467 || | |
| (curChar >= 0x2000 && curChar <= 0x206F)) { | |
| return FALSE; | |
| } | |
| return TRUE; | |
| } | |
| FX_FLOAT _NormalizeThreshold(FX_FLOAT threshold) { | |
| if (threshold < 300) { | |
| return threshold / 2.0f; | |
| } | |
| if (threshold < 500) { | |
| return threshold / 4.0f; | |
| } | |
| if (threshold < 700) { | |
| return threshold / 5.0f; | |
| } | |
| return threshold / 6.0f; | |
| } | |
| FX_FLOAT _CalculateBaseSpace(const CPDF_TextObject* pTextObj, | |
| const CFX_AffineMatrix& matrix) { | |
| FX_FLOAT baseSpace = 0.0; | |
| const int nItems = pTextObj->CountItems(); | |
| if (pTextObj->m_TextState.GetObject()->m_CharSpace && nItems >= 3) { | |
| FX_BOOL bAllChar = TRUE; | |
| FX_FLOAT spacing = matrix.TransformDistance( | |
| pTextObj->m_TextState.GetObject()->m_CharSpace); | |
| baseSpace = spacing; | |
| for (int i = 0; i < nItems; i++) { | |
| CPDF_TextObjectItem item; | |
| pTextObj->GetItemInfo(i, &item); | |
| if (item.m_CharCode == (FX_DWORD)-1) { | |
| FX_FLOAT fontsize_h = pTextObj->m_TextState.GetFontSizeH(); | |
| FX_FLOAT kerning = -fontsize_h * item.m_OriginX / 1000; | |
| baseSpace = std::min(baseSpace, kerning + spacing); | |
| bAllChar = FALSE; | |
| } | |
| } | |
| if (baseSpace < 0.0 || (nItems == 3 && !bAllChar)) { | |
| baseSpace = 0.0; | |
| } | |
| } | |
| return baseSpace; | |
| } | |
| } // namespace | |
| CPDFText_ParseOptions::CPDFText_ParseOptions() | |
| : m_bGetCharCodeOnly(FALSE), | |
| m_bNormalizeObjs(TRUE), | |
| m_bOutputHyphen(FALSE) {} | |
| IPDF_TextPage* IPDF_TextPage::CreateTextPage(const CPDF_Page* pPage, | |
| int flags) { | |
| return new CPDF_TextPage(pPage, flags); | |
| } | |
| IPDF_TextPageFind* IPDF_TextPageFind::CreatePageFind( | |
| const IPDF_TextPage* pTextPage) { | |
| return pTextPage ? new CPDF_TextPageFind(pTextPage) : nullptr; | |
| } | |
| IPDF_LinkExtract* IPDF_LinkExtract::CreateLinkExtract() { | |
| return new CPDF_LinkExtract(); | |
| } | |
| #define TEXT_BLANK_CHAR L' ' | |
| #define TEXT_LINEFEED_CHAR L'\n' | |
| #define TEXT_RETURN_CHAR L'\r' | |
| #define TEXT_EMPTY L"" | |
| #define TEXT_BLANK L" " | |
| #define TEXT_RETURN_LINEFEED L"\r\n" | |
| #define TEXT_LINEFEED L"\n" | |
| #define TEXT_CHARRATIO_GAPDELTA 0.070 | |
| CPDF_TextPage::CPDF_TextPage(const CPDF_Page* pPage, int flags) | |
| : m_pPage(pPage), | |
| m_charList(512), | |
| m_TempCharList(50), | |
| m_parserflag(flags), | |
| m_pPreTextObj(nullptr), | |
| m_bIsParsed(false), | |
| m_TextlineDir(-1), | |
| m_CurlineRect(0, 0, 0, 0) { | |
| m_TextBuf.EstimateSize(0, 10240); | |
| pPage->GetDisplayMatrix(m_DisplayMatrix, 0, 0, (int)pPage->GetPageWidth(), | |
| (int)pPage->GetPageHeight(), 0); | |
| } | |
| void CPDF_TextPage::NormalizeObjects(FX_BOOL bNormalize) { | |
| m_ParseOptions.m_bNormalizeObjs = bNormalize; | |
| } | |
| bool CPDF_TextPage::IsControlChar(const PAGECHAR_INFO& charInfo) { | |
| switch (charInfo.m_Unicode) { | |
| case 0x2: | |
| case 0x3: | |
| case 0x93: | |
| case 0x94: | |
| case 0x96: | |
| case 0x97: | |
| case 0x98: | |
| case 0xfffe: | |
| return charInfo.m_Flag != FPDFTEXT_CHAR_HYPHEN; | |
| default: | |
| return false; | |
| } | |
| } | |
| FX_BOOL CPDF_TextPage::ParseTextPage() { | |
| m_bIsParsed = false; | |
| if (!m_pPage) | |
| return FALSE; | |
| m_TextBuf.Clear(); | |
| m_charList.RemoveAll(); | |
| m_pPreTextObj = NULL; | |
| ProcessObject(); | |
| m_bIsParsed = true; | |
| if (!m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_CharIndex.RemoveAll(); | |
| int nCount = m_charList.GetSize(); | |
| if (nCount) { | |
| m_CharIndex.Add(0); | |
| } | |
| for (int i = 0; i < nCount; i++) { | |
| int indexSize = m_CharIndex.GetSize(); | |
| FX_BOOL bNormal = FALSE; | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(i); | |
| if (charinfo.m_Flag == FPDFTEXT_CHAR_GENERATED) { | |
| bNormal = TRUE; | |
| } else if (charinfo.m_Unicode == 0 || IsControlChar(charinfo)) | |
| bNormal = FALSE; | |
| else { | |
| bNormal = TRUE; | |
| } | |
| if (bNormal) { | |
| if (indexSize % 2) { | |
| m_CharIndex.Add(1); | |
| } else { | |
| if (indexSize <= 0) { | |
| continue; | |
| } | |
| m_CharIndex.SetAt(indexSize - 1, | |
| m_CharIndex.GetAt(indexSize - 1) + 1); | |
| } | |
| } else { | |
| if (indexSize % 2) { | |
| if (indexSize <= 0) { | |
| continue; | |
| } | |
| m_CharIndex.SetAt(indexSize - 1, i + 1); | |
| } else { | |
| m_CharIndex.Add(i + 1); | |
| } | |
| } | |
| } | |
| int indexSize = m_CharIndex.GetSize(); | |
| if (indexSize % 2) { | |
| m_CharIndex.RemoveAt(indexSize - 1); | |
| } | |
| } | |
| return TRUE; | |
| } | |
| int CPDF_TextPage::CountChars() const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return m_TextBuf.GetSize(); | |
| } | |
| return m_charList.GetSize(); | |
| } | |
| int CPDF_TextPage::CharIndexFromTextIndex(int TextIndex) const { | |
| int indexSize = m_CharIndex.GetSize(); | |
| int count = 0; | |
| for (int i = 0; i < indexSize; i += 2) { | |
| count += m_CharIndex.GetAt(i + 1); | |
| if (count > TextIndex) { | |
| return TextIndex - count + m_CharIndex.GetAt(i + 1) + | |
| m_CharIndex.GetAt(i); | |
| } | |
| } | |
| return -1; | |
| } | |
| int CPDF_TextPage::TextIndexFromCharIndex(int CharIndex) const { | |
| int indexSize = m_CharIndex.GetSize(); | |
| int count = 0; | |
| for (int i = 0; i < indexSize; i += 2) { | |
| count += m_CharIndex.GetAt(i + 1); | |
| if (m_CharIndex.GetAt(i + 1) + m_CharIndex.GetAt(i) > CharIndex) { | |
| if (CharIndex - m_CharIndex.GetAt(i) < 0) { | |
| return -1; | |
| } | |
| return CharIndex - m_CharIndex.GetAt(i) + count - | |
| m_CharIndex.GetAt(i + 1); | |
| } | |
| } | |
| return -1; | |
| } | |
| void CPDF_TextPage::GetRectArray(int start, | |
| int nCount, | |
| CFX_RectArray& rectArray) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return; | |
| } | |
| if (start < 0 || nCount == 0) { | |
| return; | |
| } | |
| if (!m_bIsParsed) { | |
| return; | |
| } | |
| PAGECHAR_INFO info_curchar; | |
| CPDF_TextObject* pCurObj = NULL; | |
| CFX_FloatRect rect; | |
| int curPos = start; | |
| FX_BOOL flagNewRect = TRUE; | |
| if (nCount + start > m_charList.GetSize() || nCount == -1) { | |
| nCount = m_charList.GetSize() - start; | |
| } | |
| while (nCount--) { | |
| info_curchar = *(PAGECHAR_INFO*)m_charList.GetAt(curPos++); | |
| if (info_curchar.m_Flag == FPDFTEXT_CHAR_GENERATED) { | |
| continue; | |
| } | |
| if (info_curchar.m_CharBox.Width() < 0.01 || | |
| info_curchar.m_CharBox.Height() < 0.01) { | |
| continue; | |
| } | |
| if (!pCurObj) { | |
| pCurObj = info_curchar.m_pTextObj; | |
| } | |
| if (pCurObj != info_curchar.m_pTextObj) { | |
| rectArray.Add(rect); | |
| pCurObj = info_curchar.m_pTextObj; | |
| flagNewRect = TRUE; | |
| } | |
| if (flagNewRect) { | |
| FX_FLOAT orgX = info_curchar.m_OriginX, orgY = info_curchar.m_OriginY; | |
| CFX_AffineMatrix matrix, matrix_reverse; | |
| info_curchar.m_pTextObj->GetTextMatrix(&matrix); | |
| matrix.Concat(info_curchar.m_Matrix); | |
| matrix_reverse.SetReverse(matrix); | |
| matrix_reverse.Transform(orgX, orgY); | |
| rect.left = info_curchar.m_CharBox.left; | |
| rect.right = info_curchar.m_CharBox.right; | |
| if (pCurObj->GetFont()->GetTypeDescent()) { | |
| rect.bottom = orgY + | |
| pCurObj->GetFont()->GetTypeDescent() * | |
| pCurObj->GetFontSize() / 1000; | |
| FX_FLOAT xPosTemp = orgX; | |
| matrix.Transform(xPosTemp, rect.bottom); | |
| } else { | |
| rect.bottom = info_curchar.m_CharBox.bottom; | |
| } | |
| if (pCurObj->GetFont()->GetTypeAscent()) { | |
| rect.top = | |
| orgY + | |
| pCurObj->GetFont()->GetTypeAscent() * pCurObj->GetFontSize() / 1000; | |
| FX_FLOAT xPosTemp = | |
| orgX + | |
| GetCharWidth(info_curchar.m_CharCode, pCurObj->GetFont()) * | |
| pCurObj->GetFontSize() / 1000; | |
| matrix.Transform(xPosTemp, rect.top); | |
| } else { | |
| rect.top = info_curchar.m_CharBox.top; | |
| } | |
| flagNewRect = FALSE; | |
| rect = info_curchar.m_CharBox; | |
| rect.Normalize(); | |
| } else { | |
| info_curchar.m_CharBox.Normalize(); | |
| if (rect.left > info_curchar.m_CharBox.left) { | |
| rect.left = info_curchar.m_CharBox.left; | |
| } | |
| if (rect.right < info_curchar.m_CharBox.right) { | |
| rect.right = info_curchar.m_CharBox.right; | |
| } | |
| if (rect.top < info_curchar.m_CharBox.top) { | |
| rect.top = info_curchar.m_CharBox.top; | |
| } | |
| if (rect.bottom > info_curchar.m_CharBox.bottom) { | |
| rect.bottom = info_curchar.m_CharBox.bottom; | |
| } | |
| } | |
| } | |
| rectArray.Add(rect); | |
| return; | |
| } | |
| int CPDF_TextPage::GetIndexAtPos(CPDF_Point point, | |
| FX_FLOAT xTolerance, | |
| FX_FLOAT yTolerance) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return -3; | |
| int pos = 0; | |
| int NearPos = -1; | |
| double xdif = 5000, ydif = 5000; | |
| while (pos < m_charList.GetSize()) { | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)(m_charList.GetAt(pos)); | |
| CFX_FloatRect charrect = charinfo.m_CharBox; | |
| if (charrect.Contains(point.x, point.y)) { | |
| break; | |
| } | |
| if (xTolerance > 0 || yTolerance > 0) { | |
| CFX_FloatRect charRectExt; | |
| charrect.Normalize(); | |
| charRectExt.left = charrect.left - xTolerance / 2; | |
| charRectExt.right = charrect.right + xTolerance / 2; | |
| charRectExt.top = charrect.top + yTolerance / 2; | |
| charRectExt.bottom = charrect.bottom - yTolerance / 2; | |
| if (charRectExt.Contains(point.x, point.y)) { | |
| double curXdif, curYdif; | |
| curXdif = FXSYS_fabs(point.x - charrect.left) < | |
| FXSYS_fabs(point.x - charrect.right) | |
| ? FXSYS_fabs(point.x - charrect.left) | |
| : FXSYS_fabs(point.x - charrect.right); | |
| curYdif = FXSYS_fabs(point.y - charrect.bottom) < | |
| FXSYS_fabs(point.y - charrect.top) | |
| ? FXSYS_fabs(point.y - charrect.bottom) | |
| : FXSYS_fabs(point.y - charrect.top); | |
| if (curYdif + curXdif < xdif + ydif) { | |
| ydif = curYdif; | |
| xdif = curXdif; | |
| NearPos = pos; | |
| } | |
| } | |
| } | |
| ++pos; | |
| } | |
| if (pos >= m_charList.GetSize()) { | |
| pos = NearPos; | |
| } | |
| return pos; | |
| } | |
| CFX_WideString CPDF_TextPage::GetTextByRect(const CFX_FloatRect& rect) const { | |
| CFX_WideString strText; | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return strText; | |
| int nCount = m_charList.GetSize(); | |
| int pos = 0; | |
| FX_FLOAT posy = 0; | |
| FX_BOOL IsContainPreChar = FALSE; | |
| FX_BOOL ISAddLineFeed = FALSE; | |
| while (pos < nCount) { | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(pos++); | |
| if (IsRectIntersect(rect, charinfo.m_CharBox)) { | |
| if (FXSYS_fabs(posy - charinfo.m_OriginY) > 0 && !IsContainPreChar && | |
