Smart contract metrics: a first study

Abstract

Smart contracts (SC) are software programs that reside and run over a blockchain. The code can be written in different languages with the common purpose of implementing various kinds of transactions onto the hosting blockchain. They are ruled by the blockchain infrastructure with the intent to automatically implement the typical conditions of traditional contracts. Programs must satisfy context-dependent constraints which are quite different from traditional software code. In particular, since the bytecode is uploaded in the hosting blockchain, the size, computational resources, interaction between different parts of the program are all limited. This is true even if the specific programming languages implement more or less the same constructs as that of traditional languages: there is not the same freedom as in normal software development. The working hypothesis used in this article is that Smart Contract specific constraints should be captured by specific software metrics (that may differ from traditional software metrics). We tested this hypothesis on 85K Smart Contracts written in Solidity and uploaded on the Ethereum blockchain. We analyzed Smart Contracts from two repositories “Etherscan” and “Smart Corpus” and we computed the statistics of a set of software metrics related to Smart Contracts and compared them to the metrics extracted from more traditional software projects. Our results show that generally, Smart Contract metrics have more restricted ranges than the corresponding metrics in traditional software systems. Some of the stylized facts, like power law in the tail of the distribution of some metrics, are only approximate but the lines of code follow a log-normal distribution which reminds us of the same behaviour already found in traditional software systems.