diff --git a/src/json.cc b/src/json.cc index b664ef157..f7da724ce 100644 --- a/src/json.cc +++ b/src/json.cc @@ -2151,13 +2151,13 @@ Turns a code point into it's UTF-8 representation. You should only pass numbers < 0x10ffff into this function (everything else is a invalid code point). -@return the UTF-8 representation of the given codepoint +@return the UTF-8 representation of the given code point @pre This method isn't accessing the members of the parser @post This method isn't accessing the members of the parser */ -std::string json::parser::codepointToUTF8(unsigned int codepoint) +std::string json::parser::codePointToUTF8(unsigned int codePoint) { // this method contains a lot of bit manipulations to // build the bytes for UTF-8. @@ -2171,49 +2171,49 @@ std::string json::parser::codepointToUTF8(unsigned int codepoint) // (e.g. 1 to 4 bytes) to save the reallocations. - if (codepoint <= 0x7f) + if (codePoint <= 0x7f) { - // it's just a ASCII compatible codepoint, + // it's just a ASCII compatible codePoint, // so we just interpret the point as a character // and return ASCII - return std::string(1, static_cast(codepoint)); + return std::string(1, static_cast(codePoint)); } // if true, we need two bytes to encode this as UTF-8 - else if (codepoint <= 0x7ff) + else if (codePoint <= 0x7ff) { // the 0xC0 enables the two most significant two bits // to make this a two-byte UTF-8 character. - std::string result(2, static_cast(0xC0 | ((codepoint >> 6) & 0x1F))); - result[1] = static_cast(0x80 | (codepoint & 0x3F)); + std::string result(2, static_cast(0xC0 | ((codePoint >> 6) & 0x1F))); + result[1] = static_cast(0x80 | (codePoint & 0x3F)); return result; } // if true, now we need three bytes to encode this as UTF-8 - else if (codepoint <= 0xffff) + else if (codePoint <= 0xffff) { // the 0xE0 enables the three most significant two bits // to make this a three-byte UTF-8 character. - std::string result(3, static_cast(0xE0 | ((codepoint >> 12) & 0x0F))); - result[1] = static_cast(0x80 | ((codepoint >> 6) & 0x3F)); - result[2] = static_cast(0x80 | (codepoint & 0x3F)); + std::string result(3, static_cast(0xE0 | ((codePoint >> 12) & 0x0F))); + result[1] = static_cast(0x80 | ((codePoint >> 6) & 0x3F)); + result[2] = static_cast(0x80 | (codePoint & 0x3F)); return result; } // if true, we need maximal four bytes to encode this as UTF-8 - else if (codepoint <= 0x10ffff) + else if (codePoint <= 0x10ffff) { // the 0xE0 enables the four most significant two bits // to make this a three-byte UTF-8 character. - std::string result(4, static_cast(0xF0 | ((codepoint >> 18) & 0x07))); - result[1] = static_cast(0x80 | ((codepoint >> 12) & 0x3F)); - result[2] = static_cast(0x80 | ((codepoint >> 6) & 0x3F)); - result[3] = static_cast(0x80 | (codepoint & 0x3F)); + std::string result(4, static_cast(0xF0 | ((codePoint >> 18) & 0x07))); + result[1] = static_cast(0x80 | ((codePoint >> 12) & 0x3F)); + result[2] = static_cast(0x80 | ((codePoint >> 6) & 0x3F)); + result[3] = static_cast(0x80 | (codePoint & 0x3F)); return result; } else { // Can't be tested without direct access to this private method. - std::string errorMessage = "Invalid codepoint: "; - errorMessage += codepoint; + std::string errorMessage = "Invalid codePoint: "; + errorMessage += codePoint; error(errorMessage); } } @@ -2227,7 +2227,7 @@ Parses 4 hexadecimal characters as a number. @post pos_ is pointing to the character after the 4 hexadecimal characters. */ -unsigned int json::parser::parse4HexCodepoint() +unsigned int json::parser::parse4HexCodePoint() { const auto startPos = pos_; @@ -2285,10 +2285,10 @@ std::string json::parser::parseUnicodeEscape() // jump to the first hex value pos_++; // parse the hex first hex values - unsigned int firstCodepoint = parse4HexCodepoint(); + unsigned int firstCodePoint = parse4HexCodePoint(); - if (firstCodepoint >= 0xD800 && firstCodepoint <= 0xDBFF) + if (firstCodePoint >= 0xD800 && firstCodePoint <= 0xDBFF) { // we found invalid code points, which means we either have a malformed input // or we found a high surrogate. @@ -2299,29 +2299,29 @@ std::string json::parser::parseUnicodeEscape() pos_ += 2; // try to parse the next hex values. // the method does boundary checking for us, so no need to do that here - unsigned secondCodepoint = parse4HexCodepoint(); + unsigned secondCodePoint = parse4HexCodePoint(); // ok, we have a low surrogate, check if it is a valid one - if (secondCodepoint >= 0xDC00 && secondCodepoint <= 0xDFFF) + if (secondCodePoint >= 0xDC00 && secondCodePoint <= 0xDFFF) { // calculate the final code point from the pair according to the spec unsigned int finalCodePoint = // high surrogate occupies the most significant 22 bits - (firstCodepoint << 10) + (firstCodePoint << 10) // low surrogate occupies the least significant 15 bits - + secondCodepoint + + secondCodePoint // there is still the 0xD800, 0xDC00 and 0x10000 noise in the result // so we have to substract with (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00 - 0x35FDC00; // we transform the calculated point into UTF-8 - return codepointToUTF8(finalCodePoint); + return codePointToUTF8(finalCodePoint); } else error("missing low surrogate"); } // We have Form 1, so we just interpret the XXXX as a code point - return codepointToUTF8(firstCodepoint); + return codePointToUTF8(firstCodePoint); } diff --git a/src/json.h b/src/json.h index 90c5ded88..2fa0bdf30 100644 --- a/src/json.h +++ b/src/json.h @@ -419,9 +419,9 @@ class json /// parse a quoted string inline std::string parseString(); /// transforms a unicode codepoint to it's UTF-8 presentation - inline std::string codepointToUTF8(unsigned int codepoint); - /// parses 4 hex characters that represent a unicode codepoint - inline unsigned int parse4HexCodepoint(); + inline std::string codePointToUTF8(unsigned int codePoint); + /// parses 4 hex characters that represent a unicode code point + inline unsigned int parse4HexCodePoint(); /// parses \uXXXX[\uXXXX] unicode escape characters inline std::string parseUnicodeEscape(); /// parse a Boolean "true"