| ISAddLineFeed) { | |
| posy = charinfo.m_OriginY; | |
| if (strText.GetLength() > 0) { | |
| strText += L"\r\n"; | |
| } | |
| } | |
| IsContainPreChar = TRUE; | |
| ISAddLineFeed = FALSE; | |
| if (charinfo.m_Unicode) { | |
| strText += charinfo.m_Unicode; | |
| } | |
| } else if (charinfo.m_Unicode == 32) { | |
| if (IsContainPreChar && charinfo.m_Unicode) { | |
| strText += charinfo.m_Unicode; | |
| IsContainPreChar = FALSE; | |
| ISAddLineFeed = FALSE; | |
| } | |
| } else { | |
| IsContainPreChar = FALSE; | |
| ISAddLineFeed = TRUE; | |
| } | |
| } | |
| return strText; | |
| } | |
| void CPDF_TextPage::GetRectsArrayByRect(const CFX_FloatRect& rect, | |
| CFX_RectArray& resRectArray) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return; | |
| CFX_FloatRect curRect; | |
| FX_BOOL flagNewRect = TRUE; | |
| CPDF_TextObject* pCurObj = NULL; | |
| int nCount = m_charList.GetSize(); | |
| int pos = 0; | |
| while (pos < nCount) { | |
| PAGECHAR_INFO info_curchar = *(PAGECHAR_INFO*)m_charList.GetAt(pos++); | |
| if (info_curchar.m_Flag == FPDFTEXT_CHAR_GENERATED) { | |
| continue; | |
| } | |
| if (IsRectIntersect(rect, info_curchar.m_CharBox)) { | |
| if (!pCurObj) { | |
| pCurObj = info_curchar.m_pTextObj; | |
| } | |
| if (pCurObj != info_curchar.m_pTextObj) { | |
| resRectArray.Add(curRect); | |
| pCurObj = info_curchar.m_pTextObj; | |
| flagNewRect = TRUE; | |
| } | |
| if (flagNewRect) { | |
| curRect = info_curchar.m_CharBox; | |
| flagNewRect = FALSE; | |
| curRect.Normalize(); | |
| } else { | |
| info_curchar.m_CharBox.Normalize(); | |
| if (curRect.left > info_curchar.m_CharBox.left) { | |
| curRect.left = info_curchar.m_CharBox.left; | |
| } | |
| if (curRect.right < info_curchar.m_CharBox.right) { | |
| curRect.right = info_curchar.m_CharBox.right; | |
| } | |
| if (curRect.top < info_curchar.m_CharBox.top) { | |
| curRect.top = info_curchar.m_CharBox.top; | |
| } | |
| if (curRect.bottom > info_curchar.m_CharBox.bottom) { | |
| curRect.bottom = info_curchar.m_CharBox.bottom; | |
| } | |
| } | |
| } | |
| } | |
| resRectArray.Add(curRect); | |
| return; | |
| } | |
| int CPDF_TextPage::GetIndexAtPos(FX_FLOAT x, | |
| FX_FLOAT y, | |
| FX_FLOAT xTolerance, | |
| FX_FLOAT yTolerance) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return -3; | |
| } | |
| CPDF_Point point(x, y); | |
| return GetIndexAtPos(point, xTolerance, yTolerance); | |
| } | |
| void CPDF_TextPage::GetCharInfo(int index, FPDF_CHAR_INFO& info) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return; | |
| if (index < 0 || index >= m_charList.GetSize()) | |
| return; | |
| PAGECHAR_INFO charinfo; | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(index); | |
| info.m_Charcode = charinfo.m_CharCode; | |
| info.m_OriginX = charinfo.m_OriginX; | |
| info.m_OriginY = charinfo.m_OriginY; | |
| info.m_Unicode = charinfo.m_Unicode; | |
| info.m_Flag = charinfo.m_Flag; | |
| info.m_CharBox = charinfo.m_CharBox; | |
| info.m_pTextObj = charinfo.m_pTextObj; | |
| if (charinfo.m_pTextObj && charinfo.m_pTextObj->GetFont()) { | |
| info.m_FontSize = charinfo.m_pTextObj->GetFontSize(); | |
| } | |
| info.m_Matrix.Copy(charinfo.m_Matrix); | |
| return; | |
| } | |
| void CPDF_TextPage::CheckMarkedContentObject(int32_t& start, | |
| int32_t& nCount) const { | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(start); | |
| PAGECHAR_INFO charinfo2 = | |
| *(PAGECHAR_INFO*)m_charList.GetAt(start + nCount - 1); | |
| if (FPDFTEXT_CHAR_PIECE != charinfo.m_Flag && | |
| FPDFTEXT_CHAR_PIECE != charinfo2.m_Flag) { | |
| return; | |
| } | |
| if (FPDFTEXT_CHAR_PIECE == charinfo.m_Flag) { | |
| PAGECHAR_INFO charinfo1 = charinfo; | |
| int startIndex = start; | |
| while (FPDFTEXT_CHAR_PIECE == charinfo1.m_Flag && | |
| charinfo1.m_Index == charinfo.m_Index) { | |
| startIndex--; | |
| if (startIndex < 0) { | |
| break; | |
| } | |
| charinfo1 = *(PAGECHAR_INFO*)m_charList.GetAt(startIndex); | |
| } | |
| startIndex++; | |
| start = startIndex; | |
| } | |
| if (FPDFTEXT_CHAR_PIECE == charinfo2.m_Flag) { | |
| PAGECHAR_INFO charinfo3 = charinfo2; | |
| int endIndex = start + nCount - 1; | |
| while (FPDFTEXT_CHAR_PIECE == charinfo3.m_Flag && | |
| charinfo3.m_Index == charinfo2.m_Index) { | |
| endIndex++; | |
| if (endIndex >= m_charList.GetSize()) { | |
| break; | |
| } | |
| charinfo3 = *(PAGECHAR_INFO*)m_charList.GetAt(endIndex); | |
| } | |
| endIndex--; | |
| nCount = endIndex - start + 1; | |
| } | |
| } | |
| CFX_WideString CPDF_TextPage::GetPageText(int start, int nCount) const { | |
| if (!m_bIsParsed || nCount == 0) | |
| return L""; | |
| if (start < 0) | |
| start = 0; | |
| if (nCount == -1) { | |
| nCount = m_charList.GetSize() - start; | |
| return m_TextBuf.GetWideString().Mid(start, | |
| m_TextBuf.GetWideString().GetLength()); | |
| } | |
| if (nCount <= 0 || m_charList.GetSize() <= 0) { | |
| return L""; | |
| } | |
| if (nCount + start > m_charList.GetSize() - 1) { | |
| nCount = m_charList.GetSize() - start; | |
| } | |
| if (nCount <= 0) { | |
| return L""; | |
| } | |
| CheckMarkedContentObject(start, nCount); | |
| int startindex = 0; | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(start); | |
| int startOffset = 0; | |
| while (charinfo.m_Index == -1) { | |
| startOffset++; | |
| if (startOffset > nCount || start + startOffset >= m_charList.GetSize()) { | |
| return L""; | |
| } | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(start + startOffset); | |
| } | |
| startindex = charinfo.m_Index; | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(start + nCount - 1); | |
| int nCountOffset = 0; | |
| while (charinfo.m_Index == -1) { | |
| nCountOffset++; | |
| if (nCountOffset >= nCount) { | |
| return L""; | |
| } | |
| charinfo = | |
| *(PAGECHAR_INFO*)m_charList.GetAt(start + nCount - nCountOffset - 1); | |
| } | |
| nCount = start + nCount - nCountOffset - startindex; | |
| if (nCount <= 0) { | |
| return L""; | |
| } | |
| return m_TextBuf.GetWideString().Mid(startindex, nCount); | |
| } | |
| int CPDF_TextPage::CountRects(int start, int nCount) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed || start < 0) | |
| return -1; | |
| if (nCount == -1 || nCount + start > m_charList.GetSize()) { | |
| nCount = m_charList.GetSize() - start; | |
| } | |
| m_SelRects.RemoveAll(); | |
| GetRectArray(start, nCount, m_SelRects); | |
| return m_SelRects.GetSize(); | |
| } | |
| void CPDF_TextPage::GetRect(int rectIndex, | |
| FX_FLOAT& left, | |
| FX_FLOAT& top, | |
| FX_FLOAT& right, | |
| FX_FLOAT& bottom) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return; | |
| if (rectIndex < 0 || rectIndex >= m_SelRects.GetSize()) | |
| return; | |
| left = m_SelRects.GetAt(rectIndex).left; | |
| top = m_SelRects.GetAt(rectIndex).top; | |
| right = m_SelRects.GetAt(rectIndex).right; | |
| bottom = m_SelRects.GetAt(rectIndex).bottom; | |
| } | |
| FX_BOOL CPDF_TextPage::GetBaselineRotate(int start, int end, int& Rotate) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return FALSE; | |
| } | |
| if (end == start) { | |
| return FALSE; | |
| } | |
| FX_FLOAT dx, dy; | |
| FPDF_CHAR_INFO info1, info2; | |
| GetCharInfo(start, info1); | |
| GetCharInfo(end, info2); | |
| while (info2.m_CharBox.Width() == 0 || info2.m_CharBox.Height() == 0) { | |
| end--; | |
| if (end <= start) { | |
| return FALSE; | |
| } | |
| GetCharInfo(end, info2); | |
| } | |
| dx = (info2.m_OriginX - info1.m_OriginX); | |
| dy = (info2.m_OriginY - info1.m_OriginY); | |
| if (dx == 0) { | |
| if (dy > 0) { | |
| Rotate = 90; | |
| } else if (dy < 0) { | |
| Rotate = 270; | |
| } else { | |
| Rotate = 0; | |
| } | |
| } else { | |
| float a = FXSYS_atan2(dy, dx); | |
| Rotate = (int)(a * 180 / FX_PI + 0.5); | |
| } | |
| if (Rotate < 0) { | |
| Rotate = -Rotate; | |
| } else if (Rotate > 0) { | |
| Rotate = 360 - Rotate; | |
| } | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_TextPage::GetBaselineRotate(const CFX_FloatRect& rect, | |
| int& Rotate) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return FALSE; | |
| } | |
| int start, end, count, | |
| n = CountBoundedSegments(rect.left, rect.top, rect.right, rect.bottom, | |
| TRUE); | |
| if (n < 1) { | |
| return FALSE; | |
| } | |
| if (n > 1) { | |
| GetBoundedSegment(n - 1, start, count); | |
| end = start + count - 1; | |
| GetBoundedSegment(0, start, count); | |
| } else { | |
| GetBoundedSegment(0, start, count); | |
| end = start + count - 1; | |
| } | |
| return GetBaselineRotate(start, end, Rotate); | |
| } | |
| FX_BOOL CPDF_TextPage::GetBaselineRotate(int rectIndex, int& Rotate) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return FALSE; | |
| if (rectIndex < 0 || rectIndex > m_SelRects.GetSize()) | |
| return FALSE; | |
| CFX_FloatRect rect = m_SelRects.GetAt(rectIndex); | |
| return GetBaselineRotate(rect, Rotate); | |
| } | |
| int CPDF_TextPage::CountBoundedSegments(FX_FLOAT left, | |
| FX_FLOAT top, | |
| FX_FLOAT right, | |
| FX_FLOAT bottom, | |
| FX_BOOL bContains) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) | |
| return -1; | |
| m_Segment.RemoveAll(); | |
| if (!m_bIsParsed) | |
| return -1; | |
| CFX_FloatRect rect(left, bottom, right, top); | |
| rect.Normalize(); | |
| int nCount = m_charList.GetSize(); | |
| int pos = 0; | |
| FPDF_SEGMENT segment; | |
| segment.m_Start = 0; | |
| segment.m_nCount = 0; | |
| int segmentStatus = 0; | |
| FX_BOOL IsContainPreChar = FALSE; | |
| while (pos < nCount) { | |
| PAGECHAR_INFO charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(pos); | |
| if (bContains && rect.Contains(charinfo.m_CharBox)) { | |
| if (segmentStatus == 0 || segmentStatus == 2) { | |
| segment.m_Start = pos; | |
| segment.m_nCount = 1; | |
| segmentStatus = 1; | |
| } else if (segmentStatus == 1) { | |
| segment.m_nCount++; | |
| } | |
| IsContainPreChar = TRUE; | |
| } else if (!bContains && | |
| (IsRectIntersect(rect, charinfo.m_CharBox) || | |
| rect.Contains(charinfo.m_OriginX, charinfo.m_OriginY))) { | |
| if (segmentStatus == 0 || segmentStatus == 2) { | |
| segment.m_Start = pos; | |
| segment.m_nCount = 1; | |
| segmentStatus = 1; | |
| } else if (segmentStatus == 1) { | |
| segment.m_nCount++; | |
| } | |
| IsContainPreChar = TRUE; | |
| } else if (charinfo.m_Unicode == 32) { | |
| if (IsContainPreChar == TRUE) { | |
| if (segmentStatus == 0 || segmentStatus == 2) { | |
| segment.m_Start = pos; | |
| segment.m_nCount = 1; | |
| segmentStatus = 1; | |
| } else if (segmentStatus == 1) { | |
| segment.m_nCount++; | |
| } | |
| IsContainPreChar = FALSE; | |
| } else { | |
| if (segmentStatus == 1) { | |
