| Index: third_party/WebKit/Source/wtf/text/Base64.cpp
|
| diff --git a/third_party/WebKit/Source/wtf/text/Base64.cpp b/third_party/WebKit/Source/wtf/text/Base64.cpp
|
| index b179184fb8ce63ae1f9912927de1db5c200104f9..541f925028771310eb18c3b2cbdd7a56cc69614f 100644
|
| --- a/third_party/WebKit/Source/wtf/text/Base64.cpp
|
| +++ b/third_party/WebKit/Source/wtf/text/Base64.cpp
|
| @@ -37,8 +37,7 @@ static const char base64EncMap[64] = {
|
| 0x67, 0x68, 0x69, 0x6A, 0x6B, 0x6C, 0x6D, 0x6E,
|
| 0x6F, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76,
|
| 0x77, 0x78, 0x79, 0x7A, 0x30, 0x31, 0x32, 0x33,
|
| - 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x2B, 0x2F
|
| -};
|
| + 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x2B, 0x2F};
|
|
|
| static const char base64DecMap[128] = {
|
| 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
| @@ -56,196 +55,186 @@ static const char base64DecMap[128] = {
|
| 0x00, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F, 0x20,
|
| 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,
|
| 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F, 0x30,
|
| - 0x31, 0x32, 0x33, 0x00, 0x00, 0x00, 0x00, 0x00
|
| -};
|
| -
|
| -String base64Encode(const char* data, unsigned length, Base64EncodePolicy policy)
|
| -{
|
| - Vector<char> result;
|
| - base64Encode(data, length, result, policy);
|
| - return String(result.data(), result.size());
|
| + 0x31, 0x32, 0x33, 0x00, 0x00, 0x00, 0x00, 0x00};
|
| +
|
| +String base64Encode(const char* data, unsigned length, Base64EncodePolicy policy) {
|
| + Vector<char> result;
|
| + base64Encode(data, length, result, policy);
|
| + return String(result.data(), result.size());
|
| }
|
|
|
| -void base64Encode(const char* data, unsigned len, Vector<char>& out, Base64EncodePolicy policy)
|
| -{
|
| - out.clear();
|
| - if (!len)
|
| - return;
|
| -
|
| - // If the input string is pathologically large, just return nothing.
|
| - // Note: Keep this in sync with the "outLength" computation below.
|
| - // Rather than being perfectly precise, this is a bit conservative.
|
| - const unsigned maxInputBufferSize = UINT_MAX / 77 * 76 / 4 * 3 - 2;
|
| - if (len > maxInputBufferSize)
|
| - return;
|
| -
|
| - unsigned sidx = 0;
|
| - unsigned didx = 0;
|
| -
|
| - unsigned outLength = ((len + 2) / 3) * 4;
|
| -
|
| - // Deal with the 76 character per line limit specified in RFC 2045.
|
| - bool insertLFs = (policy == Base64InsertLFs && outLength > 76);
|
| - if (insertLFs)
|
| - outLength += ((outLength - 1) / 76);
|
| -
|
| - int count = 0;
|
| - out.grow(outLength);
|
| -
|
| - // 3-byte to 4-byte conversion + 0-63 to ascii printable conversion
|
| - if (len > 1) {
|
| - while (sidx < len - 2) {
|
| - if (insertLFs) {
|
| - if (count && !(count % 76))
|
| - out[didx++] = '\n';
|
| - count += 4;
|
| - }
|
| - out[didx++] = base64EncMap[(data[sidx] >> 2) & 077];
|
| - out[didx++] = base64EncMap[((data[sidx + 1] >> 4) & 017) | ((data[sidx] << 4) & 077)];
|
| - out[didx++] = base64EncMap[((data[sidx + 2] >> 6) & 003) | ((data[sidx + 1] << 2) & 077)];
|
| - out[didx++] = base64EncMap[data[sidx + 2] & 077];
|
| - sidx += 3;
|
| - }
|
| +void base64Encode(const char* data, unsigned len, Vector<char>& out, Base64EncodePolicy policy) {
|
| + out.clear();
|
| + if (!len)
|
| + return;
|
| +
|
| + // If the input string is pathologically large, just return nothing.
|
| + // Note: Keep this in sync with the "outLength" computation below.
|
| + // Rather than being perfectly precise, this is a bit conservative.
