The extended whole program path (eWPP) representation is described in this paper. It captures both the control flow history and data dependence history of program execution. The objective is to capture data dependence history that cannot be recovered from the control trace. For this, the authors use disambiguation checks.
The approach is described clearly and thoroughly with test data. A positive result is the reduced size (one-fourth) of eWPP, as compared to combined compressed control flow and address traces. The runtime overhead (five times) for their collection makes the approach not universally applicable--keeping in mind that this is an already optimized solution (for example, the authors use binary search instead of linear search).
The use of tools, such as Phoenix CF, can automate instrumentation code insertion. Overall, there is no clear statement of benefits from the approach. Specifically, what is the trade-off between smaller size versus high-runtime overhead?