| segmentStatus = 2; | |
| m_Segment.Add(segment); | |
| segment.m_Start = 0; | |
| segment.m_nCount = 0; | |
| } | |
| } | |
| } else { | |
| if (segmentStatus == 1) { | |
| segmentStatus = 2; | |
| m_Segment.Add(segment); | |
| segment.m_Start = 0; | |
| segment.m_nCount = 0; | |
| } | |
| IsContainPreChar = FALSE; | |
| } | |
| pos++; | |
| } | |
| if (segmentStatus == 1) { | |
| segmentStatus = 2; | |
| m_Segment.Add(segment); | |
| segment.m_Start = 0; | |
| segment.m_nCount = 0; | |
| } | |
| return m_Segment.GetSize(); | |
| } | |
| void CPDF_TextPage::GetBoundedSegment(int index, int& start, int& count) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| return; | |
| } | |
| if (index < 0 || index >= m_Segment.GetSize()) { | |
| return; | |
| } | |
| start = m_Segment.GetAt(index).m_Start; | |
| count = m_Segment.GetAt(index).m_nCount; | |
| } | |
| int CPDF_TextPage::GetWordBreak(int index, int direction) const { | |
| if (m_ParseOptions.m_bGetCharCodeOnly || !m_bIsParsed) | |
| return -1; | |
| if (direction != FPDFTEXT_LEFT && direction != FPDFTEXT_RIGHT) | |
| return -1; | |
| if (index < 0 || index >= m_charList.GetSize()) | |
| return -1; | |
| PAGECHAR_INFO charinfo; | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(index); | |
| if (charinfo.m_Index == -1 || charinfo.m_Flag == FPDFTEXT_CHAR_GENERATED) { | |
| return index; | |
| } | |
| if (!IsLetter(charinfo.m_Unicode)) { | |
| return index; | |
| } | |
| int breakPos = index; | |
| if (direction == FPDFTEXT_LEFT) { | |
| while (--breakPos > 0) { | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(breakPos); | |
| if (!IsLetter(charinfo.m_Unicode)) { | |
| return breakPos; | |
| } | |
| } | |
| } else if (direction == FPDFTEXT_RIGHT) { | |
| while (++breakPos < m_charList.GetSize()) { | |
| charinfo = *(PAGECHAR_INFO*)m_charList.GetAt(breakPos); | |
| if (!IsLetter(charinfo.m_Unicode)) { | |
| return breakPos; | |
| } | |
| } | |
| } | |
| return breakPos; | |
| } | |
| int32_t CPDF_TextPage::FindTextlineFlowDirection() { | |
| if (!m_pPage) { | |
| return -1; | |
| } | |
| const int32_t nPageWidth = (int32_t)((CPDF_Page*)m_pPage)->GetPageWidth(); | |
| const int32_t nPageHeight = (int32_t)((CPDF_Page*)m_pPage)->GetPageHeight(); | |
| CFX_ByteArray nHorizontalMask; | |
| if (!nHorizontalMask.SetSize(nPageWidth)) { | |
| return -1; | |
| } | |
| uint8_t* pDataH = nHorizontalMask.GetData(); | |
| CFX_ByteArray nVerticalMask; | |
| if (!nVerticalMask.SetSize(nPageHeight)) { | |
| return -1; | |
| } | |
| uint8_t* pDataV = nVerticalMask.GetData(); | |
| int32_t index = 0; | |
| FX_FLOAT fLineHeight = 0.0f; | |
| CPDF_PageObject* pPageObj = NULL; | |
| FX_POSITION pos = NULL; | |
| pos = m_pPage->GetFirstObjectPosition(); | |
| if (!pos) { | |
| return -1; | |
| } | |
| while (pos) { | |
| pPageObj = m_pPage->GetNextObject(pos); | |
| if (NULL == pPageObj) { | |
| continue; | |
| } | |
| if (PDFPAGE_TEXT != pPageObj->m_Type) { | |
| continue; | |
| } | |
| int32_t minH = | |
| (int32_t)pPageObj->m_Left < 0 ? 0 : (int32_t)pPageObj->m_Left; | |
| int32_t maxH = (int32_t)pPageObj->m_Right > nPageWidth | |
| ? nPageWidth | |
| : (int32_t)pPageObj->m_Right; | |
| int32_t minV = | |
| (int32_t)pPageObj->m_Bottom < 0 ? 0 : (int32_t)pPageObj->m_Bottom; | |
| int32_t maxV = (int32_t)pPageObj->m_Top > nPageHeight | |
| ? nPageHeight | |
| : (int32_t)pPageObj->m_Top; | |
| if (minH >= maxH || minV >= maxV) { | |
| continue; | |
| } | |
| FXSYS_memset(pDataH + minH, 1, maxH - minH); | |
| FXSYS_memset(pDataV + minV, 1, maxV - minV); | |
| if (fLineHeight <= 0.0f) { | |
| fLineHeight = pPageObj->m_Top - pPageObj->m_Bottom; | |
| } | |
| pPageObj = NULL; | |
| } | |
| int32_t nStartH = 0; | |
| int32_t nEndH = 0; | |
| FX_FLOAT nSumH = 0.0f; | |
| for (index = 0; index < nPageWidth; index++) | |
| if (1 == nHorizontalMask[index]) { | |
| break; | |
| } | |
| nStartH = index; | |
| for (index = nPageWidth; index > 0; index--) | |
| if (1 == nHorizontalMask[index - 1]) { | |
| break; | |
| } | |
| nEndH = index; | |
| for (index = nStartH; index < nEndH; index++) { | |
| nSumH += nHorizontalMask[index]; | |
| } | |
| nSumH /= nEndH - nStartH; | |
| int32_t nStartV = 0; | |
| int32_t nEndV = 0; | |
| FX_FLOAT nSumV = 0.0f; | |
| for (index = 0; index < nPageHeight; index++) | |
| if (1 == nVerticalMask[index]) { | |
| break; | |
| } | |
| nStartV = index; | |
| for (index = nPageHeight; index > 0; index--) | |
| if (1 == nVerticalMask[index - 1]) { | |
| break; | |
| } | |
| nEndV = index; | |
| for (index = nStartV; index < nEndV; index++) { | |
| nSumV += nVerticalMask[index]; | |
| } | |
| nSumV /= nEndV - nStartV; | |
| if ((nEndV - nStartV) < (int32_t)(2 * fLineHeight)) { | |
| return 0; | |
| } | |
| if ((nEndH - nStartH) < (int32_t)(2 * fLineHeight)) { | |
| return 1; | |
| } | |
| if (nSumH > 0.8f) { | |
| return 0; | |
| } | |
| if (nSumH - nSumV > 0.0f) { | |
| return 0; | |
| } | |
| if (nSumV - nSumH > 0.0f) { | |
| return 1; | |
| } | |
| return -1; | |
| } | |
| void CPDF_TextPage::ProcessObject() { | |
| CPDF_PageObject* pPageObj = NULL; | |
| if (!m_pPage) { | |
| return; | |
| } | |
| FX_POSITION pos; | |
| pos = m_pPage->GetFirstObjectPosition(); | |
| if (!pos) { | |
| return; | |
| } | |
| m_TextlineDir = FindTextlineFlowDirection(); | |
| int nCount = 0; | |
| while (pos) { | |
| pPageObj = m_pPage->GetNextObject(pos); | |
| if (pPageObj) { | |
| if (pPageObj->m_Type == PDFPAGE_TEXT) { | |
| CFX_AffineMatrix matrix; | |
| ProcessTextObject((CPDF_TextObject*)pPageObj, matrix, pos); | |
| nCount++; | |
| } else if (pPageObj->m_Type == PDFPAGE_FORM) { | |
| CFX_AffineMatrix formMatrix(1, 0, 0, 1, 0, 0); | |
| ProcessFormObject((CPDF_FormObject*)pPageObj, formMatrix); | |
| } | |
| } | |
| pPageObj = NULL; | |
| } | |
| int count = m_LineObj.GetSize(); | |
| for (int i = 0; i < count; i++) { | |
| ProcessTextObject(m_LineObj.GetAt(i)); | |
| } | |
| m_LineObj.RemoveAll(); | |
| CloseTempLine(); | |
| } | |
| void CPDF_TextPage::ProcessFormObject(CPDF_FormObject* pFormObj, | |
| const CFX_AffineMatrix& formMatrix) { | |
| CPDF_PageObject* pPageObj = NULL; | |
| FX_POSITION pos; | |
| if (!pFormObj) { | |
| return; | |
| } | |
| pos = pFormObj->m_pForm->GetFirstObjectPosition(); | |
| if (!pos) { | |
| return; | |
| } | |
| CFX_AffineMatrix curFormMatrix; | |
| curFormMatrix.Copy(pFormObj->m_FormMatrix); | |
| curFormMatrix.Concat(formMatrix); | |
| while (pos) { | |
| pPageObj = pFormObj->m_pForm->GetNextObject(pos); | |
| if (pPageObj) { | |
| if (pPageObj->m_Type == PDFPAGE_TEXT) { | |
| ProcessTextObject((CPDF_TextObject*)pPageObj, curFormMatrix, pos); | |
| } else if (pPageObj->m_Type == PDFPAGE_FORM) { | |
| ProcessFormObject((CPDF_FormObject*)pPageObj, curFormMatrix); | |
| } | |
| } | |
| pPageObj = NULL; | |
| } | |
| } | |
| int CPDF_TextPage::GetCharWidth(FX_DWORD charCode, CPDF_Font* pFont) const { | |
| if (charCode == -1) { | |
| return 0; | |
| } | |
| int w = pFont->GetCharWidthF(charCode); | |
| if (w == 0) { | |
| CFX_ByteString str; | |
| pFont->AppendChar(str, charCode); | |
| w = pFont->GetStringWidth(str, 1); | |
| if (w == 0) { | |
| FX_RECT BBox; | |
| pFont->GetCharBBox(charCode, BBox); | |
| w = BBox.right - BBox.left; | |
| } | |
| } | |
| return w; | |
| } | |
| void CPDF_TextPage::OnPiece(CFX_BidiChar* pBidi, CFX_WideString& str) { | |
| int32_t start, count; | |
| CFX_BidiChar::Direction ret = pBidi->GetBidiInfo(&start, &count); | |
| if (ret == CFX_BidiChar::RIGHT) { | |
| for (int i = start + count - 1; i >= start; i--) { | |
| m_TextBuf.AppendChar(str.GetAt(i)); | |
| m_charList.Add(*(PAGECHAR_INFO*)m_TempCharList.GetAt(i)); | |
| } | |
| } else { | |
| int end = start + count; | |
| for (int i = start; i < end; i++) { | |
| m_TextBuf.AppendChar(str.GetAt(i)); | |
| m_charList.Add(*(PAGECHAR_INFO*)m_TempCharList.GetAt(i)); | |
| } | |
| } | |
| } | |
| void CPDF_TextPage::AddCharInfoByLRDirection(CFX_WideString& str, int i) { | |
| PAGECHAR_INFO Info = *(PAGECHAR_INFO*)m_TempCharList.GetAt(i); | |
| FX_WCHAR wChar = str.GetAt(i); | |
| if (!IsControlChar(Info)) { | |
| Info.m_Index = m_TextBuf.GetLength(); | |
| if (wChar >= 0xFB00 && wChar <= 0xFB06) { | |
| FX_WCHAR* pDst = NULL; | |
| FX_STRSIZE nCount = FX_Unicode_GetNormalization(wChar, pDst); | |
| if (nCount >= 1) { | |
| pDst = FX_Alloc(FX_WCHAR, nCount); | |
| FX_Unicode_GetNormalization(wChar, pDst); | |
| for (int nIndex = 0; nIndex < nCount; nIndex++) { | |
| PAGECHAR_INFO Info2 = Info; | |
| Info2.m_Unicode = pDst[nIndex]; | |
| Info2.m_Flag = FPDFTEXT_CHAR_PIECE; | |
| m_TextBuf.AppendChar(Info2.m_Unicode); | |
| if (!m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_charList.Add(Info2); | |
| } | |
| } | |
| FX_Free(pDst); | |
| return; | |
| } | |
| } | |
| m_TextBuf.AppendChar(wChar); | |
| } else { | |
| Info.m_Index = -1; | |
| } | |
| if (!m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_charList.Add(Info); | |
| } | |
| } | |
| void CPDF_TextPage::AddCharInfoByRLDirection(CFX_WideString& str, int i) { | |
| PAGECHAR_INFO Info = *(PAGECHAR_INFO*)m_TempCharList.GetAt(i); | |
| if (!IsControlChar(Info)) { | |
| Info.m_Index = m_TextBuf.GetLength(); | |
| FX_WCHAR wChar = FX_GetMirrorChar(str.GetAt(i), TRUE, FALSE); | |
| FX_WCHAR* pDst = NULL; | |
| FX_STRSIZE nCount = FX_Unicode_GetNormalization(wChar, pDst); | |
| if (nCount >= 1) { | |
| pDst = FX_Alloc(FX_WCHAR, nCount); | |
| FX_Unicode_GetNormalization(wChar, pDst); | |
| for (int nIndex = 0; nIndex < nCount; nIndex++) { | |
| PAGECHAR_INFO Info2 = Info; | |
| Info2.m_Unicode = pDst[nIndex]; | |
| Info2.m_Flag = FPDFTEXT_CHAR_PIECE; | |
| m_TextBuf.AppendChar(Info2.m_Unicode); | |
| if (!m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_charList.Add(Info2); | |
| } | |
| } | |
| FX_Free(pDst); | |
| return; | |
| } | |
| Info.m_Unicode = wChar; | |
| m_TextBuf.AppendChar(Info.m_Unicode); | |
| } else { | |
| Info.m_Index = -1; | |
| } | |
| if (!m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_charList.Add(Info); | |
| } | |
| } | |
| void CPDF_TextPage::CloseTempLine() { | |
| int count1 = m_TempCharList.GetSize(); | |
| if (count1 <= 0) { | |
| return; | |
| } | |
| nonstd::unique_ptr<CFX_BidiChar> pBidiChar(new CFX_BidiChar); | |
| CFX_WideString str = m_TempTextBuf.GetWideString(); | |
| CFX_WordArray order; | |
| FX_BOOL bR2L = FALSE; | |
| int32_t start = 0, count = 0; | |
| int nR2L = 0, nL2R = 0; | |
| FX_BOOL bPrevSpace = FALSE; | |
| for (int i = 0; i < str.GetLength(); i++) { | |
| if (str.GetAt(i) == 32) { | |
| if (bPrevSpace) { | |
| m_TempTextBuf.Delete(i, 1); | |
