Trace-driven simulation is a widely used technique to evaluate microprocessor architecture and/or memory hierarchy performance. However, software trace collection induces significant execution slowdown. As on-line simulation induces an even more disastrous slowdown, simulating applications incurring trillions of instructions can not be considered without trace sampling. Therefore continuous software trace collection of the complete workload of a desktop computer cannot be envisaged. The Code Cloning Tracing method is aimed at allowing trace collection on time consuming applications. Sampling the execution is performed through a static duplication and instrumentation of the source code. A first copy of the code is nearly identical to the original code and will execute most of the time. The second code copy is further instrumented by the trace user and will be parsimoniously executed in order to collect traces without slowing down too much the overall application. At execution time, some events make the execution switch to the instrumented code copy. Other events make the execution return to the original-like code copy. The choice of these events determines the execution overhead, therefore making the approach very flexible. This paper presents in details the Code Cloning Tracing method. We also present calvin, the first prototype implemented to validate the approach. Experiments show that Code Cloning Tracing could be applied on all applications running on desktop computer without significantly affecting user response time when traces are not collected. This opens the way to punctual (let us say one second of CPU) collection of the whole activity of a desktop computer.