It Will Never Work in Theory: Live!

Our next set of online lightning talks is happening April 25-26, 2023. Check out the speakers and get your ticket now!

Can Static Analysis Tools Find More Defects?

Reviewed by Greg Wilson / 2023-03-08
Keywords: Static Analysis

Sorting algorithms, data compression, engine efficiency: in each case, we can compare how well we're doing to a provable optimum. In situations where we don't know what the upper bound is, we can still sometimes estimate how much room there is for improvement. This paper does that for static analysis tools that look for patterns (or anti-patterns) in code that indicate bugs. By reformulating issues found in manual code review as rules, the authors show that static analysis might be able to detect as much as three quarters of all bugs, which is considerably better than what current-generation linters do. Work like this can guide development of better tools, but it also tells us how much we still have to gain from them.

Sahar Mehrpour and Thomas D. LaToza. Can static analysis tools find more defects? Empirical Software Engineering, Nov 2022. doi:10.1007/s10664-022-10232-4.

Static analysis tools find defects in code, checking code against rules to reveal potential defects. Many studies have evaluated these tools by measuring their ability to detect known defects in code. But these studies measure the current state of tools rather than their future potential to find more defects. To investigate the prospects for tools to find more defects, we conducted a study where we formulated each issue raised by a code reviewer as a violation of a rule, which we then compared to what static analysis tools might potentially check. We first gathered a corpus of 1323 defects found through code review. Through a qualitative analysis process, for each defect we identified a violated rule and the type of Static Analysis Tool (SAT) which might check this rule. We found that SATs might, in principle, be used to detect as many as 76% of code review defects, considerably more than current tools have been demonstrated to successfully detect. Among a variety of types of SATs, Style Checkers and AST Pattern Checkers had the broadest coverage of defects, each with the potential to detect 25% of all code review defects. We found that static analysis tools might be able to detect more code review defects by better supporting the creation of project-specific rules. We also investigated the characteristics of code review defects not detectable by traditional static analysis techniques, which to detect might require tools which simulate human judgements about code.