| m_TempCharList.Delete(i); | |
| str.Delete(i); | |
| count1--; | |
| i--; | |
| continue; | |
| } | |
| bPrevSpace = TRUE; | |
| } else { | |
| bPrevSpace = FALSE; | |
| } | |
| if (pBidiChar->AppendChar(str.GetAt(i))) { | |
| CFX_BidiChar::Direction ret = pBidiChar->GetBidiInfo(&start, &count); | |
| order.Add(start); | |
| order.Add(count); | |
| order.Add(ret); | |
| if (!bR2L) { | |
| if (ret == CFX_BidiChar::RIGHT) { | |
| nR2L++; | |
| } else if (ret == CFX_BidiChar::LEFT) { | |
| nL2R++; | |
| } | |
| } | |
| } | |
| } | |
| if (pBidiChar->EndChar()) { | |
| CFX_BidiChar::Direction ret = pBidiChar->GetBidiInfo(&start, &count); | |
| order.Add(start); | |
| order.Add(count); | |
| order.Add(ret); | |
| if (!bR2L) { | |
| if (ret == CFX_BidiChar::RIGHT) { | |
| nR2L++; | |
| } else if (ret == CFX_BidiChar::LEFT) { | |
| nL2R++; | |
| } | |
| } | |
| } | |
| if (nR2L > 0 && nR2L >= nL2R) { | |
| bR2L = TRUE; | |
| } | |
| if (m_parserflag == FPDFTEXT_RLTB || bR2L) { | |
| int count = order.GetSize(); | |
| for (int i = count - 1; i > 0; i -= 3) { | |
| int ret = order.GetAt(i); | |
| int start = order.GetAt(i - 2); | |
| int count1 = order.GetAt(i - 1); | |
| if (ret == 2 || ret == 0) { | |
| for (int j = start + count1 - 1; j >= start; j--) { | |
| AddCharInfoByRLDirection(str, j); | |
| } | |
| } else { | |
| int j = i; | |
| FX_BOOL bSymbol = FALSE; | |
| while (j > 0 && order.GetAt(j) != 2) { | |
| bSymbol = !order.GetAt(j); | |
| j -= 3; | |
| } | |
| int end = start + count1; | |
| int n = 0; | |
| if (bSymbol) { | |
| n = j + 6; | |
| } else { | |
| n = j + 3; | |
| } | |
| if (n >= i) { | |
| for (int m = start; m < end; m++) { | |
| AddCharInfoByLRDirection(str, m); | |
| } | |
| } else { | |
| j = i; | |
| i = n; | |
| for (; n <= j; n += 3) { | |
| int start = order.GetAt(n - 2); | |
| int count1 = order.GetAt(n - 1); | |
| int end = start + count1; | |
| for (int m = start; m < end; m++) { | |
| AddCharInfoByLRDirection(str, m); | |
| } | |
| } | |
| } | |
| } | |
| } | |
| } else { | |
| int count = order.GetSize(); | |
| FX_BOOL bL2R = FALSE; | |
| for (int i = 0; i < count; i += 3) { | |
| int ret = order.GetAt(i + 2); | |
| int start = order.GetAt(i); | |
| int count1 = order.GetAt(i + 1); | |
| if (ret == 2 || (i == 0 && ret == 0 && !bL2R)) { | |
| int j = i + 3; | |
| while (bR2L && j < count) { | |
| if (order.GetAt(j + 2) == 1) { | |
| break; | |
| } else { | |
| j += 3; | |
| } | |
| } | |
| if (j == 3) { | |
| i = -3; | |
| bL2R = TRUE; | |
| continue; | |
| } | |
| int end = m_TempCharList.GetSize() - 1; | |
| if (j < count) { | |
| end = order.GetAt(j) - 1; | |
| } | |
| i = j - 3; | |
| for (int n = end; n >= start; n--) { | |
| AddCharInfoByRLDirection(str, n); | |
| } | |
| } else { | |
| int end = start + count1; | |
| for (int n = start; n < end; n++) { | |
| AddCharInfoByLRDirection(str, n); | |
| } | |
| } | |
| } | |
| } | |
| order.RemoveAll(); | |
| m_TempCharList.RemoveAll(); | |
| m_TempTextBuf.Delete(0, m_TempTextBuf.GetLength()); | |
| } | |
| void CPDF_TextPage::ProcessTextObject(CPDF_TextObject* pTextObj, | |
| const CFX_AffineMatrix& formMatrix, | |
| FX_POSITION ObjPos) { | |
| CFX_FloatRect re(pTextObj->m_Left, pTextObj->m_Bottom, pTextObj->m_Right, | |
| pTextObj->m_Top); | |
| if (FXSYS_fabs(pTextObj->m_Right - pTextObj->m_Left) < 0.01f) { | |
| return; | |
| } | |
| int count = m_LineObj.GetSize(); | |
| PDFTEXT_Obj Obj; | |
| Obj.m_pTextObj = pTextObj; | |
| Obj.m_formMatrix = formMatrix; | |
| if (count == 0) { | |
| m_LineObj.Add(Obj); | |
| return; | |
| } | |
| if (IsSameAsPreTextObject(pTextObj, ObjPos)) { | |
| return; | |
| } | |
| PDFTEXT_Obj prev_Obj = m_LineObj.GetAt(count - 1); | |
| CPDF_TextObjectItem item; | |
| int nItem = prev_Obj.m_pTextObj->CountItems(); | |
| prev_Obj.m_pTextObj->GetItemInfo(nItem - 1, &item); | |
| FX_FLOAT prev_width = | |
| GetCharWidth(item.m_CharCode, prev_Obj.m_pTextObj->GetFont()) * | |
| prev_Obj.m_pTextObj->GetFontSize() / 1000; | |
| CFX_AffineMatrix prev_matrix; | |
| prev_Obj.m_pTextObj->GetTextMatrix(&prev_matrix); | |
| prev_width = FXSYS_fabs(prev_width); | |
| prev_matrix.Concat(prev_Obj.m_formMatrix); | |
| prev_width = prev_matrix.TransformDistance(prev_width); | |
| pTextObj->GetItemInfo(0, &item); | |
| FX_FLOAT this_width = GetCharWidth(item.m_CharCode, pTextObj->GetFont()) * | |
| pTextObj->GetFontSize() / 1000; | |
| this_width = FXSYS_fabs(this_width); | |
| CFX_AffineMatrix this_matrix; | |
| pTextObj->GetTextMatrix(&this_matrix); | |
| this_width = FXSYS_fabs(this_width); | |
| this_matrix.Concat(formMatrix); | |
| this_width = this_matrix.TransformDistance(this_width); | |
| FX_FLOAT threshold = | |
| prev_width > this_width ? prev_width / 4 : this_width / 4; | |
| FX_FLOAT prev_x = prev_Obj.m_pTextObj->GetPosX(), | |
| prev_y = prev_Obj.m_pTextObj->GetPosY(); | |
| prev_Obj.m_formMatrix.Transform(prev_x, prev_y); | |
| m_DisplayMatrix.Transform(prev_x, prev_y); | |
| FX_FLOAT this_x = pTextObj->GetPosX(), this_y = pTextObj->GetPosY(); | |
| formMatrix.Transform(this_x, this_y); | |
| m_DisplayMatrix.Transform(this_x, this_y); | |
| if (FXSYS_fabs(this_y - prev_y) > threshold * 2) { | |
| for (int i = 0; i < count; i++) { | |
| ProcessTextObject(m_LineObj.GetAt(i)); | |
| } | |
| m_LineObj.RemoveAll(); | |
| m_LineObj.Add(Obj); | |
| return; | |
| } | |
| int i = 0; | |
| if (m_ParseOptions.m_bNormalizeObjs) { | |
| for (i = count - 1; i >= 0; i--) { | |
| PDFTEXT_Obj prev_Obj = m_LineObj.GetAt(i); | |
| CFX_AffineMatrix prev_matrix; | |
| prev_Obj.m_pTextObj->GetTextMatrix(&prev_matrix); | |
| FX_FLOAT Prev_x = prev_Obj.m_pTextObj->GetPosX(), | |
| Prev_y = prev_Obj.m_pTextObj->GetPosY(); | |
| prev_Obj.m_formMatrix.Transform(Prev_x, Prev_y); | |
| m_DisplayMatrix.Transform(Prev_x, Prev_y); | |
| if (this_x >= Prev_x) { | |
| if (i == count - 1) { | |
| m_LineObj.Add(Obj); | |
| } else { | |
| m_LineObj.InsertAt(i + 1, Obj); | |
| } | |
| break; | |
| } | |
| } | |
| if (i < 0) { | |
| m_LineObj.InsertAt(0, Obj); | |
| } | |
| } else { | |
| m_LineObj.Add(Obj); | |
| } | |
| } | |
| int32_t CPDF_TextPage::PreMarkedContent(PDFTEXT_Obj Obj) { | |
| CPDF_TextObject* pTextObj = Obj.m_pTextObj; | |
| CPDF_ContentMarkData* pMarkData = | |
| (CPDF_ContentMarkData*)pTextObj->m_ContentMark.GetObject(); | |
| if (!pMarkData) { | |
| return FPDFTEXT_MC_PASS; | |
| } | |
| int nContentMark = pMarkData->CountItems(); | |
| if (nContentMark < 1) { | |
| return FPDFTEXT_MC_PASS; | |
| } | |
| CFX_WideString actText; | |
| FX_BOOL bExist = FALSE; | |
| CPDF_Dictionary* pDict = NULL; | |
| int n = 0; | |
| for (n = 0; n < nContentMark; n++) { | |
| CPDF_ContentMarkItem& item = pMarkData->GetItem(n); | |
| CFX_ByteString tagStr = (CFX_ByteString)item.GetName(); | |
| pDict = (CPDF_Dictionary*)item.GetParam(); | |
| CPDF_String* temp = | |
| (CPDF_String*)(pDict ? pDict->GetElement(FX_BSTRC("ActualText")) | |
| : NULL); | |
| if (temp) { | |
| bExist = TRUE; | |
| actText = temp->GetUnicodeText(); | |
| } | |
| } | |
| if (!bExist) { | |
| return FPDFTEXT_MC_PASS; | |
| } | |
| if (m_pPreTextObj) { | |
| if (CPDF_ContentMarkData* pPreMarkData = | |
| (CPDF_ContentMarkData*)m_pPreTextObj->m_ContentMark.GetObject()) { | |
| if (pPreMarkData->CountItems() == n) { | |
| CPDF_ContentMarkItem& item = pPreMarkData->GetItem(n - 1); | |
| if (pDict == item.GetParam()) { | |
| return FPDFTEXT_MC_DONE; | |
| } | |
| } | |
| } | |
| } | |
| CPDF_Font* pFont = pTextObj->GetFont(); | |
| FX_STRSIZE nItems = actText.GetLength(); | |
| if (nItems < 1) { | |
| return FPDFTEXT_MC_PASS; | |
| } | |
| bExist = FALSE; | |
| for (FX_STRSIZE i = 0; i < nItems; i++) { | |
| FX_WCHAR wChar = actText.GetAt(i); | |
| if (-1 == pFont->CharCodeFromUnicode(wChar)) { | |
| continue; | |
| } else { | |
| bExist = TRUE; | |
| break; | |
| } | |
| } | |
| if (!bExist) { | |
| return FPDFTEXT_MC_PASS; | |
| } | |
| bExist = FALSE; | |
| for (FX_STRSIZE i = 0; i < nItems; i++) { | |
| FX_WCHAR wChar = actText.GetAt(i); | |
| if ((wChar > 0x80 && wChar < 0xFFFD) || (wChar <= 0x80 && isprint(wChar))) { | |
| bExist = TRUE; | |
| break; | |
| } | |
| } | |
| if (!bExist) { | |
| return FPDFTEXT_MC_DONE; | |
| } | |
| return FPDFTEXT_MC_DELAY; | |
| } | |
| void CPDF_TextPage::ProcessMarkedContent(PDFTEXT_Obj Obj) { | |
| CPDF_TextObject* pTextObj = Obj.m_pTextObj; | |
| CPDF_ContentMarkData* pMarkData = | |
| (CPDF_ContentMarkData*)pTextObj->m_ContentMark.GetObject(); | |
| if (!pMarkData) { | |
| return; | |
| } | |
| int nContentMark = pMarkData->CountItems(); | |
| if (nContentMark < 1) { | |
| return; | |
| } | |
| CFX_WideString actText; | |
| CPDF_Dictionary* pDict = NULL; | |
| int n = 0; | |
| for (n = 0; n < nContentMark; n++) { | |
| CPDF_ContentMarkItem& item = pMarkData->GetItem(n); | |
| CFX_ByteString tagStr = (CFX_ByteString)item.GetName(); | |
| pDict = (CPDF_Dictionary*)item.GetParam(); | |
| CPDF_String* temp = | |
| (CPDF_String*)(pDict ? pDict->GetElement(FX_BSTRC("ActualText")) | |
| : NULL); | |
| if (temp) { | |
| actText = temp->GetUnicodeText(); | |
| } | |
| } | |
| FX_STRSIZE nItems = actText.GetLength(); | |
| if (nItems < 1) { | |
| return; | |
| } | |
| CPDF_Font* pFont = pTextObj->GetFont(); | |
| CFX_AffineMatrix formMatrix = Obj.m_formMatrix; | |
| CFX_AffineMatrix matrix; | |
| pTextObj->GetTextMatrix(&matrix); | |
| matrix.Concat(formMatrix); | |
| FX_FLOAT fPosX = pTextObj->GetPosX(); | |
| FX_FLOAT fPosY = pTextObj->GetPosY(); | |
| int nCharInfoIndex = m_TextBuf.GetLength(); | |
| CFX_FloatRect charBox; | |
| charBox.top = pTextObj->m_Top; | |
| charBox.left = pTextObj->m_Left; | |
| charBox.right = pTextObj->m_Right; | |
| charBox.bottom = pTextObj->m_Bottom; | |
| for (FX_STRSIZE k = 0; k < nItems; k++) { | |
| FX_WCHAR wChar = actText.GetAt(k); | |
| if (wChar <= 0x80 && !isprint(wChar)) { | |
| wChar = 0x20; | |
| } | |
| if (wChar >= 0xFFFD) { | |
| continue; | |
| } | |
| PAGECHAR_INFO charinfo; | |
| charinfo.m_OriginX = fPosX; | |
| charinfo.m_OriginY = fPosY; | |
| charinfo.m_Index = nCharInfoIndex; | |
| charinfo.m_Unicode = wChar; | |
| charinfo.m_CharCode = pFont->CharCodeFromUnicode(wChar); | |
| charinfo.m_Flag = FPDFTEXT_CHAR_PIECE; | |
| charinfo.m_pTextObj = pTextObj; | |
| charinfo.m_CharBox.top = charBox.top; | |
| charinfo.m_CharBox.left = charBox.left; | |
| charinfo.m_CharBox.right = charBox.right; | |
| charinfo.m_CharBox.bottom = charBox.bottom; | |
| charinfo.m_Matrix.Copy(matrix); | |
| m_TempTextBuf.AppendChar(wChar); | |
| m_TempCharList.Add(charinfo); | |
| } | |
| } | |
| void CPDF_TextPage::FindPreviousTextObject(void) { | |
