Hi Team, While going through Blink's commit, I came across following refactoring / optimization: Blink Commit - https://chromium.googlesource.com/chromium/blink/+/f9411437595a5782eb1adba1f719d4448530bd93 WebKit Source - https://searchfox.org/wubkat/source/Source/WebCore/svg/SVGPathByteStreamSource.h#57 ''' This avoids having the compiler beating around the bush trying to recognize the memcpy, and results in smaller code (roughly 50% smaller measured for GCC 4.8). All at the affordable price of upsetting a few language purists. ''' Appreciate if someone can share input. Thanks!
<rdar://problem/104552389>
I think there is no much saving in the Blink's commit. This function just reads bool, short or float. So the for-loop in SVGPathByteStreamSource::readType() runs 1, 2 or 4 times. In my opinion using memcpy is just a cleaner solution and nice refactoring but it is not a perf or code size optimization. So if we want to clean this area, I would suggest something like that instead: // SVGPathByteStreamSource.h template<typename DataType> DataType readType() { DataType data; size_t dataSize = sizeof(DataType); ASSERT(m_streamCurrent + dataSize <= m_streamEnd); memcpy(&data, m_streamCurrent, dataSize); m_streamCurrent += dataSize; return data; } // SVGPathByteStreamBuilder.h template<typename DataType> void writeType(DataType data) { union { DataType value; unsigned char bytes[sizeof(DataType)]; } ByteType; ByteType type = { data }; size_t typeSize = sizeof(ByteType); for (size_t i = 0; i < typeSize; ++i) m_byteStream.append(type.bytes[i]); } And we can get rid of BoolByte, FloatByte and UnsignedShortByte.
Pull request: https://github.com/WebKit/WebKit/pull/17264
Committed 267523@main (e2aa0c20fc47): <https://commits.webkit.org/267523@main> Reviewed commits have been landed. Closing PR #17264 and removing active labels.