A Fault Detecting Diagnostic Tool for Python-driven Multi-language Scienti耀c Code

Traditional debugging tools are limited in their ability to relate fault exceptions, such as numerical or memory errors, to their sources in a multi-language application, such as those being developed under the CREATE program. A new tool for multi-language scienti�耀c computing applications driven by Python is presented that reports useful diagnostic and performance information when the program experiences some form of anomalous operation. Built upon the TAU performance system, it operates across all code layers and generates a diagnostic �耀le containing information about memory usage, call-stack, and input/output (I/O) gathered during execution. The tool is useful to both software developers as well as users who experience software issues, as it provides a way to exchange execution information between the user and the development team without requiring disclosure of potentially-sensitive application data. Furthermore, it provides support for retaining performance measurements and other execution information that would otherwise be lost.