| if (m_TempCharList.GetSize() < 1 && m_charList.GetSize() < 1) { | |
| return; | |
| } | |
| PAGECHAR_INFO preChar; | |
| if (m_TempCharList.GetSize() >= 1) { | |
| preChar = | |
| *(PAGECHAR_INFO*)m_TempCharList.GetAt(m_TempCharList.GetSize() - 1); | |
| } else { | |
| preChar = *(PAGECHAR_INFO*)m_charList.GetAt(m_charList.GetSize() - 1); | |
| } | |
| if (preChar.m_pTextObj) { | |
| m_pPreTextObj = preChar.m_pTextObj; | |
| } | |
| } | |
| void CPDF_TextPage::ProcessTextObject(PDFTEXT_Obj Obj) { | |
| CPDF_TextObject* pTextObj = Obj.m_pTextObj; | |
| if (FXSYS_fabs(pTextObj->m_Right - pTextObj->m_Left) < 0.01f) { | |
| return; | |
| } | |
| CFX_AffineMatrix formMatrix = Obj.m_formMatrix; | |
| CPDF_Font* pFont = pTextObj->GetFont(); | |
| CFX_AffineMatrix matrix; | |
| pTextObj->GetTextMatrix(&matrix); | |
| matrix.Concat(formMatrix); | |
| int32_t bPreMKC = PreMarkedContent(Obj); | |
| if (FPDFTEXT_MC_DONE == bPreMKC) { | |
| m_pPreTextObj = pTextObj; | |
| m_perMatrix.Copy(formMatrix); | |
| return; | |
| } | |
| int result = 0; | |
| if (m_pPreTextObj) { | |
| result = ProcessInsertObject(pTextObj, formMatrix); | |
| if (2 == result) { | |
| m_CurlineRect = | |
| CFX_FloatRect(Obj.m_pTextObj->m_Left, Obj.m_pTextObj->m_Bottom, | |
| Obj.m_pTextObj->m_Right, Obj.m_pTextObj->m_Top); | |
| } else { | |
| m_CurlineRect.Union( | |
| CFX_FloatRect(Obj.m_pTextObj->m_Left, Obj.m_pTextObj->m_Bottom, | |
| Obj.m_pTextObj->m_Right, Obj.m_pTextObj->m_Top)); | |
| } | |
| PAGECHAR_INFO generateChar; | |
| if (result == 1) { | |
| if (GenerateCharInfo(TEXT_BLANK_CHAR, generateChar)) { | |
| if (!formMatrix.IsIdentity()) { | |
| generateChar.m_Matrix.Copy(formMatrix); | |
| } | |
| m_TempTextBuf.AppendChar(TEXT_BLANK_CHAR); | |
| m_TempCharList.Add(generateChar); | |
| } | |
| } else if (result == 2) { | |
| CloseTempLine(); | |
| if (m_TextBuf.GetSize()) { | |
| if (m_ParseOptions.m_bGetCharCodeOnly) { | |
| m_TextBuf.AppendChar(TEXT_RETURN_CHAR); | |
| m_TextBuf.AppendChar(TEXT_LINEFEED_CHAR); | |
| } else { | |
| if (GenerateCharInfo(TEXT_RETURN_CHAR, generateChar)) { | |
| m_TextBuf.AppendChar(TEXT_RETURN_CHAR); | |
| if (!formMatrix.IsIdentity()) { | |
| generateChar.m_Matrix.Copy(formMatrix); | |
| } | |
| m_charList.Add(generateChar); | |
| } | |
| if (GenerateCharInfo(TEXT_LINEFEED_CHAR, generateChar)) { | |
| m_TextBuf.AppendChar(TEXT_LINEFEED_CHAR); | |
| if (!formMatrix.IsIdentity()) { | |
| generateChar.m_Matrix.Copy(formMatrix); | |
| } | |
| m_charList.Add(generateChar); | |
| } | |
| } | |
| } | |
| } else if (result == 3 && !m_ParseOptions.m_bOutputHyphen) { | |
| int32_t nChars = pTextObj->CountChars(); | |
| if (nChars == 1) { | |
| CPDF_TextObjectItem item; | |
| pTextObj->GetCharInfo(0, &item); | |
| CFX_WideString wstrItem = | |
| pTextObj->GetFont()->UnicodeFromCharCode(item.m_CharCode); | |
| if (wstrItem.IsEmpty()) { | |
| wstrItem += (FX_WCHAR)item.m_CharCode; | |
| } | |
| FX_WCHAR curChar = wstrItem.GetAt(0); | |
| if (0x2D == curChar || 0xAD == curChar) { | |
| return; | |
| } | |
| } | |
| while (m_TempTextBuf.GetSize() > 0 && | |
| m_TempTextBuf.GetWideString().GetAt(m_TempTextBuf.GetLength() - | |
| 1) == 0x20) { | |
| m_TempTextBuf.Delete(m_TempTextBuf.GetLength() - 1, 1); | |
| m_TempCharList.Delete(m_TempCharList.GetSize() - 1); | |
| } | |
| PAGECHAR_INFO* cha = | |
| (PAGECHAR_INFO*)m_TempCharList.GetAt(m_TempCharList.GetSize() - 1); | |
| m_TempTextBuf.Delete(m_TempTextBuf.GetLength() - 1, 1); | |
| cha->m_Unicode = 0x2; | |
| cha->m_Flag = FPDFTEXT_CHAR_HYPHEN; | |
| m_TempTextBuf.AppendChar(0xfffe); | |
| } | |
| } else { | |
| m_CurlineRect = | |
| CFX_FloatRect(Obj.m_pTextObj->m_Left, Obj.m_pTextObj->m_Bottom, | |
| Obj.m_pTextObj->m_Right, Obj.m_pTextObj->m_Top); | |
| } | |
| if (FPDFTEXT_MC_DELAY == bPreMKC) { | |
| ProcessMarkedContent(Obj); | |
| m_pPreTextObj = pTextObj; | |
| m_perMatrix.Copy(formMatrix); | |
| return; | |
| } | |
| m_pPreTextObj = pTextObj; | |
| m_perMatrix.Copy(formMatrix); | |
| int nItems = pTextObj->CountItems(); | |
| FX_FLOAT baseSpace = _CalculateBaseSpace(pTextObj, matrix); | |
| const FX_BOOL bR2L = IsRightToLeft(pTextObj, pFont, nItems); | |
| const FX_BOOL bIsBidiAndMirrorInverse = | |
| bR2L && (matrix.a * matrix.d - matrix.b * matrix.c) < 0; | |
| int32_t iBufStartAppend = m_TempTextBuf.GetLength(); | |
| int32_t iCharListStartAppend = m_TempCharList.GetSize(); | |
| FX_FLOAT spacing = 0; | |
| 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(); | |
| } | |
| if (str.IsEmpty() || str.GetAt(str.GetLength() - 1) == TEXT_BLANK_CHAR) { | |
| continue; | |
| } | |
| FX_FLOAT fontsize_h = pTextObj->m_TextState.GetFontSizeH(); | |
| spacing = -fontsize_h * item.m_OriginX / 1000; | |
| continue; | |
| } | |
| FX_FLOAT charSpace = pTextObj->m_TextState.GetObject()->m_CharSpace; | |
| if (charSpace > 0.001) { | |
| spacing += matrix.TransformDistance(charSpace); | |
| } else if (charSpace < -0.001) { | |
| spacing -= matrix.TransformDistance(FXSYS_fabs(charSpace)); | |
| } | |
| spacing -= baseSpace; | |
| if (spacing && i > 0) { | |
| int last_width = 0; | |
| FX_FLOAT fontsize_h = pTextObj->m_TextState.GetFontSizeH(); | |
| FX_DWORD space_charcode = pFont->CharCodeFromUnicode(' '); | |
| FX_FLOAT threshold = 0; | |
| if (space_charcode != -1) { | |
| threshold = fontsize_h * pFont->GetCharWidthF(space_charcode) / 1000; | |
| } | |
| if (threshold > fontsize_h / 3) { | |
| threshold = 0; | |
| } else { | |
| threshold /= 2; | |
| } | |
| if (threshold == 0) { | |
| threshold = fontsize_h; | |
| int this_width = FXSYS_abs(GetCharWidth(item.m_CharCode, pFont)); | |
| threshold = this_width > last_width ? (FX_FLOAT)this_width | |
| : (FX_FLOAT)last_width; | |
| threshold = _NormalizeThreshold(threshold); | |
| threshold = fontsize_h * threshold / 1000; | |
| } | |
| if (threshold && (spacing && spacing >= threshold)) { | |
| charinfo.m_Unicode = TEXT_BLANK_CHAR; | |
| charinfo.m_Flag = FPDFTEXT_CHAR_GENERATED; | |
| charinfo.m_pTextObj = pTextObj; | |
| charinfo.m_Index = m_TextBuf.GetLength(); | |
| m_TempTextBuf.AppendChar(TEXT_BLANK_CHAR); | |
| charinfo.m_CharCode = -1; | |
| charinfo.m_Matrix.Copy(formMatrix); | |
| matrix.Transform(item.m_OriginX, item.m_OriginY, charinfo.m_OriginX, | |
| charinfo.m_OriginY); | |
| charinfo.m_CharBox = | |
| CFX_FloatRect(charinfo.m_OriginX, charinfo.m_OriginY, | |
| charinfo.m_OriginX, charinfo.m_OriginY); | |
| m_TempCharList.Add(charinfo); | |
| } | |
| if (item.m_CharCode == (FX_DWORD)-1) { | |
| continue; | |
| } | |
| } | |
| spacing = 0; | |
| CFX_WideString wstrItem = pFont->UnicodeFromCharCode(item.m_CharCode); | |
| FX_BOOL bNoUnicode = FALSE; | |
| FX_WCHAR wChar = wstrItem.GetAt(0); | |
| if ((wstrItem.IsEmpty() || wChar == 0) && item.m_CharCode) { | |
| if (wstrItem.IsEmpty()) { | |
| wstrItem += (FX_WCHAR)item.m_CharCode; | |
| } else { | |
| wstrItem.SetAt(0, (FX_WCHAR)item.m_CharCode); | |
| } | |
| bNoUnicode = TRUE; | |
| } | |
| charinfo.m_Index = -1; | |
| charinfo.m_CharCode = item.m_CharCode; | |
| if (bNoUnicode) { | |
| charinfo.m_Flag = FPDFTEXT_CHAR_UNUNICODE; | |
| } else { | |
| charinfo.m_Flag = FPDFTEXT_CHAR_NORMAL; | |
| } | |
| charinfo.m_pTextObj = pTextObj; | |
| charinfo.m_OriginX = 0, charinfo.m_OriginY = 0; | |
| matrix.Transform(item.m_OriginX, item.m_OriginY, charinfo.m_OriginX, | |
| charinfo.m_OriginY); | |
| FX_RECT rect(0, 0, 0, 0); | |
| rect.Intersect(0, 0, 0, 0); | |
| charinfo.m_pTextObj->GetFont()->GetCharBBox(charinfo.m_CharCode, rect); | |
| charinfo.m_CharBox.top = | |
| rect.top * pTextObj->GetFontSize() / 1000 + item.m_OriginY; | |
| charinfo.m_CharBox.left = | |
| rect.left * pTextObj->GetFontSize() / 1000 + item.m_OriginX; | |
| charinfo.m_CharBox.right = | |
| rect.right * pTextObj->GetFontSize() / 1000 + item.m_OriginX; | |
| charinfo.m_CharBox.bottom = | |
| rect.bottom * pTextObj->GetFontSize() / 1000 + item.m_OriginY; | |
| if (fabsf(charinfo.m_CharBox.top - charinfo.m_CharBox.bottom) < 0.01f) { | |
| charinfo.m_CharBox.top = | |
| charinfo.m_CharBox.bottom + pTextObj->GetFontSize(); | |
| } | |
| if (fabsf(charinfo.m_CharBox.right - charinfo.m_CharBox.left) < 0.01f) { | |
| charinfo.m_CharBox.right = | |
| charinfo.m_CharBox.left + pTextObj->GetCharWidth(charinfo.m_CharCode); | |
| } | |
| matrix.TransformRect(charinfo.m_CharBox); | |
| charinfo.m_Matrix.Copy(matrix); | |
| if (wstrItem.IsEmpty()) { | |
| charinfo.m_Unicode = 0; | |
| m_TempCharList.Add(charinfo); | |
| m_TempTextBuf.AppendChar(0xfffe); | |
| continue; | |
| } else { | |
| int nTotal = wstrItem.GetLength(); | |
| FX_BOOL bDel = FALSE; | |
| const int count = std::min(m_TempCharList.GetSize(), 7); | |
| FX_FLOAT threshold = charinfo.m_Matrix.TransformXDistance( | |
| (FX_FLOAT)TEXT_CHARRATIO_GAPDELTA * pTextObj->GetFontSize()); | |
| for (int n = m_TempCharList.GetSize(); | |
| n > m_TempCharList.GetSize() - count; n--) { | |
| PAGECHAR_INFO* charinfo1 = (PAGECHAR_INFO*)m_TempCharList.GetAt(n - 1); | |
| if (charinfo1->m_CharCode == charinfo.m_CharCode && | |
| charinfo1->m_pTextObj->GetFont() == | |
| charinfo.m_pTextObj->GetFont() && | |
| FXSYS_fabs(charinfo1->m_OriginX - charinfo.m_OriginX) < threshold && | |
| FXSYS_fabs(charinfo1->m_OriginY - charinfo.m_OriginY) < threshold) { | |
| bDel = TRUE; | |
| break; | |
| } | |
| } | |
| if (!bDel) { | |
| for (int nIndex = 0; nIndex < nTotal; nIndex++) { | |
| charinfo.m_Unicode = wstrItem.GetAt(nIndex); | |
| if (charinfo.m_Unicode) { | |
| charinfo.m_Index = m_TextBuf.GetLength(); | |
| m_TempTextBuf.AppendChar(charinfo.m_Unicode); | |
| } else { | |
| m_TempTextBuf.AppendChar(0xfffe); | |
| } | |
| m_TempCharList.Add(charinfo); | |
| } | |
| } 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 (bIsBidiAndMirrorInverse) { | |
| SwapTempTextBuf(iCharListStartAppend, iBufStartAppend); | |
| } | |
| } | |
| void CPDF_TextPage::SwapTempTextBuf(int32_t iCharListStartAppend, | |
| int32_t iBufStartAppend) { | |
| int32_t i, j; | |
| i = iCharListStartAppend; | |
| j = m_TempCharList.GetSize() - 1; | |
| for (; i < j; i++, j--) { | |
| std::swap(m_TempCharList[i], m_TempCharList[j]); | |
| std::swap(m_TempCharList[i].m_Index, m_TempCharList[j].m_Index); | |
| } | |
| FX_WCHAR* pTempBuffer = m_TempTextBuf.GetBuffer(); | |
| i = iBufStartAppend; | |
| j = m_TempTextBuf.GetLength() - 1; | |
| for (; i < j; i++, j--) { | |
| std::swap(pTempBuffer[i], pTempBuffer[j]); | |
| } | |
| } | |
| FX_BOOL CPDF_TextPage::IsRightToLeft(const CPDF_TextObject* pTextObj, | |
| const CPDF_Font* pFont, | |
| int nItems) const { | |