|
| + const unsigned maxInputBufferSize = UINT_MAX / 77 * 76 / 4 * 3 - 2;
|
| + if (len > maxInputBufferSize)
|
| + return;
|
| +
|
| + unsigned sidx = 0;
|
| + unsigned didx = 0;
|
| +
|
| + unsigned outLength = ((len + 2) / 3) * 4;
|
| +
|
| + // Deal with the 76 character per line limit specified in RFC 2045.
|
| + bool insertLFs = (policy == Base64InsertLFs && outLength > 76);
|
| + if (insertLFs)
|
| + outLength += ((outLength - 1) / 76);
|
| +
|
| + int count = 0;
|
| + out.grow(outLength);
|
| +
|
| + // 3-byte to 4-byte conversion + 0-63 to ascii printable conversion
|
| + if (len > 1) {
|
| + while (sidx < len - 2) {
|
| + if (insertLFs) {
|
| + if (count && !(count % 76))
|
| + out[didx++] = '\n';
|
| + count += 4;
|
| + }
|
| + out[didx++] = base64EncMap[(data[sidx] >> 2) & 077];
|
| + out[didx++] = base64EncMap[((data[sidx + 1] >> 4) & 017) | ((data[sidx] << 4) & 077)];
|
| + out[didx++] = base64EncMap[((data[sidx + 2] >> 6) & 003) | ((data[sidx + 1] << 2) & 077)];
|
| + out[didx++] = base64EncMap[data[sidx + 2] & 077];
|
| + sidx += 3;
|
| }
|
| -
|
| - if (sidx < len) {
|
| - if (insertLFs && (count > 0) && !(count % 76))
|
| - out[didx++] = '\n';
|
| -
|
| - out[didx++] = base64EncMap[(data[sidx] >> 2) & 077];
|
| - if (sidx < len - 1) {
|
| - out[didx++] = base64EncMap[((data[sidx + 1] >> 4) & 017) | ((data[sidx] << 4) & 077)];
|
| - out[didx++] = base64EncMap[(data[sidx + 1] << 2) & 077];
|
| - } else {
|
| - out[didx++] = base64EncMap[(data[sidx] << 4) & 077];
|
| - }
|
| + }
|
| +
|
| + if (sidx < len) {
|
| + if (insertLFs && (count > 0) && !(count % 76))
|
| + out[didx++] = '\n';
|
| +
|
| + out[didx++] = base64EncMap[(data[sidx] >> 2) & 077];
|
| + if (sidx < len - 1) {
|
| + out[didx++] = base64EncMap[((data[sidx + 1] >> 4) & 017) | ((data[sidx] << 4) & 077)];
|
| + out[didx++] = base64EncMap[(data[sidx + 1] << 2) & 077];
|
| + } else {
|
| + out[didx++] = base64EncMap[(data[sidx] << 4) & 077];
|
| }
|
| + }
|
|
|
| - // Add padding
|
| - while (didx < out.size()) {
|
| - out[didx] = '=';
|
| - ++didx;
|
| - }
|
| + // Add padding
|
| + while (didx < out.size()) {
|
| + out[didx] = '=';
|
| + ++didx;
|
| + }
|
| }
|
|
|
| -bool base64Decode(const Vector<char>& in, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy)
|
| -{
|
| - out.clear();
|
| +bool base64Decode(const Vector<char>& in, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy) {
|
| + out.clear();
|
|
|
| - // If the input string is pathologically large, just return nothing.
|
| - if (in.size() > UINT_MAX)
|
| - return false;
|
| + // If the input string is pathologically large, just return nothing.
|
| + if (in.size() > UINT_MAX)
|
| + return false;
|
|
|
| - return base64Decode(in.data(), in.size(), out, shouldIgnoreCharacter, policy);
|
| + return base64Decode(in.data(), in.size(), out, shouldIgnoreCharacter, policy);
|
| }
|
|
|
| -template<typename T>
|
| -static inline bool base64DecodeInternal(const T* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy)
|
| -{
|
| - out.clear();
|
| - if (!length)
|
| - return true;
|
| -
|
| - out.grow(length);
|
| -
|
| - unsigned equalsSignCount = 0;
|
| - unsigned outLength = 0;
|
| - bool hadError = false;
|
| - for (unsigned idx = 0; idx < length; ++idx) {
|
| - UChar ch = data[idx];
|
| - if (ch == '=') {
|
| - ++equalsSignCount;
|
| - // There should never be more than 2 padding characters.
