No abstract

Elusive bugs are difficult to observe and to reproduce. They are often caused by non-deterministic or unexpected events, inputs or computations. To track elusive bugs, it is necessary to narrow down the scope of the bug investigation. This helps reproducing and observing very specific aspects of the program's state and behavior. In object-oriented programs, it often comes down to finding and debugging specific objects. However, some objects are particularly hard to find. Typical hard cases are finding a temporary object or a single particular instance of a given class. This is the case when debugging UI elements, or programs with non-deterministic state. This capability of identifying objects of interest is crucial to the debugging of objects. Yet, debuggers addressing this problem only provide manual or limited ways to find such objects. In this paper we present Object Miners: a non-intrusive object-centric debugging approach for acquiring, capturing and replaying objects. We show how the miners acquire objects at run time from the sub-results of an instrumented expression. Miners capture objects with their execution context and replay them to freeze strategic parts of the execution, eliminating non-determinism. We present a Pharo implementation of Object Miners along with a performance and memory overhead evaluation. We present a debugger built on top of Object Miners, and demonstrate through a series of examples how the application of Object Miners facilitates the tracking of elusive bugs. These examples include the fixing of a non-deterministic bug in an IOT application, tooling support for object-centric debugging, and the tracking of a bug in a real-world program.