| nonstd::unique_ptr<CFX_BidiChar> pBidiChar(new CFX_BidiChar); | |
| int32_t nR2L = 0; | |
| int32_t nL2R = 0; | |
| int32_t start = 0, count = 0; | |
| CPDF_TextObjectItem item; | |
| for (int32_t 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) { | |
| wChar = (FX_WCHAR)item.m_CharCode; | |
| } | |
| if (!wChar) { | |
| continue; | |
| } | |
| if (pBidiChar->AppendChar(wChar)) { | |
| CFX_BidiChar::Direction ret = pBidiChar->GetBidiInfo(&start, &count); | |
| if (ret == CFX_BidiChar::RIGHT) { | |
| nR2L++; | |
| } else if (ret == CFX_BidiChar::LEFT) { | |
| nL2R++; | |
| } | |
| } | |
| } | |
| if (pBidiChar->EndChar()) { | |
| CFX_BidiChar::Direction ret = pBidiChar->GetBidiInfo(&start, &count); | |
| if (ret == CFX_BidiChar::RIGHT) { | |
| nR2L++; | |
| } else if (ret == CFX_BidiChar::LEFT) { | |
| nL2R++; | |
| } | |
| } | |
| return (nR2L > 0 && nR2L >= nL2R); | |
| } | |
| int32_t CPDF_TextPage::GetTextObjectWritingMode( | |
| const CPDF_TextObject* pTextObj) { | |
| int32_t nChars = pTextObj->CountChars(); | |
| if (nChars == 1) { | |
| return m_TextlineDir; | |
| } | |
| CPDF_TextObjectItem first, last; | |
| pTextObj->GetCharInfo(0, &first); | |
| pTextObj->GetCharInfo(nChars - 1, &last); | |
| CFX_Matrix textMatrix; | |
| pTextObj->GetTextMatrix(&textMatrix); | |
| textMatrix.TransformPoint(first.m_OriginX, first.m_OriginY); | |
| textMatrix.TransformPoint(last.m_OriginX, last.m_OriginY); | |
| FX_FLOAT dX = FXSYS_fabs(last.m_OriginX - first.m_OriginX); | |
| FX_FLOAT dY = FXSYS_fabs(last.m_OriginY - first.m_OriginY); | |
| if (dX <= 0.0001f && dY <= 0.0001f) { | |
| return -1; | |
| } | |
| CFX_VectorF v; | |
| v.Set(dX, dY); | |
| v.Normalize(); | |
| if (v.y <= 0.0872f) { | |
| return v.x <= 0.0872f ? m_TextlineDir : 0; | |
| } | |
| if (v.x <= 0.0872f) { | |
| return 1; | |
| } | |
| return m_TextlineDir; | |
| } | |
| FX_BOOL CPDF_TextPage::IsHyphen(FX_WCHAR curChar) { | |
| CFX_WideString strCurText = m_TempTextBuf.GetWideString(); | |
| if (strCurText.GetLength() == 0) { | |
| strCurText = m_TextBuf.GetWideString(); | |
| } | |
| FX_STRSIZE nCount = strCurText.GetLength(); | |
| int nIndex = nCount - 1; | |
| FX_WCHAR wcTmp = strCurText.GetAt(nIndex); | |
| while (wcTmp == 0x20 && nIndex <= nCount - 1 && nIndex >= 0) { | |
| wcTmp = strCurText.GetAt(--nIndex); | |
| } | |
| if (0x2D == wcTmp || 0xAD == wcTmp) { | |
| if (--nIndex > 0) { | |
| FX_WCHAR preChar = strCurText.GetAt((nIndex)); | |
| if (((preChar >= L'A' && preChar <= L'Z') || | |
| (preChar >= L'a' && preChar <= L'z')) && | |
| ((curChar >= L'A' && curChar <= L'Z') || | |
| (curChar >= L'a' && curChar <= L'z'))) { | |
| return TRUE; | |
| } | |
| } | |
| int size = m_TempCharList.GetSize(); | |
| PAGECHAR_INFO preChar; | |
| if (size) { | |
| preChar = (PAGECHAR_INFO)m_TempCharList[size - 1]; | |
| } else { | |
| size = m_charList.GetSize(); | |
| if (size == 0) { | |
| return FALSE; | |
| } | |
| preChar = (PAGECHAR_INFO)m_charList[size - 1]; | |
| } | |
| if (FPDFTEXT_CHAR_PIECE == preChar.m_Flag) | |
| if (0xAD == preChar.m_Unicode || 0x2D == preChar.m_Unicode) { | |
| return TRUE; | |
| } | |
| } | |
| return FALSE; | |
| } | |
| int CPDF_TextPage::ProcessInsertObject(const CPDF_TextObject* pObj, | |
| const CFX_AffineMatrix& formMatrix) { | |
| FindPreviousTextObject(); | |
| FX_BOOL bNewline = FALSE; | |
| int WritingMode = GetTextObjectWritingMode(pObj); | |
| if (WritingMode == -1) { | |
| WritingMode = GetTextObjectWritingMode(m_pPreTextObj); | |
| } | |
| CFX_FloatRect this_rect(pObj->m_Left, pObj->m_Bottom, pObj->m_Right, | |
| pObj->m_Top); | |
| CFX_FloatRect prev_rect(m_pPreTextObj->m_Left, m_pPreTextObj->m_Bottom, | |
| m_pPreTextObj->m_Right, m_pPreTextObj->m_Top); | |
| CPDF_TextObjectItem PrevItem, item; | |
| int nItem = m_pPreTextObj->CountItems(); | |
| m_pPreTextObj->GetItemInfo(nItem - 1, &PrevItem); | |
| pObj->GetItemInfo(0, &item); | |
| CFX_WideString wstrItem = | |
| pObj->GetFont()->UnicodeFromCharCode(item.m_CharCode); | |
| if (wstrItem.IsEmpty()) { | |
| wstrItem += (FX_WCHAR)item.m_CharCode; | |
| } | |
| FX_WCHAR curChar = wstrItem.GetAt(0); | |
| if (WritingMode == 0) { | |
| if (this_rect.Height() > 4.5 && prev_rect.Height() > 4.5) { | |
| FX_FLOAT top = | |
| this_rect.top < prev_rect.top ? this_rect.top : prev_rect.top; | |
| FX_FLOAT bottom = this_rect.bottom > prev_rect.bottom ? this_rect.bottom | |
| : prev_rect.bottom; | |
| if (bottom >= top) { | |
| if (IsHyphen(curChar)) { | |
| return 3; | |
| } | |
| return 2; | |
| } | |
| } | |
| } else if (WritingMode == 1) { | |
| if (this_rect.Width() > pObj->GetFontSize() * 0.1f && | |
| prev_rect.Width() > m_pPreTextObj->GetFontSize() * 0.1f) { | |
| FX_FLOAT left = this_rect.left > m_CurlineRect.left ? this_rect.left | |
| : m_CurlineRect.left; | |
| FX_FLOAT right = this_rect.right < m_CurlineRect.right | |
| ? this_rect.right | |
| : m_CurlineRect.right; | |
| if (right <= left) { | |
| if (IsHyphen(curChar)) { | |
| return 3; | |
| } | |
| return 2; | |
| } | |
| } | |
| } | |
| FX_FLOAT last_pos = PrevItem.m_OriginX; | |
| int nLastWidth = GetCharWidth(PrevItem.m_CharCode, m_pPreTextObj->GetFont()); | |
| FX_FLOAT last_width = nLastWidth * m_pPreTextObj->GetFontSize() / 1000; | |
| last_width = FXSYS_fabs(last_width); | |
| int nThisWidth = GetCharWidth(item.m_CharCode, pObj->GetFont()); | |
| FX_FLOAT this_width = nThisWidth * pObj->GetFontSize() / 1000; | |
| this_width = FXSYS_fabs(this_width); | |
| FX_FLOAT threshold = | |
| last_width > this_width ? last_width / 4 : this_width / 4; | |
| CFX_AffineMatrix prev_matrix, prev_reverse; | |
| m_pPreTextObj->GetTextMatrix(&prev_matrix); | |
| prev_matrix.Concat(m_perMatrix); | |
| prev_reverse.SetReverse(prev_matrix); | |
| FX_FLOAT x = pObj->GetPosX(); | |
| FX_FLOAT y = pObj->GetPosY(); | |
| formMatrix.Transform(x, y); | |
| prev_reverse.Transform(x, y); | |
| if (last_width < this_width) { | |
| threshold = prev_reverse.TransformDistance(threshold); | |
| } | |
| CFX_FloatRect rect1(m_pPreTextObj->m_Left, pObj->m_Bottom, | |
| m_pPreTextObj->m_Right, pObj->m_Top); | |
| CFX_FloatRect rect2(m_pPreTextObj->m_Left, m_pPreTextObj->m_Bottom, | |
| m_pPreTextObj->m_Right, m_pPreTextObj->m_Top); | |
| CFX_FloatRect rect3 = rect1; | |
| rect1.Intersect(rect2); | |
| if (WritingMode == 0) { | |
| if ((rect1.IsEmpty() && rect2.Height() > 5 && rect3.Height() > 5) || | |
| ((y > threshold * 2 || y < threshold * -3) && | |
| (FXSYS_fabs(y) < 1 ? FXSYS_fabs(x) < FXSYS_fabs(y) : TRUE))) { | |
| bNewline = TRUE; | |
| if (nItem > 1) { | |
| CPDF_TextObjectItem tempItem; | |
| m_pPreTextObj->GetItemInfo(0, &tempItem); | |
| CFX_AffineMatrix m; | |
| m_pPreTextObj->GetTextMatrix(&m); | |
| if (PrevItem.m_OriginX > tempItem.m_OriginX && | |
| m_DisplayMatrix.a > 0.9 && m_DisplayMatrix.b < 0.1 && | |
| m_DisplayMatrix.c < 0.1 && m_DisplayMatrix.d < -0.9 && m.b < 0.1 && | |
| m.c < 0.1) { | |
| CFX_FloatRect re(0, m_pPreTextObj->m_Bottom, 1000, | |
| m_pPreTextObj->m_Top); | |
| if (re.Contains(pObj->GetPosX(), pObj->GetPosY())) { | |
| bNewline = FALSE; | |
| } else { | |
| CFX_FloatRect re(0, pObj->m_Bottom, 1000, pObj->m_Top); | |
| if (re.Contains(m_pPreTextObj->GetPosX(), | |
| m_pPreTextObj->GetPosY())) { | |
| bNewline = FALSE; | |
| } | |
| } | |
| } | |
| } | |
| } | |
| } | |
| if (bNewline) { | |
| if (IsHyphen(curChar)) { | |
| return 3; | |
| } | |
| return 2; | |
| } | |
| int32_t nChars = pObj->CountChars(); | |
| if (nChars == 1 && (0x2D == curChar || 0xAD == curChar)) | |
| if (IsHyphen(curChar)) { | |
| return 3; | |
| } | |
| CFX_WideString PrevStr = | |
| m_pPreTextObj->GetFont()->UnicodeFromCharCode(PrevItem.m_CharCode); | |
| FX_WCHAR preChar = PrevStr.GetAt(PrevStr.GetLength() - 1); | |
| CFX_AffineMatrix matrix; | |
| pObj->GetTextMatrix(&matrix); | |
| matrix.Concat(formMatrix); | |
| threshold = (FX_FLOAT)(nLastWidth > nThisWidth ? nLastWidth : nThisWidth); | |
| threshold = threshold > 400 | |
| ? (threshold < 700 | |
| ? threshold / 4 | |
| : (threshold > 800 ? threshold / 6 : threshold / 5)) | |
| : (threshold / 2); | |
| if (nLastWidth >= nThisWidth) { | |
| threshold *= FXSYS_fabs(m_pPreTextObj->GetFontSize()); | |
| } else { | |
| threshold *= FXSYS_fabs(pObj->GetFontSize()); | |
| threshold = matrix.TransformDistance(threshold); | |
| threshold = prev_reverse.TransformDistance(threshold); | |
| } | |
| threshold /= 1000; | |
| if ((threshold < 1.4881 && threshold > 1.4879) || | |
| (threshold < 1.39001 && threshold > 1.38999)) { | |
| threshold *= 1.5; | |
| } | |
| if (FXSYS_fabs(last_pos + last_width - x) > threshold && curChar != L' ' && | |
| preChar != L' ') | |
| if (curChar != L' ' && preChar != L' ') { | |
| if ((x - last_pos - last_width) > threshold || | |
| (last_pos - x - last_width) > threshold) { | |
| return 1; | |
| } | |
| if (x < 0 && (last_pos - x - last_width) > threshold) { | |
| return 1; | |
| } | |
| if ((x - last_pos - last_width) > this_width || | |
| (x - last_pos - this_width) > last_width) { | |
| return 1; | |
| } | |
| } | |
| return 0; | |
| } | |
| FX_BOOL CPDF_TextPage::IsSameTextObject(CPDF_TextObject* pTextObj1, | |
| CPDF_TextObject* pTextObj2) { | |
| if (!pTextObj1 || !pTextObj2) { | |
| return FALSE; | |
| } | |
| CFX_FloatRect rcPreObj(pTextObj2->m_Left, pTextObj2->m_Bottom, | |
| pTextObj2->m_Right, pTextObj2->m_Top); | |
| CFX_FloatRect rcCurObj(pTextObj1->m_Left, pTextObj1->m_Bottom, | |
| pTextObj1->m_Right, pTextObj1->m_Top); | |
| if (rcPreObj.IsEmpty() && rcCurObj.IsEmpty() && | |
| !m_ParseOptions.m_bGetCharCodeOnly) { | |
| FX_FLOAT dbXdif = FXSYS_fabs(rcPreObj.left - rcCurObj.left); | |
| int nCount = m_charList.GetSize(); | |
| if (nCount >= 2) { | |
| PAGECHAR_INFO perCharTemp = (PAGECHAR_INFO)m_charList[nCount - 2]; | |
| FX_FLOAT dbSpace = perCharTemp.m_CharBox.Width(); | |
| if (dbXdif > dbSpace) { | |
| return FALSE; | |
| } | |
| } | |
| } | |
| if (!rcPreObj.IsEmpty() || !rcCurObj.IsEmpty()) { | |
| rcPreObj.Intersect(rcCurObj); | |
| if (rcPreObj.IsEmpty()) { | |
| return FALSE; | |
| } | |
| if (FXSYS_fabs(rcPreObj.Width() - rcCurObj.Width()) > | |
| rcCurObj.Width() / 2) { | |
| return FALSE; | |
| } | |
| if (pTextObj2->GetFontSize() != pTextObj1->GetFontSize()) { | |
| return FALSE; | |
| } | |
| } | |
| int nPreCount = pTextObj2->CountItems(); | |
| int nCurCount = pTextObj1->CountItems(); | |
| if (nPreCount != nCurCount) { | |
| return FALSE; | |
| } | |
| CPDF_TextObjectItem itemPer, itemCur; | |
| for (int i = 0; i < nPreCount; i++) { | |
| pTextObj2->GetItemInfo(i, &itemPer); | |
| pTextObj1->GetItemInfo(i, &itemCur); | |
