diff options
author | dan sinclair <dsinclair@chromium.org> | 2018-03-08 15:21:59 +0000 |
---|---|---|
committer | Chromium commit bot <commit-bot@chromium.org> | 2018-03-08 15:21:59 +0000 |
commit | f2a069413d98749b99d7d51430c284375eb2b56c (patch) | |
tree | 4e250c19ba251dbf505c5cbdc7cb1cc1849cb7bf /core/fpdfapi | |
parent | 7544a4b4fd98b4cf6e06e717758e237ffd822671 (diff) | |
download | pdfium-f2a069413d98749b99d7d51430c284375eb2b56c.tar.xz |
Cleanup CPDF_SimpleParser
Minor code cleanup.
Change-Id: Ibc3ea17d270de517058c41e9b466bc8fb042ecd7
Reviewed-on: https://pdfium-review.googlesource.com/28230
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Diffstat (limited to 'core/fpdfapi')
-rw-r--r-- | core/fpdfapi/parser/cpdf_simple_parser.cpp | 107 |
1 files changed, 61 insertions, 46 deletions
diff --git a/core/fpdfapi/parser/cpdf_simple_parser.cpp b/core/fpdfapi/parser/cpdf_simple_parser.cpp index b49f8423c3..9e3bf54022 100644 --- a/core/fpdfapi/parser/cpdf_simple_parser.cpp +++ b/core/fpdfapi/parser/cpdf_simple_parser.cpp @@ -6,6 +6,8 @@ #include "core/fpdfapi/parser/cpdf_simple_parser.h" +#include <vector> + #include "core/fpdfapi/parser/fpdf_parser_utility.h" CPDF_SimpleParser::CPDF_SimpleParser(const uint8_t* pData, uint32_t dwSize) @@ -18,9 +20,12 @@ std::pair<const uint8_t*, uint32_t> CPDF_SimpleParser::ParseWord() { const uint8_t* pStart = nullptr; uint8_t dwSize = 0; uint8_t ch; + + // Skip whitespace and comment lines. while (1) { if (m_dwSize <= m_dwCurPos) return std::make_pair(pStart, dwSize); + ch = m_pData[m_dwCurPos++]; while (PDFCharIsWhitespace(ch)) { if (m_dwSize <= m_dwCurPos) @@ -34,6 +39,7 @@ std::pair<const uint8_t*, uint32_t> CPDF_SimpleParser::ParseWord() { while (1) { if (m_dwSize <= m_dwCurPos) return std::make_pair(pStart, dwSize); + ch = m_pData[m_dwCurPos++]; if (PDFCharIsLineEnding(ch)) break; @@ -43,36 +49,41 @@ std::pair<const uint8_t*, uint32_t> CPDF_SimpleParser::ParseWord() { uint32_t start_pos = m_dwCurPos - 1; pStart = m_pData + start_pos; if (PDFCharIsDelimiter(ch)) { + // Find names if (ch == '/') { while (1) { if (m_dwSize <= m_dwCurPos) - return std::make_pair(pStart, dwSize); + break; + ch = m_pData[m_dwCurPos++]; if (!PDFCharIsOther(ch) && !PDFCharIsNumeric(ch)) { m_dwCurPos--; dwSize = m_dwCurPos - start_pos; - return std::make_pair(pStart, dwSize); + break; } } - } else { - dwSize = 1; - if (ch == '<') { - if (m_dwSize <= m_dwCurPos) - return std::make_pair(pStart, dwSize); - ch = m_pData[m_dwCurPos++]; - if (ch == '<') - dwSize = 2; - else - m_dwCurPos--; - } else if (ch == '>') { - if (m_dwSize <= m_dwCurPos) - return std::make_pair(pStart, dwSize); - ch = m_pData[m_dwCurPos++]; - if (ch == '>') - dwSize = 2; - else - m_dwCurPos--; - } + return std::make_pair(pStart, dwSize); + } + + dwSize = 1; + if (ch == '<') { + if (m_dwSize <= m_dwCurPos) + return std::make_pair(pStart, dwSize); + + ch = m_pData[m_dwCurPos++]; + if (ch == '<') + dwSize = 2; + else + m_dwCurPos--; + } else if (ch == '>') { + if (m_dwSize <= m_dwCurPos) + return std::make_pair(pStart, dwSize); + + ch = m_pData[m_dwCurPos++]; + if (ch == '>') + dwSize = 2; + else + m_dwCurPos--; } return std::make_pair(pStart, dwSize); } @@ -96,41 +107,46 @@ ByteStringView CPDF_SimpleParser::GetWord() { const uint8_t* pStart; uint32_t dwSize; std::tie(pStart, dwSize) = ParseWord(); - if (dwSize == 1 && pStart[0] == '<') { - while (m_dwCurPos < m_dwSize && m_pData[m_dwCurPos] != '>') { + + if (dwSize != 1) + return ByteStringView(pStart, dwSize); + + if (pStart[0] == '<') { + while (m_dwCurPos < m_dwSize && m_pData[m_dwCurPos] != '>') m_dwCurPos++; - } - if (m_dwCurPos < m_dwSize) { + + if (m_dwCurPos < m_dwSize) m_dwCurPos++; - } + return ByteStringView(pStart, static_cast<size_t>(m_dwCurPos - (pStart - m_pData))); } - if (dwSize == 1 && pStart[0] == '(') { + + if (pStart[0] == '(') { int level = 1; while (m_dwCurPos < m_dwSize) { if (m_pData[m_dwCurPos] == ')') { level--; - if (level == 0) { + if (level == 0) break; - } } + if (m_pData[m_dwCurPos] == '\\') { - if (m_dwSize <= m_dwCurPos) { + if (m_dwSize <= m_dwCurPos) break; - } + m_dwCurPos++; } else if (m_pData[m_dwCurPos] == '(') { level++; } - if (m_dwSize <= m_dwCurPos) { + if (m_dwSize <= m_dwCurPos) break; - } + m_dwCurPos++; } - if (m_dwCurPos < m_dwSize) { + if (m_dwCurPos < m_dwSize) m_dwCurPos++; - } + return ByteStringView(pStart, static_cast<size_t>(m_dwCurPos - (pStart - m_pData))); } @@ -140,30 +156,29 @@ ByteStringView CPDF_SimpleParser::GetWord() { bool CPDF_SimpleParser::FindTagParamFromStart(const ByteStringView& token, int nParams) { nParams++; - uint32_t* pBuf = FX_Alloc(uint32_t, nParams); + + std::vector<uint32_t> pBuf(nParams); int buf_index = 0; int buf_count = 0; m_dwCurPos = 0; while (1) { pBuf[buf_index++] = m_dwCurPos; - if (buf_index == nParams) { + if (buf_index == nParams) buf_index = 0; - } + buf_count++; - if (buf_count > nParams) { + if (buf_count > nParams) buf_count = nParams; - } + ByteStringView word = GetWord(); - if (word.IsEmpty()) { - FX_Free(pBuf); + if (word.IsEmpty()) return false; - } + if (word == token) { - if (buf_count < nParams) { + if (buf_count < nParams) continue; - } + m_dwCurPos = pBuf[buf_index]; - FX_Free(pBuf); return true; } } |