|
| - if (policy == Base64ValidatePadding && equalsSignCount > 2) {
|
| - hadError = true;
|
| - break;
|
| - }
|
| - } else if (('0' <= ch && ch <= '9') || ('A' <= ch && ch <= 'Z') || ('a' <= ch && ch <= 'z') || ch == '+' || ch == '/') {
|
| - if (equalsSignCount) {
|
| - hadError = true;
|
| - break;
|
| - }
|
| - out[outLength++] = base64DecMap[ch];
|
| - } else if (!shouldIgnoreCharacter || !shouldIgnoreCharacter(ch)) {
|
| - hadError = true;
|
| - break;
|
| - }
|
| - }
|
| +template <typename T>
|
| +static inline bool base64DecodeInternal(const T* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy) {
|
| + out.clear();
|
| + if (!length)
|
| + return true;
|
|
|
| - if (outLength < out.size())
|
| - out.shrink(outLength);
|
| -
|
| - if (hadError)
|
| - return false;
|
| -
|
| - if (!outLength)
|
| - return !equalsSignCount;
|
| -
|
| - // There should be no padding if length is a multiple of 4.
|
| - // We use (outLength + equalsSignCount) instead of length because we don't want to account for ignored characters.
|
| - if (policy == Base64ValidatePadding && equalsSignCount && (outLength + equalsSignCount) % 4)
|
| - return false;
|
| -
|
| - // Valid data is (n * 4 + [0,2,3]) characters long.
|
| - if ((outLength % 4) == 1)
|
| - return false;
|
| -
|
| - // 4-byte to 3-byte conversion
|
| - outLength -= (outLength + 3) / 4;
|
| - if (!outLength)
|
| - return false;
|
| -
|
| - unsigned sidx = 0;
|
| - unsigned didx = 0;
|
| - if (outLength > 1) {
|
| - while (didx < outLength - 2) {
|
| - out[didx] = (((out[sidx] << 2) & 255) | ((out[sidx + 1] >> 4) & 003));
|
| - out[didx + 1] = (((out[sidx + 1] << 4) & 255) | ((out[sidx + 2] >> 2) & 017));
|
| - out[didx + 2] = (((out[sidx + 2] << 6) & 255) | (out[sidx + 3] & 077));
|
| - sidx += 4;
|
| - didx += 3;
|
| - }
|
| + out.grow(length);
|
| +
|
| + unsigned equalsSignCount = 0;
|
| + unsigned outLength = 0;
|
| + bool hadError = false;
|
| + for (unsigned idx = 0; idx < length; ++idx) {
|
| + UChar ch = data[idx];
|
| + if (ch == '=') {
|
| + ++equalsSignCount;
|
| + // There should never be more than 2 padding characters.
|
| + if (policy == Base64ValidatePadding && equalsSignCount > 2) {
|
| + hadError = true;
|
| + break;
|
| + }
|
| + } else if (('0' <= ch && ch <= '9') || ('A' <= ch && ch <= 'Z') || ('a' <= ch && ch <= 'z') || ch == '+' || ch == '/') {
|
| + if (equalsSignCount) {
|
| + hadError = true;
|
| + break;
|
| + }
|
| + out[outLength++] = base64DecMap[ch];
|
| + } else if (!shouldIgnoreCharacter || !shouldIgnoreCharacter(ch)) {
|
| + hadError = true;
|
| + break;
|
| + }
|
| + }
|
| +
|
| + if (outLength < out.size())
|
| + out.shrink(outLength);
|
| +
|
| + if (hadError)
|
| + return false;
|
| +
|
| + if (!outLength)
|
| + return !equalsSignCount;
|
| +
|
| + // There should be no padding if length is a multiple of 4.
|
| + // We use (outLength + equalsSignCount) instead of length because we don't want to account for ignored characters.
|
| + if (policy == Base64ValidatePadding && equalsSignCount && (outLength + equalsSignCount) % 4)
|
| + return false;
|
| +
|
| + // Valid data is (n * 4 + [0,2,3]) characters long.