| if (itemCur.m_CharCode != itemPer.m_CharCode) { | |
| return FALSE; | |
| } | |
| } | |
| if (FXSYS_fabs(pTextObj1->GetPosX() - pTextObj2->GetPosX()) > | |
| GetCharWidth(itemPer.m_CharCode, pTextObj2->GetFont()) * | |
| pTextObj2->GetFontSize() / 1000 * 0.9 || | |
| FXSYS_fabs(pTextObj1->GetPosY() - pTextObj2->GetPosY()) > | |
| FX_MAX(FX_MAX(rcPreObj.Height(), rcPreObj.Width()), | |
| pTextObj2->GetFontSize()) / | |
| 8) { | |
| return FALSE; | |
| } | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_TextPage::IsSameAsPreTextObject(CPDF_TextObject* pTextObj, | |
| FX_POSITION ObjPos) { | |
| if (!pTextObj) { | |
| return FALSE; | |
| } | |
| int i = 0; | |
| if (!ObjPos) { | |
| ObjPos = m_pPage->GetLastObjectPosition(); | |
| } | |
| CPDF_PageObject* pObj = m_pPage->GetPrevObject(ObjPos); | |
| while (i < 5 && ObjPos) { | |
| pObj = m_pPage->GetPrevObject(ObjPos); | |
| if (pObj == pTextObj) { | |
| continue; | |
| } | |
| if (pObj->m_Type != PDFPAGE_TEXT) { | |
| continue; | |
| } | |
| if (IsSameTextObject((CPDF_TextObject*)pObj, pTextObj)) { | |
| return TRUE; | |
| } | |
| i++; | |
| } | |
| return FALSE; | |
| } | |
| FX_BOOL CPDF_TextPage::GenerateCharInfo(FX_WCHAR unicode, PAGECHAR_INFO& info) { | |
| int size = m_TempCharList.GetSize(); | |
| PAGECHAR_INFO preChar; | |
| if (size) { | |
| preChar = (PAGECHAR_INFO)m_TempCharList[size - 1]; | |
| } else { | |
| size = m_charList.GetSize(); | |
| if (size == 0) { | |
| return FALSE; | |
| } | |
| preChar = (PAGECHAR_INFO)m_charList[size - 1]; | |
| } | |
| info.m_Index = m_TextBuf.GetLength(); | |
| info.m_Unicode = unicode; | |
| info.m_pTextObj = NULL; | |
| info.m_CharCode = -1; | |
| info.m_Flag = FPDFTEXT_CHAR_GENERATED; | |
| int preWidth = 0; | |
| if (preChar.m_pTextObj && preChar.m_CharCode != (FX_DWORD)-1) { | |
| preWidth = GetCharWidth(preChar.m_CharCode, preChar.m_pTextObj->GetFont()); | |
| } | |
| FX_FLOAT fs = 0; | |
| if (preChar.m_pTextObj) { | |
| fs = preChar.m_pTextObj->GetFontSize(); | |
| } else { | |
| fs = preChar.m_CharBox.Height(); | |
| } | |
| if (!fs) { | |
| fs = 1; | |
| } | |
| info.m_OriginX = preChar.m_OriginX + preWidth * (fs) / 1000; | |
| info.m_OriginY = preChar.m_OriginY; | |
| info.m_CharBox = CFX_FloatRect(info.m_OriginX, info.m_OriginY, info.m_OriginX, | |
| info.m_OriginY); | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_TextPage::IsRectIntersect(const CFX_FloatRect& rect1, | |
| const CFX_FloatRect& rect2) { | |
| CFX_FloatRect rect = rect1; | |
| rect.Intersect(rect2); | |
| return !rect.IsEmpty(); | |
| } | |
| FX_BOOL CPDF_TextPage::IsLetter(FX_WCHAR unicode) { | |
| if (unicode < L'A') { | |
| return FALSE; | |
| } | |
| if (unicode > L'Z' && unicode < L'a') { | |
| return FALSE; | |
| } | |
| if (unicode > L'z') { | |
| return FALSE; | |
| } | |
| return TRUE; | |
| } | |
| CPDF_TextPageFind::CPDF_TextPageFind(const IPDF_TextPage* pTextPage) | |
| : m_pTextPage(pTextPage), | |
| m_flags(0), | |
| m_findNextStart(-1), | |
| m_findPreStart(-1), | |
| m_bMatchCase(FALSE), | |
| m_bMatchWholeWord(FALSE), | |
| m_resStart(0), | |
| m_resEnd(-1), | |
| m_IsFind(FALSE) { | |
| m_strText = m_pTextPage->GetPageText(); | |
| int nCount = pTextPage->CountChars(); | |
| if (nCount) { | |
| m_CharIndex.Add(0); | |
| } | |
| for (int i = 0; i < nCount; i++) { | |
| FPDF_CHAR_INFO info; | |
| pTextPage->GetCharInfo(i, info); | |
| int indexSize = m_CharIndex.GetSize(); | |
| if (info.m_Flag == CHAR_NORMAL || info.m_Flag == CHAR_GENERATED) { | |
| if (indexSize % 2) { | |
| m_CharIndex.Add(1); | |
| } else { | |
| if (indexSize <= 0) { | |
| continue; | |
| } | |
| m_CharIndex.SetAt(indexSize - 1, m_CharIndex.GetAt(indexSize - 1) + 1); | |
| } | |
| } else { | |
| if (indexSize % 2) { | |
| if (indexSize <= 0) { | |
| continue; | |
| } | |
| m_CharIndex.SetAt(indexSize - 1, i + 1); | |
| } else { | |
| m_CharIndex.Add(i + 1); | |
| } | |
| } | |
| } | |
| int indexSize = m_CharIndex.GetSize(); | |
| if (indexSize % 2) { | |
| m_CharIndex.RemoveAt(indexSize - 1); | |
| } | |
| } | |
| int CPDF_TextPageFind::GetCharIndex(int index) const { | |
| return m_pTextPage->CharIndexFromTextIndex(index); | |
| int indexSize = m_CharIndex.GetSize(); | |
| int count = 0; | |
| for (int i = 0; i < indexSize; i += 2) { | |
| count += m_CharIndex.GetAt(i + 1); | |
| if (count > index) { | |
| return index - count + m_CharIndex.GetAt(i + 1) + m_CharIndex.GetAt(i); | |
| } | |
| } | |
| return -1; | |
| } | |
| FX_BOOL CPDF_TextPageFind::FindFirst(const CFX_WideString& findwhat, | |
| int flags, | |
| int startPos) { | |
| if (!m_pTextPage) { | |
| return FALSE; | |
| } | |
| if (m_strText.IsEmpty() || m_bMatchCase != (flags & FPDFTEXT_MATCHCASE)) { | |
| m_strText = m_pTextPage->GetPageText(); | |
| } | |
| CFX_WideString findwhatStr = findwhat; | |
| m_findWhat = findwhatStr; | |
| m_flags = flags; | |
| m_bMatchCase = flags & FPDFTEXT_MATCHCASE; | |
| if (m_strText.IsEmpty()) { | |
| m_IsFind = FALSE; | |
| return TRUE; | |
| } | |
| FX_STRSIZE len = findwhatStr.GetLength(); | |
| if (!m_bMatchCase) { | |
| findwhatStr.MakeLower(); | |
| m_strText.MakeLower(); | |
| } | |
| m_bMatchWholeWord = flags & FPDFTEXT_MATCHWHOLEWORD; | |
| m_findNextStart = startPos; | |
| if (startPos == -1) { | |
| m_findPreStart = m_strText.GetLength() - 1; | |
| } else { | |
| m_findPreStart = startPos; | |
| } | |
| m_csFindWhatArray.RemoveAll(); | |
| int i = 0; | |
| while (i < len) { | |
| if (findwhatStr.GetAt(i) != ' ') { | |
| break; | |
| } | |
| i++; | |
| } | |
| if (i < len) { | |
| ExtractFindWhat(findwhatStr); | |
| } else { | |
| m_csFindWhatArray.Add(findwhatStr); | |
| } | |
| if (m_csFindWhatArray.GetSize() <= 0) { | |
| return FALSE; | |
| } | |
| m_IsFind = TRUE; | |
| m_resStart = 0; | |
| m_resEnd = -1; | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_TextPageFind::FindNext() { | |
| if (!m_pTextPage) { | |
| return FALSE; | |
| } | |
| m_resArray.RemoveAll(); | |
| if (m_findNextStart == -1) { | |
| return FALSE; | |
| } | |
| if (m_strText.IsEmpty()) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| int strLen = m_strText.GetLength(); | |
| if (m_findNextStart > strLen - 1) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| int nCount = m_csFindWhatArray.GetSize(); | |
| int nResultPos = 0; | |
| int nStartPos = 0; | |
| nStartPos = m_findNextStart; | |
| FX_BOOL bSpaceStart = FALSE; | |
| for (int iWord = 0; iWord < nCount; iWord++) { | |
| CFX_WideString csWord = m_csFindWhatArray[iWord]; | |
| if (csWord.IsEmpty()) { | |
| if (iWord == nCount - 1) { | |
| FX_WCHAR strInsert = m_strText.GetAt(nStartPos); | |
| if (strInsert == TEXT_LINEFEED_CHAR || strInsert == TEXT_BLANK_CHAR || | |
| strInsert == TEXT_RETURN_CHAR || strInsert == 160) { | |
| nResultPos = nStartPos + 1; | |
| break; | |
| } | |
| iWord = -1; | |
| } else if (iWord == 0) { | |
| bSpaceStart = TRUE; | |
| } | |
| continue; | |
| } | |
| int endIndex; | |
| nResultPos = m_strText.Find(csWord.c_str(), nStartPos); | |
| if (nResultPos == -1) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| endIndex = nResultPos + csWord.GetLength() - 1; | |
| if (iWord == 0) { | |
| m_resStart = nResultPos; | |
| } | |
| FX_BOOL bMatch = TRUE; | |
| if (iWord != 0 && !bSpaceStart) { | |
| int PreResEndPos = nStartPos; | |
| int curChar = csWord.GetAt(0); | |
| CFX_WideString lastWord = m_csFindWhatArray[iWord - 1]; | |
| int lastChar = lastWord.GetAt(lastWord.GetLength() - 1); | |
| if (nStartPos == nResultPos && | |
| !(_IsIgnoreSpaceCharacter(lastChar) || | |
| _IsIgnoreSpaceCharacter(curChar))) { | |
| bMatch = FALSE; | |
| } | |
| for (int d = PreResEndPos; d < nResultPos; d++) { | |
| FX_WCHAR strInsert = m_strText.GetAt(d); | |
| if (strInsert != TEXT_LINEFEED_CHAR && strInsert != TEXT_BLANK_CHAR && | |
| strInsert != TEXT_RETURN_CHAR && strInsert != 160) { | |
| bMatch = FALSE; | |
| break; | |
| } | |
| } | |
| } else if (bSpaceStart) { | |
| if (nResultPos > 0) { | |
| FX_WCHAR strInsert = m_strText.GetAt(nResultPos - 1); | |
| if (strInsert != TEXT_LINEFEED_CHAR && strInsert != TEXT_BLANK_CHAR && | |
| strInsert != TEXT_RETURN_CHAR && strInsert != 160) { | |
| bMatch = FALSE; | |
| m_resStart = nResultPos; | |
| } else { | |
| m_resStart = nResultPos - 1; | |
| } | |
| } | |
| } | |
| if (m_bMatchWholeWord && bMatch) { | |
| bMatch = IsMatchWholeWord(m_strText, nResultPos, endIndex); | |
| } | |
| nStartPos = endIndex + 1; | |
| if (!bMatch) { | |
| iWord = -1; | |
| if (bSpaceStart) { | |
| nStartPos = m_resStart + m_csFindWhatArray[1].GetLength(); | |
| } else { | |
| nStartPos = m_resStart + m_csFindWhatArray[0].GetLength(); | |
| } | |
| } | |
| } | |
| m_resEnd = nResultPos + | |
| m_csFindWhatArray[m_csFindWhatArray.GetSize() - 1].GetLength() - 1; | |
| m_IsFind = TRUE; | |
| int resStart = GetCharIndex(m_resStart); | |
| int resEnd = GetCharIndex(m_resEnd); | |
| m_pTextPage->GetRectArray(resStart, resEnd - resStart + 1, m_resArray); | |
| if (m_flags & FPDFTEXT_CONSECUTIVE) { | |
| m_findNextStart = m_resStart + 1; | |
| m_findPreStart = m_resEnd - 1; | |
| } else { | |
| m_findNextStart = m_resEnd + 1; | |
| m_findPreStart = m_resStart - 1; | |
| } | |
| return m_IsFind; | |
| } | |
| FX_BOOL CPDF_TextPageFind::FindPrev() { | |
| if (!m_pTextPage) { | |
| return FALSE; | |
| } | |
| m_resArray.RemoveAll(); | |
| if (m_strText.IsEmpty() || m_findPreStart < 0) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| CPDF_TextPageFind findEngine(m_pTextPage); | |
| FX_BOOL ret = findEngine.FindFirst(m_findWhat, m_flags); | |
| if (!ret) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| int order = -1, MatchedCount = 0; | |
| while (ret) { | |
| ret = findEngine.FindNext(); | |
| if (ret) { | |
| int order1 = findEngine.GetCurOrder(); | |
| int MatchedCount1 = findEngine.GetMatchedCount(); | |
| if (((order1 + MatchedCount1) - 1) > m_findPreStart) { | |
| break; | |
| } | |
| order = order1; | |
| MatchedCount = MatchedCount1; | |
| } | |
| } | |
| if (order == -1) { | |
| m_IsFind = FALSE; | |
| return m_IsFind; | |
| } | |
| m_resStart = m_pTextPage->TextIndexFromCharIndex(order); | |
| m_resEnd = m_pTextPage->TextIndexFromCharIndex(order + MatchedCount - 1); | |
| m_IsFind = TRUE; | |
| m_pTextPage->GetRectArray(order, MatchedCount, m_resArray); | |
| if (m_flags & FPDFTEXT_CONSECUTIVE) { | |
| m_findNextStart = m_resStart + 1; | |
| m_findPreStart = m_resEnd - 1; | |
| } else { | |
| m_findNextStart = m_resEnd + 1; | |
| m_findPreStart = m_resStart - 1; | |
| } | |
| return m_IsFind; | |
| } | |
| void CPDF_TextPageFind::ExtractFindWhat(const CFX_WideString& findwhat) { | |
