Donald Bren School of Information and Computer Sciences

A common paradigm for the development of process-control or embedded computer software is to do most of the implementation and testing on a large host computer, then retarget the code for final checkout and production execution on the target machine. The host machine is usually large and provides a variety of program development tools, while the target may be a small, bare machine. A difficulty with the paradigm arises when the software developed has real-time constraints and is composed of multiple communicating processes. If a test execution on the target fails, it may be exceptionally tedious to determine the cause of the failure. Host machine debuggers cannot normally be applied, because the same program processing the same data will frequently exhibit different behavior on the host. Differences in processor speed, scheduling algorithm, and the like, account for the disparity. This paper proposes a partial solution to this problem, in which the errant execution reconstructed and made amenable to source language level debugging on the host. The solution involves the integrated application of a static concurrency analyzer, an interactive interpreter, and a graphical program visualization aid. Though generally applicable, the solution is described here in the context of multi-tasked real-time Ada* programs.