diff options
author | Ryan Harrison <rharrison@chromium.org> | 2017-09-27 10:53:11 -0400 |
---|---|---|
committer | Chromium commit bot <commit-bot@chromium.org> | 2017-09-27 16:11:38 +0000 |
commit | 875e98c581952478f3a3ccef9b2f2e3ed06c5346 (patch) | |
tree | 8e0d7e032056bf4c73d6da43c0f3ce4eadb74dfd /core/fpdftext/cpdf_textpage.cpp | |
parent | cc3a3ee3ebcc1baabdfa7ffca5876dbbfa3980c1 (diff) | |
download | pdfium-875e98c581952478f3a3ccef9b2f2e3ed06c5346.tar.xz |
Remove FX_STRSIZE and replace with size_t
BUG=pdfium:828
Change-Id: I5c40237433ebabaeabdb43aec9cdf783e41dfe16
Reviewed-on: https://pdfium-review.googlesource.com/13230
Reviewed-by: dsinclair <dsinclair@chromium.org>
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
Diffstat (limited to 'core/fpdftext/cpdf_textpage.cpp')
-rw-r--r-- | core/fpdftext/cpdf_textpage.cpp | 26 |
1 files changed, 12 insertions, 14 deletions
diff --git a/core/fpdftext/cpdf_textpage.cpp b/core/fpdftext/cpdf_textpage.cpp index f73793cd72..fd33fb2f2f 100644 --- a/core/fpdftext/cpdf_textpage.cpp +++ b/core/fpdftext/cpdf_textpage.cpp @@ -66,7 +66,7 @@ float CalculateBaseSpace(const CPDF_TextObject* pTextObj, return baseSpace; } -FX_STRSIZE Unicode_GetNormalization(wchar_t wch, wchar_t* pDst) { +size_t Unicode_GetNormalization(wchar_t wch, wchar_t* pDst) { wch = wch & 0xFFFF; wchar_t wFind = g_UnicodeData_Normalization[wch]; if (!wFind) { @@ -93,7 +93,7 @@ FX_STRSIZE Unicode_GetNormalization(wchar_t wch, wchar_t* pDst) { while (n--) *pDst++ = *pMap++; } - return (FX_STRSIZE)wFind; + return static_cast<size_t>(wFind); } float MaskPercentFilled(const std::vector<bool>& mask, @@ -649,11 +649,11 @@ void CPDF_TextPage::AddCharInfoByLRDirection(wchar_t wChar, info.m_Index = m_TextBuf.GetLength(); if (wChar >= 0xFB00 && wChar <= 0xFB06) { wchar_t* pDst = nullptr; - FX_STRSIZE nCount = Unicode_GetNormalization(wChar, pDst); + size_t nCount = Unicode_GetNormalization(wChar, pDst); if (nCount >= 1) { pDst = FX_Alloc(wchar_t, nCount); Unicode_GetNormalization(wChar, pDst); - for (FX_STRSIZE nIndex = 0; nIndex < nCount; nIndex++) { + for (size_t nIndex = 0; nIndex < nCount; nIndex++) { PAGECHAR_INFO info2 = info; info2.m_Unicode = pDst[nIndex]; info2.m_Flag = FPDFTEXT_CHAR_PIECE; @@ -679,11 +679,11 @@ void CPDF_TextPage::AddCharInfoByRLDirection(wchar_t wChar, info.m_Index = m_TextBuf.GetLength(); wChar = FX_GetMirrorChar(wChar); wchar_t* pDst = nullptr; - FX_STRSIZE nCount = Unicode_GetNormalization(wChar, pDst); + size_t nCount = Unicode_GetNormalization(wChar, pDst); if (nCount >= 1) { pDst = FX_Alloc(wchar_t, nCount); Unicode_GetNormalization(wChar, pDst); - for (FX_STRSIZE nIndex = 0; nIndex < nCount; nIndex++) { + for (size_t nIndex = 0; nIndex < nCount; nIndex++) { PAGECHAR_INFO info2 = info; info2.m_Unicode = pDst[nIndex]; info2.m_Flag = FPDFTEXT_CHAR_PIECE; @@ -704,7 +704,7 @@ void CPDF_TextPage::CloseTempLine() { WideString str = m_TempTextBuf.MakeString(); bool bPrevSpace = false; - for (FX_STRSIZE i = 0; i < str.GetLength(); i++) { + for (size_t i = 0; i < str.GetLength(); i++) { if (str[i] != ' ') { bPrevSpace = false; continue; @@ -838,13 +838,12 @@ FPDFText_MarkedContent CPDF_TextPage::PreMarkedContent(PDFTEXT_Obj Obj) { return FPDFText_MarkedContent::Done; } - FX_STRSIZE nItems = actText.GetLength(); - if (nItems < 1) + if (actText.IsEmpty()) return FPDFText_MarkedContent::Pass; CPDF_Font* pFont = pTextObj->GetFont(); bExist = false; - for (FX_STRSIZE i = 0; i < nItems; i++) { + for (size_t i = 0; i < actText.GetLength(); i++) { if (pFont->CharCodeFromUnicode(actText[i]) != CPDF_Font::kInvalidCharCode) { bExist = true; break; @@ -854,7 +853,7 @@ FPDFText_MarkedContent CPDF_TextPage::PreMarkedContent(PDFTEXT_Obj Obj) { return FPDFText_MarkedContent::Pass; bExist = false; - for (FX_STRSIZE i = 0; i < nItems; i++) { + for (size_t i = 0; i < actText.GetLength(); i++) { wchar_t wChar = actText[i]; if ((wChar > 0x80 && wChar < 0xFFFD) || (wChar <= 0x80 && isprint(wChar))) { bExist = true; @@ -883,15 +882,14 @@ void CPDF_TextPage::ProcessMarkedContent(PDFTEXT_Obj Obj) { if (pDict) actText = pDict->GetUnicodeTextFor("ActualText"); } - FX_STRSIZE nItems = actText.GetLength(); - if (nItems < 1) + if (actText.IsEmpty()) return; CPDF_Font* pFont = pTextObj->GetFont(); CFX_Matrix matrix = pTextObj->GetTextMatrix(); matrix.Concat(Obj.m_formMatrix); - for (FX_STRSIZE k = 0; k < nItems; k++) { + for (size_t k = 0; k < actText.GetLength(); k++) { wchar_t wChar = actText[k]; if (wChar <= 0x80 && !isprint(wChar)) wChar = 0x20; |