CodeFuse-Query Documentation

CodeFuse-Query (also referred to as CodeQuery) is a centralized, data-centric static code analysis platform designed for large-scale software development. Unlike traditional tool-centric static analysis tools (like Clang Static Analyzer or PMD) that are often purpose-built for specific objectives, CodeFuse-Query treats source code and analysis results as data and the execution process as big data processing.

Key Characteristics

• Highly Scalable: Capable of handling massive codebases and adapting to varying analysis needs (e.g., security-focused taint analysis vs. productivity-focused cyclomatic complexity metrics).
• Data-Centric: Views code as a highly structured, interconnected dataset, making it suitable for environments where codebases evolve incrementally.
• Highly Integrated: Designed to integrate with enterprise-grade big data and cloud infrastructure, including:
  • Data warehouses
  • Data computation facilities (e.g., MaxCompute, Hive)
  • OSS object storage
  • Flexible computing resources (e.g., Kubernetes)
• Supports Diverse Needs: Accommodates Quality of Service (QoS) analysis, cross-language analysis, and varying algorithmic/performance requirements.

The COREF (Code Relation) data model represents code structures as relations. The available relations vary depending on the programming language being analyzed. Use these relations to build queries that traverse code hierarchies, identify declarations, and map locations.

Java Relations

• field(id, tid, name, locId): A field with its ID, type ID, name, and location ID.
• method(id, tid, name): A method with its ID, type ID, and name.
• parent(id1, id2): A parent-child relationship between two entities.
• class(id, name): A class definition.
• interface(id, name): An interface definition.
• type(id, n): A type definition.
• javaLoc(locId, s, e, path): Location details (start line, end line, file path).

XML Relations

• xmlEle(): XML element.
• xmlAttri(): XML attribute.

JS/TS Relations

• enclose(exprId, methodId): An expression enclosed within a method.
• callExpr(exprId, methodId, argExprId, callSiteId): A function call expression.
• reference(refId, defId): A reference to a definition.
• service(id, appId, locId): A service associated with an app and location.
• jsLoc(locId, s, e, path): Location details (start line, end line, file path).
• app(id, name): An application definition.
• api(id, apiName, apiSignature, appId): An API definition.
• functionLikeDeclaration(id, name): A function-like declaration.
• parent(id1, id2): Parent-child relationship.
• class(id, name): Class definition.

// Example Java Relations
field(id, tid, name, locId)
method(id, tid, name)
parent(id1, id2)
class(id, name)
interface(id, name)
type(id, n)
javaLoc(locId, s, e, path)

// Example JS/TS Relations
enclose(exprId, methodId)
callExpr(exprId, methodId, argExprId, callSiteId)
reference(refId, defId)
service(id, appId, locId)
jsLoc(locId, s, e, path)
app(id, name)
api(id, apiName, apiSignature, appId)
functionLikeDeclaration(id, name)
parent(id1, id2)
class(id, name)

CodeFuse-Query uses a custom Domain Specific Language (DSL) called Gödel to perform code analysis on the COREF data.

Key Characteristics of Gödel:

• Logic-based: It is a logic reasoning language based on Datalog. It works by describing "facts" and "rules" to derive new facts.
• Declarative: Users describe what they want to find rather than how to compute it, leaving the implementation to the engine.
• Performance: By inheriting Datalog's properties, it offers monotonicity and termination, providing higher performance for complex queries compared to standard SQL.

Comparison:

• vs. SDK: Gödel is easier to learn and use because it is declarative; users don't need to manage intermediate computation steps.
• vs. SQL: Gödel has stronger expressive power and faster computation for complex tasks like recursive algorithms and multi-table joins, which are difficult in SQL.