Mining Software Specifications

Software is a ubiquitous component in our daily life. It ranges from large software systems like operating systems to small embedded systems like vending machines, both of which we frequently interact with. Reducing software related costs and ensuring correctness and dependability of software are certainly worthwhile goals to pursue. Due to the short-time-to-market requirement imposed on many software projects, documented software specifications are often lacking, incomplete and outdated (Deelstra, Sinnema & Bosch 2004). Lack of documented software specifications contributes to difficulties in understanding existing systems. The latter is termed program comprehension and is estimated to contribute up to 45% of total software cost which goes to billions of dollars (Erlikh 2000, Standish 1984; Canfora & Cimitile 2002; BEA 2007). Lack of specifications also hampers automated effort of program verification and testing (Ammons, Bodik & Larus 2002). One solution to address the above problems is mining (or automatic extraction of) software specification from program execution traces. Given a set of program traces, candidate partial specifications pertaining to the behavior a piece of software obeys can be mined. In this chapter, we will describe recent studies on mining software specifications. Software specification mining has been one of the new directions in data mining (Lo, Khoo & Liu 2007a, Lo & Khoo 2007). Existing specification mining techniques can be categorized based on the form of specifications they mine. We will categorize and describe specification mining algorithms for mining five different target formalisms: Boolean expressions, automata (Hopcroft, Motwani & Ullman 2001), Linear Temporal Logic (Huth & Ryan 2003), frequent patterns (Han & Kamber 2006) and Live Sequence Charts (Harel & Marelly 2003).