On 12/4/2010 4:33 PM, Peter Foelsche wrote:
"Jeffrey Lee Hellrung, Jr." <jhellrung@ucla.edu> wrote in message news:4CFADA65.40708@ucla.edu...
I think the point is that copy-on-write is rarely optimal, often suboptimal, and increases complexity; hence it should not be encouraged.
increases complexity -- wrong -- I only replaced the matching types -- the using code was not modified. The number of lines for the wrapper are minimal -- I only implemented the methods I'm using. I don't care about complexity the compiler has to manage. I care about the complexity I have to manage. I could introduce some filtering to avoid storing identical instances which would even more decrease the memory foot print.
Rather than address these points directly, I will simply refer to Herb Sutter's discussions [1,2,3,4] on copy-on-write. I'm not claiming that copy-on-write is never a good solution, only that it should not be one's knee-jerk reaction (to say the least) to reducing unnecessary copies. There are solutions that, much more often than not, are superior that one should explore first, especially now with emulated move semantics or true move semantics. [1] http://www.gotw.ca/gotw/043.htm [2] http://www.gotw.ca/gotw/044.htm [3] http://www.gotw.ca/gotw/045.htm [4] http://www.gotw.ca/publications/optimizations.htm - Jeff