A GSoC 2013 Proposal Enhancing Giri: Dynamic Slicing in LLVM

Program slice contains all statements in a program that directly or indirectly act the value of a variable occurrence [14], the criteria of which is a pair of statement and variables. We can further narrow the notion of slice, which contains statements that influence the value of a variable occurrence for special program inputs. This is referred as dynamic program slicing [1]. It works on a single execution and outputs the executed statements (traces) relevant to the slicing criterion. There are many applications that use (or could benefit from) dynamic slicing, both by research and industry organizations (e.g. Microsoft, IBM). For example, it’s long been used in software debugging model [2, 5] and testing [3]. Sahoo et. al. from UIUC use dynamic program slicing to generate likely invariants for automated software fault localization [10]. Differential slicing [6] was a joint work by UC Berkeley, IMDEA Software Institute and CMU. It uses dynamic slicing to establish the sequence of value differences that affect the target, which can help them identifying causal execution differences for security applications. Gupta et. al. from the University of Arizona employed dynamic program slicing to narrow down the search for faulty code [4]. Our group from Tsinghua University would benefit from dynamic slicing in automatically finding manual configuration errors [15] in software deployment (see below). Recently, researchers from National University of Singapore and Microsoft used dynamic slicing to debug evolving programs [8]. Dennis Jeffrey from Google built a system for debugging via online tracing and dynamic slicing with other guys from university [7]. Researchers from IBM Research[9] who work on fault localization for data-centric programs, split the trace into multiple slices by applying dynamic slicing.