diff options
author | LogicParrot <LogicParrot@users.noreply.github.com> | 2016-02-05 22:45:45 +0100 |
---|---|---|
committer | LogicParrot <LogicParrot@users.noreply.github.com> | 2016-02-05 22:50:18 +0100 |
commit | ca6ef58b1ee8521e4b940ee4883dee714960e413 (patch) | |
tree | 8532add455224b07c07a759e3d906f50c0695888 /src/HTTPServer/NameValueParser.cpp | |
parent | Merge pull request #2972 from marvinkopf/PlayerAutoComplete (diff) | |
download | cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar.gz cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar.bz2 cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar.lz cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar.xz cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.tar.zst cuberite-ca6ef58b1ee8521e4b940ee4883dee714960e413.zip |
Diffstat (limited to '')
-rw-r--r-- | src/HTTPServer/NameValueParser.cpp | 30 |
1 files changed, 15 insertions, 15 deletions
diff --git a/src/HTTPServer/NameValueParser.cpp b/src/HTTPServer/NameValueParser.cpp index b345fef88..f759c4d21 100644 --- a/src/HTTPServer/NameValueParser.cpp +++ b/src/HTTPServer/NameValueParser.cpp @@ -29,18 +29,18 @@ public: Parser2.Parse(Data + i, 1); } Parser2.Finish(); - + // Parse as a single chunk of data: cNameValueParser Parser(Data, sizeof(Data) - 1); - + // Use the debugger to inspect the Parser variable - + // Check that the two parsers have the same content: for (cNameValueParser::const_iterator itr = Parser.begin(), end = Parser.end(); itr != end; ++itr) { ASSERT(Parser2[itr->first] == itr->second); } // for itr - Parser[] - + // Try parsing in 2-char chunks: cNameValueParser Parser3; for (int i = 0; i < sizeof(Data) - 2; i += 2) @@ -52,13 +52,13 @@ public: Parser3.Parse(Data + sizeof(Data) - 2, 1); } Parser3.Finish(); - + // Check that the third parser has the same content: for (cNameValueParser::const_iterator itr = Parser.begin(), end = Parser.end(); itr != end; ++itr) { ASSERT(Parser3[itr->first] == itr->second); } // for itr - Parser[] - + printf("cNameValueParserTest done"); } } g_Test; @@ -96,7 +96,7 @@ cNameValueParser::cNameValueParser(const char * a_Data, size_t a_Size, bool a_Al void cNameValueParser::Parse(const char * a_Data, size_t a_Size) { ASSERT(m_State != psFinished); // Calling Parse() after Finish() is wrong! - + size_t Last = 0; for (size_t i = 0; i < a_Size;) { @@ -107,7 +107,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) { return; } - + case psKeySpace: { // Skip whitespace until a non-whitespace is found, then start the key: @@ -122,7 +122,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } break; } - + case psKey: { // Read the key until whitespace or an equal sign: @@ -174,7 +174,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } break; } - + case psEqualSpace: { // The space before the expected equal sign; the current key is already assigned @@ -211,7 +211,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } // while (i < a_Size) break; } // case psEqualSpace - + case psEqual: { // just parsed the equal-sign @@ -256,7 +256,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } // while (i < a_Size) break; } // case psEqual - + case psValueInDQuotes: { while (i < a_Size) @@ -280,7 +280,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } break; } // case psValueInDQuotes - + case psValueInSQuotes: { while (i < a_Size) @@ -304,7 +304,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } break; } // case psValueInSQuotes - + case psValueRaw: { while (i < a_Size) @@ -328,7 +328,7 @@ void cNameValueParser::Parse(const char * a_Data, size_t a_Size) } break; } // case psValueRaw - + case psAfterValue: { // Between the closing DQuote or SQuote and the terminating semicolon |