[Paper Review] M3: An Open Model for Measuring Code Artifacts

Anastasia Izmaylova, Paul Klint+2 more|arXiv (Cornell University)|Dec 4, 2013

Software Engineering Research9 citations

TL;DR

M³ proposes an open, extensible, and composable model for representing source code artifacts using immutable, typed relations and URI-based source locations. Built within the Rascal programming language, it enables language-agnostic code analysis by decoupling parsers from metrics consumers through a standardized, serializable model that supports type inference, source tracing, and incremental composition of code metrics across projects and dependencies.

ABSTRACT

This document details design considerations of M3: a meta model for source code artifacts

Motivation & Objective

• To address the challenge of measuring code quality across diverse programming languages and metrics by creating a unified, extensible model for code artifacts.
• To decouple code extractors (parsers) from metrics and visualization tools through a standardized, language-agnostic model.
• To enable precise traceability of metrics back to source code via physical and logical URI-based locations.
• To support incremental and composable analysis of code across files, projects, and dependencies using immutable, persistent data structures.
• To provide a foundation for reusable, extensible code analysis pipelines in software analytics and evolution research.

Proposed method

• M³ models are built as immutable, typed relations between source code artifacts, using Rascal’s native support for sets, relations, and pattern matching.
• Source locations are modeled as URIs with schemes for physical (e.g., file://, project://) and logical (e.g., java+class://) identifiers, enabling traceability and stability across code movements.
• Core relations include containment (artifact hierarchy), declarations (mapping logical to physical locations), and uses (dependency tracking).
• Abstract syntax trees (ASTs) are reconstructed on-demand from model relations, with nodes annotated by source location and type information.
• Type symbols are modeled as first-class values to support parametric type reasoning and metric computation over generic types.
• Models are composed via explicit union and link operations: union fuses relations, while link updates authority fields to enable cross-project referencing.

Experimental results

Research questions

• RQ1 How can a single, extensible model represent source code artifacts across multiple programming languages while preserving traceability to source code?
• RQ2 What design principles enable decoupling of code extractors from metrics and visualization tools in a language-agnostic way?
• RQ3 How can source location identifiers be designed to be both human-readable and stable under non-functional code changes?
• RQ4 What mechanisms support incremental and composable analysis of code metrics across files, projects, and dependencies?
• RQ5 How can type systems be modeled uniformly to support metrics involving parametric types and type instantiation?

Key findings

• M³ enables the construction of language-agnostic code analysis pipelines by standardizing on a minimal set of algebraic sorts (Expression, Declaration, Statement, Type, Modifier) for abstract syntax.
• The use of URI-based source locations allows precise, human-readable, and stable linking of derived metrics back to source code, supporting debugging and explanation of results.
• The model supports incremental analysis: volume metrics can be computed before fusing external dependencies, while depth-of-inheritance metrics require post-fusion access to all declarations.
• Type symbols are modeled as first-class values, enabling computation of metrics such as the number of possible instantiations of a generic type.
• The model has been successfully applied in academic courses and research projects at UvA and CWI, demonstrating usability in real-world software analytics workflows.
• Initial support for Java, C#, and PHP is underway, with extensions to control flow and program dependence relations planned.