| if (findwhat.IsEmpty()) { | |
| return; | |
| } | |
| int index = 0; | |
| while (1) { | |
| CFX_WideString csWord = TEXT_EMPTY; | |
| int ret = | |
| ExtractSubString(csWord, findwhat.c_str(), index, TEXT_BLANK_CHAR); | |
| if (csWord.IsEmpty()) { | |
| if (ret) { | |
| m_csFindWhatArray.Add(CFX_WideString(L"")); | |
| index++; | |
| continue; | |
| } else { | |
| break; | |
| } | |
| } | |
| int pos = 0; | |
| while (pos < csWord.GetLength()) { | |
| CFX_WideString curStr = csWord.Mid(pos, 1); | |
| FX_WCHAR curChar = csWord.GetAt(pos); | |
| if (_IsIgnoreSpaceCharacter(curChar)) { | |
| if (pos > 0 && curChar == 0x2019) { | |
| pos++; | |
| continue; | |
| } | |
| if (pos > 0) { | |
| CFX_WideString preStr = csWord.Mid(0, pos); | |
| m_csFindWhatArray.Add(preStr); | |
| } | |
| m_csFindWhatArray.Add(curStr); | |
| if (pos == csWord.GetLength() - 1) { | |
| csWord.Empty(); | |
| break; | |
| } | |
| csWord = csWord.Right(csWord.GetLength() - pos - 1); | |
| pos = 0; | |
| continue; | |
| } | |
| pos++; | |
| } | |
| if (!csWord.IsEmpty()) { | |
| m_csFindWhatArray.Add(csWord); | |
| } | |
| index++; | |
| } | |
| } | |
| FX_BOOL CPDF_TextPageFind::IsMatchWholeWord(const CFX_WideString& csPageText, | |
| int startPos, | |
| int endPos) { | |
| int char_left = 0; | |
| int char_right = 0; | |
| int char_count = endPos - startPos + 1; | |
| if (char_count < 1) { | |
| return FALSE; | |
| } | |
| if (char_count == 1 && csPageText.GetAt(startPos) > 255) { | |
| return TRUE; | |
| } | |
| if (startPos - 1 >= 0) { | |
| char_left = csPageText.GetAt(startPos - 1); | |
| } | |
| if (startPos + char_count < csPageText.GetLength()) { | |
| char_right = csPageText.GetAt(startPos + char_count); | |
| } | |
| if ((char_left > 'A' && char_left < 'a') || | |
| (char_left > 'a' && char_left < 'z') || | |
| (char_left > 0xfb00 && char_left < 0xfb06) || | |
| (char_left >= '0' && char_left <= '9') || | |
| (char_right > 'A' && char_right < 'a') || | |
| (char_right > 'a' && char_right < 'z') || | |
| (char_right > 0xfb00 && char_right < 0xfb06) || | |
| (char_right >= '0' && char_right <= '9')) { | |
| return FALSE; | |
| } | |
| if (!(('A' > char_left || char_left > 'Z') && | |
| ('a' > char_left || char_left > 'z') && | |
| ('A' > char_right || char_right > 'Z') && | |
| ('a' > char_right || char_right > 'z'))) { | |
| return FALSE; | |
| } | |
| if (char_count > 0) { | |
| if (csPageText.GetAt(startPos) >= L'0' && | |
| csPageText.GetAt(startPos) <= L'9' && char_left >= L'0' && | |
| char_left <= L'9') { | |
| return FALSE; | |
| } | |
| if (csPageText.GetAt(endPos) >= L'0' && csPageText.GetAt(endPos) <= L'9' && | |
| char_right >= L'0' && char_right <= L'9') { | |
| return FALSE; | |
| } | |
| } | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_TextPageFind::ExtractSubString(CFX_WideString& rString, | |
| const FX_WCHAR* lpszFullString, | |
| int iSubString, | |
| FX_WCHAR chSep) { | |
| if (lpszFullString == NULL) { | |
| return FALSE; | |
| } | |
| while (iSubString--) { | |
| lpszFullString = FXSYS_wcschr(lpszFullString, chSep); | |
| if (lpszFullString == NULL) { | |
| rString.Empty(); | |
| return FALSE; | |
| } | |
| lpszFullString++; | |
| while (*lpszFullString == chSep) { | |
| lpszFullString++; | |
| } | |
| } | |
| const FX_WCHAR* lpchEnd = FXSYS_wcschr(lpszFullString, chSep); | |
| int nLen = (lpchEnd == NULL) ? (int)FXSYS_wcslen(lpszFullString) | |
| : (int)(lpchEnd - lpszFullString); | |
| ASSERT(nLen >= 0); | |
| FXSYS_memcpy(rString.GetBuffer(nLen), lpszFullString, | |
| nLen * sizeof(FX_WCHAR)); | |
| rString.ReleaseBuffer(); | |
| return TRUE; | |
| } | |
| CFX_WideString CPDF_TextPageFind::MakeReverse(const CFX_WideString& str) { | |
| CFX_WideString str2; | |
| str2.Empty(); | |
| int nlen = str.GetLength(); | |
| for (int i = nlen - 1; i >= 0; i--) { | |
| str2 += str.GetAt(i); | |
| } | |
| return str2; | |
| } | |
| void CPDF_TextPageFind::GetRectArray(CFX_RectArray& rects) const { | |
| rects.Copy(m_resArray); | |
| } | |
| int CPDF_TextPageFind::GetCurOrder() const { | |
| return GetCharIndex(m_resStart); | |
| } | |
| int CPDF_TextPageFind::GetMatchedCount() const { | |
| int resStart = GetCharIndex(m_resStart); | |
| int resEnd = GetCharIndex(m_resEnd); | |
| return resEnd - resStart + 1; | |
| } | |
| CPDF_LinkExtract::CPDF_LinkExtract() | |
| : m_pTextPage(nullptr), m_bIsParsed(false) { | |
| } | |
| CPDF_LinkExtract::~CPDF_LinkExtract() { | |
| DeleteLinkList(); | |
| } | |
| FX_BOOL CPDF_LinkExtract::ExtractLinks(const IPDF_TextPage* pTextPage) { | |
| if (!pTextPage || !pTextPage->IsParsed()) | |
| return FALSE; | |
| m_pTextPage = (const CPDF_TextPage*)pTextPage; | |
| m_strPageText = m_pTextPage->GetPageText(0, -1); | |
| DeleteLinkList(); | |
| if (m_strPageText.IsEmpty()) { | |
| return FALSE; | |
| } | |
| ParseLink(); | |
| m_bIsParsed = true; | |
| return TRUE; | |
| } | |
| void CPDF_LinkExtract::DeleteLinkList() { | |
| while (m_LinkList.GetSize()) { | |
| CPDF_LinkExt* linkinfo = NULL; | |
| linkinfo = m_LinkList.GetAt(0); | |
| m_LinkList.RemoveAt(0); | |
| delete linkinfo; | |
| } | |
| m_LinkList.RemoveAll(); | |
| } | |
| int CPDF_LinkExtract::CountLinks() const { | |
| if (!m_bIsParsed) { | |
| return -1; | |
| } | |
| return m_LinkList.GetSize(); | |
| } | |
| void CPDF_LinkExtract::ParseLink() { | |
| int start = 0, pos = 0; | |
| int TotalChar = m_pTextPage->CountChars(); | |
| while (pos < TotalChar) { | |
| FPDF_CHAR_INFO pageChar; | |
| m_pTextPage->GetCharInfo(pos, pageChar); | |
| if (pageChar.m_Flag == CHAR_GENERATED || pageChar.m_Unicode == 0x20 || | |
| pos == TotalChar - 1) { | |
| int nCount = pos - start; | |
| if (pos == TotalChar - 1) { | |
| nCount++; | |
| } | |
| CFX_WideString strBeCheck; | |
| strBeCheck = m_pTextPage->GetPageText(start, nCount); | |
| if (strBeCheck.GetLength() > 5) { | |
| while (strBeCheck.GetLength() > 0) { | |
| FX_WCHAR ch = strBeCheck.GetAt(strBeCheck.GetLength() - 1); | |
| if (ch == L')' || ch == L',' || ch == L'>' || ch == L'.') { | |
| strBeCheck = strBeCheck.Mid(0, strBeCheck.GetLength() - 1); | |
| nCount--; | |
| } else { | |
| break; | |
| } | |
| } | |
| if (nCount > 5 && | |
| (CheckWebLink(strBeCheck) || CheckMailLink(strBeCheck))) { | |
| if (!AppendToLinkList(start, nCount, strBeCheck)) { | |
| break; | |
| } | |
| } | |
| } | |
| start = ++pos; | |
| } else { | |
| pos++; | |
| } | |
| } | |
| } | |
| FX_BOOL CPDF_LinkExtract::CheckWebLink(CFX_WideString& strBeCheck) { | |
| CFX_WideString str = strBeCheck; | |
| str.MakeLower(); | |
| if (str.Find(L"http://www.") != -1) { | |
| strBeCheck = strBeCheck.Right(str.GetLength() - str.Find(L"http://www.")); | |
| return TRUE; | |
| } | |
| if (str.Find(L"http://") != -1) { | |
| strBeCheck = strBeCheck.Right(str.GetLength() - str.Find(L"http://")); | |
| return TRUE; | |
| } | |
| if (str.Find(L"https://www.") != -1) { | |
| strBeCheck = strBeCheck.Right(str.GetLength() - str.Find(L"https://www.")); | |
| return TRUE; | |
| } | |
| if (str.Find(L"https://") != -1) { | |
| strBeCheck = strBeCheck.Right(str.GetLength() - str.Find(L"https://")); | |
| return TRUE; | |
| } | |
| if (str.Find(L"www.") != -1) { | |
| strBeCheck = strBeCheck.Right(str.GetLength() - str.Find(L"www.")); | |
| strBeCheck = L"http://" + strBeCheck; | |
| return TRUE; | |
| } | |
| return FALSE; | |
| } | |
| FX_BOOL CPDF_LinkExtract::CheckMailLink(CFX_WideString& str) { | |
| str.MakeLower(); | |
| int aPos = str.Find(L'@'); | |
| if (aPos < 1) { | |
| return FALSE; | |
| } | |
| if (str.GetAt(aPos - 1) == L'.' || str.GetAt(aPos - 1) == L'_') { | |
| return FALSE; | |
| } | |
| int i; | |
| for (i = aPos - 1; i >= 0; i--) { | |
| FX_WCHAR ch = str.GetAt(i); | |
| if (ch == L'_' || ch == L'.' || (ch >= L'a' && ch <= L'z') || | |
| (ch >= L'0' && ch <= L'9')) { | |
| continue; | |
| } else { | |
| if (i == aPos - 1) { | |
| return FALSE; | |
| } | |
| str = str.Right(str.GetLength() - i - 1); | |
| break; | |
| } | |
| } | |
| aPos = str.Find(L'@'); | |
| if (aPos < 1) { | |
| return FALSE; | |
| } | |
| CFX_WideString strtemp = L""; | |
| for (i = 0; i < aPos; i++) { | |
| FX_WCHAR wch = str.GetAt(i); | |
| if (wch >= L'a' && wch <= L'z') { | |
| break; | |
| } else { | |
| strtemp = str.Right(str.GetLength() - i + 1); | |
| } | |
| } | |
| if (strtemp != L"") { | |
| str = strtemp; | |
| } | |
| aPos = str.Find(L'@'); | |
| if (aPos < 1) { | |
| return FALSE; | |
| } | |
| str.TrimRight(L'.'); | |
| strtemp = str; | |
| int ePos = str.Find(L'.'); | |
| if (ePos == -1) { | |
| return FALSE; | |
| } | |
| while (ePos != -1) { | |
| strtemp = strtemp.Right(strtemp.GetLength() - ePos - 1); | |
| ePos = strtemp.Find('.'); | |
| } | |
| ePos = strtemp.GetLength(); | |
| for (i = 0; i < ePos; i++) { | |
| FX_WCHAR wch = str.GetAt(i); | |
| if ((wch >= L'a' && wch <= L'z') || (wch >= L'0' && wch <= L'9')) { | |
| continue; | |
| } else { | |
| str = str.Left(str.GetLength() - ePos + i + 1); | |
| ePos = ePos - i - 1; | |
| break; | |
| } | |
| } | |
| int nLen = str.GetLength(); | |
| for (i = aPos + 1; i < nLen - ePos; i++) { | |
| FX_WCHAR wch = str.GetAt(i); | |
| if (wch == L'-' || wch == L'.' || (wch >= L'a' && wch <= L'z') || | |
| (wch >= L'0' && wch <= L'9')) { | |
| continue; | |
| } else { | |
| return FALSE; | |
| } | |
| } | |
| if (str.Find(L"mailto:") == -1) { | |
| str = L"mailto:" + str; | |
| } | |
| return TRUE; | |
| } | |
| FX_BOOL CPDF_LinkExtract::AppendToLinkList(int start, | |
| int count, | |
| const CFX_WideString& strUrl) { | |
| CPDF_LinkExt* linkInfo = new CPDF_LinkExt; | |
| linkInfo->m_strUrl = strUrl; | |
| linkInfo->m_Start = start; | |
| linkInfo->m_Count = count; | |
| m_LinkList.Add(linkInfo); | |
| return TRUE; | |
| } | |
| CFX_WideString CPDF_LinkExtract::GetURL(int index) const { | |
| if (!m_bIsParsed || index < 0 || index >= m_LinkList.GetSize()) { | |
| return L""; | |
| } | |
| CPDF_LinkExt* link = NULL; | |
| link = m_LinkList.GetAt(index); | |
| if (!link) { | |
| return L""; | |
| } | |
| return link->m_strUrl; | |
| } | |
| void CPDF_LinkExtract::GetBoundedSegment(int index, | |
| int& start, | |
| int& count) const { | |
| if (!m_bIsParsed || index < 0 || index >= m_LinkList.GetSize()) { | |
| return; | |
| } | |
| CPDF_LinkExt* link = NULL; | |
| link = m_LinkList.GetAt(index); | |
| if (!link) { | |
| return; | |
| } | |
| start = link->m_Start; | |
| count = link->m_Count; | |
| } | |
| void CPDF_LinkExtract::GetRects(int index, CFX_RectArray& rects) const { | |
| if (!m_bIsParsed || 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); | |
| } |