Talk:Code Racer

I've gone ahead and created this Code Racer page -- I think the approach is complete enough to be worth discussing - e.g., I get insight into llGetSubString snippets. However, something fiddly is still wrong with the implementation, I guess. As you can see in the sample results, often the sums of wins exceed the scale. -- Ppaatt Lynagh 18:08, 16 October 2007 (PDT)

Ha! Those new insights were half-grounded in nonsense! Half the win counts were wrong, courtesy one typo. Fixed now, maybe. -- Ppaatt Lynagh 21:18, 16 October 2007 (PDT)

Talk:Hex suggests that we should tweak this instrument to call llGetTime rather than llGetTimestamp, in order to more accurately measure elapsed run time when time dilates. -- Ppaatt Lynagh 21:41, 19 October 2007 (PDT)

Change complete! Runs wonderfully faster too. The run time cost of the harness has dropped below two milliseconds. -- Ppaatt Lynagh 07:32, 20 October 2007 (PDT)

This should run indefinitely many versions of code against each other in parallel, so that we can complete an analysis like comparing all the too many versions of code at hex without attending the experiment. Also the last version of code could be an empty block of code, to measure the overhead of the harness, to throw it out, like the Efficiency Tester does. -- Ppaatt Lynagh 13:14, 22 October 2007 (PDT)