|
| + if ((outLength % 4) == 1)
|
| + return false;
|
| +
|
| + // 4-byte to 3-byte conversion
|
| + outLength -= (outLength + 3) / 4;
|
| + if (!outLength)
|
| + return false;
|
| +
|
| + unsigned sidx = 0;
|
| + unsigned didx = 0;
|
| + if (outLength > 1) {
|
| + while (didx < outLength - 2) {
|
| + out[didx] = (((out[sidx] << 2) & 255) | ((out[sidx + 1] >> 4) & 003));
|
| + out[didx + 1] = (((out[sidx + 1] << 4) & 255) | ((out[sidx + 2] >> 2) & 017));
|
| + out[didx + 2] = (((out[sidx + 2] << 6) & 255) | (out[sidx + 3] & 077));
|
| + sidx += 4;
|
| + didx += 3;
|
| }
|
| + }
|
|
|
| - if (didx < outLength)
|
| - out[didx] = (((out[sidx] << 2) & 255) | ((out[sidx + 1] >> 4) & 003));
|
| + if (didx < outLength)
|
| + out[didx] = (((out[sidx] << 2) & 255) | ((out[sidx + 1] >> 4) & 003));
|
|
|
| - if (++didx < outLength)
|
| - out[didx] = (((out[sidx + 1] << 4) & 255) | ((out[sidx + 2] >> 2) & 017));
|
| + if (++didx < outLength)
|
| + out[didx] = (((out[sidx + 1] << 4) & 255) | ((out[sidx + 2] >> 2) & 017));
|
|
|
| - if (outLength < out.size())
|
| - out.shrink(outLength);
|
| + if (outLength < out.size())
|
| + out.shrink(outLength);
|
|
|
| - return true;
|
| + return true;
|
| }
|
|
|
| -bool base64Decode(const char* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy)
|
| -{
|
| - return base64DecodeInternal<LChar>(reinterpret_cast<const LChar*>(data), length, out, shouldIgnoreCharacter, policy);
|
| +bool base64Decode(const char* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy) {
|
| + return base64DecodeInternal<LChar>(reinterpret_cast<const LChar*>(data), length, out, shouldIgnoreCharacter, policy);
|
| }
|
|
|
| -bool base64Decode(const UChar* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy)
|
| -{
|
| - return base64DecodeInternal<UChar>(data, length, out, shouldIgnoreCharacter, policy);
|
| +bool base64Decode(const UChar* data, unsigned length, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy) {
|
| + return base64DecodeInternal<UChar>(data, length, out, shouldIgnoreCharacter, policy);
|
| }
|
|
|
| -bool base64Decode(const String& in, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy)
|
| -{
|
| - if (in.isEmpty())
|
| - return base64DecodeInternal<LChar>(0, 0, out, shouldIgnoreCharacter, policy);
|
| - if (in.is8Bit())
|
| - return base64DecodeInternal<LChar>(in.characters8(), in.length(), out, shouldIgnoreCharacter, policy);
|
| - return base64DecodeInternal<UChar>(in.characters16(), in.length(), out, shouldIgnoreCharacter, policy);
|
| +bool base64Decode(const String& in, Vector<char>& out, CharacterMatchFunctionPtr shouldIgnoreCharacter, Base64DecodePolicy policy) {
|
| + if (in.isEmpty())
|
| + return base64DecodeInternal<LChar>(0, 0, out, shouldIgnoreCharacter, policy);
|
| + if (in.is8Bit())
|
| + return base64DecodeInternal<LChar>(in.characters8(), in.length(), out, shouldIgnoreCharacter, policy);
|
| + return base64DecodeInternal<UChar>(in.characters16(), in.length(), out, shouldIgnoreCharacter, policy);
|
| }
|
|
|
| -String base64URLEncode(const char* data, unsigned length, Base64EncodePolicy policy)
|
| -{
|
| - return base64Encode(data, length, policy).replace('+', '-').replace('/', '_');
|
| +String base64URLEncode(const char* data, unsigned length, Base64EncodePolicy policy) {
|
| + return base64Encode(data, length, policy).replace('+', '-').replace('/', '_');
|
| }
|
|
|
| -String normalizeToBase64(const String& encoding)
|
| -{
|
| - return String(encoding).replace('-', '+').replace('_', '/');
|
| +String normalizeToBase64(const String& encoding) {
|
| + return String(encoding).replace('-', '+').replace('_', '/');
|
| }
|
|
|
| -} // namespace WTF
|
| +} // namespace WTF
|
|
|