Modeling The Heap: A Practical Approach

Опубликовано 17 августа 2016, 20:34

At the heart of most software engineering and optimization applications is some model of runtime program behavior. The widespread adoption of object-oriented programming and memory managed languages has lead to programs that make extensive use of rich pointer structures and that use data as a central part of organizing the program structure. Thus, understanding the memory state of the program is an increasingly critical part of understanding the behavior of the program. The increasing difficulty (and importance) of understanding the behavior of a program is occurring at the same time as increasing program complexity, the growing need for these programs to take advantage of parallel hardware (plus complex memory hierarchies), and the decreasing tolerance for software defects. This talk will introduce a model we have developed for representing the program heap and the classes of information that can be represented with this model (shape, regions, heap use/mod, reachabliltiy, collection/iterator state, sharing, nullity, and escape information). Using this model we will briefly demonstrate how it can be used for static analysis and debugger support. The static analysis is currently able to handle the majority of the managed .net bytecode and can analyze programs up to 15KLoc in a few minutes and a ~200MB of memory (which is exceptionally fast for shape analysis techniques). These results can the be used directly to aid program understanding, for generating rich type signatures, in optimization applications, or as the basis for other analysis tools. I will also demonstrate how the model can be used in dynamic applications, such as a debugger where we use the model to produce compact visualizations of the heap to aid the programmer in understanding the program state.