Skip to main content
mehen supports eleven source languages, Markdown, and SQL. Per-file language detection is by extension, and the matching analyzer crate owns parsing and metric interpretation. Each language uses the parser that gives mehen the best semantic coverage for that ecosystem.

​
Source languages

TypeScript is a superset of JavaScript, so mehen analyzes .js / .mjs / .cjs through the Oxc TypeScript front-end and .jsx through Oxc’s TSX front-end.

​
Why these parsers

  • Ruff parses Python with full support for current syntax (3.13/3.14 features, f-strings, match/case, exception groups, async constructs) and exposes a typed AST plus a semantic model.
  • Oxc is a JavaScript/TypeScript toolchain in Rust and one of the fastest TS parsers in production. It handles decorators, class fields, parameter properties, JSX, satisfies, using, and dynamic import natively.
  • Mago is a PHP toolchain in Rust. It understands attributes, promoted properties, enums, traits, anonymous classes, readonly members, null-safe calls, and match expressions out of the box.
  • Prism is the official Ruby parser maintained by the Ruby core team. It covers blocks, lambdas, numbered parameters, modifier conditionals, rescue modifiers, endless methods, pattern matching, and safe navigation.
  • ra_ap_syntax is the syntax tree library used by rust-analyzer; it gives mehen exactly the syntactic granularity rust-analyzer uses for its own analyses.
  • ANTLR parses Kotlin, Java, and C#, generated to Rust by ophi-dev/antlr-rust-runtime. Kotlin uses the official Kotlin specification grammar, which models Kotlin constructs (when entries, the elvis ?: and safe-call ?. operators, catch blocks, labeled jumps, property accessors) as first-class, semantically-named rules. Java uses the community grammars-v4 Java grammar, which covers modern Java (records, sealed types, switch expressions, text blocks, pattern matching, modules). C# uses Roslyn’s own generated grammar, which the C# compiler generates from the same Syntax.xml model that produces its syntax node classes — so it tracks C# as implemented: records, is not, and/or and relational patterns, list patterns, collection expressions, raw strings, primary constructors, required members, and the C# 14 additions (field, extension blocks). It replaced the community grammars-v4 C# grammar, which stops at C# 7 and parsed 93 of 322 System.Text.Json files against 317 for the derived Roslyn grammar; see crates/mehen-csharp-parser/grammar/PROVENANCE.md for the pinned revision and what the derivation repairs. All three are richer than a tree-sitter CST.
  • Tree-sitter is mehen’s pick for Go, C, and PowerShell, where its grammar quality and ecosystem maturity make it the best fit.

​
Documentation

Markdown gets a dedicated documentation metric suite — code-style metrics like cyclomatic complexity and NOM/NPA/NPM/WMC do not apply because Markdown has no functions or classes. See Markdown metrics for the full set.

​
SQL

A dedicated SQL analyzer (mehen-sql) ships its own metric family — query-block structure, CTE graphs, join graphs, predicate complexity, object-touch risk, dialect portability, and an SQL-flavored Halstead. Code-style metrics like cyclomatic complexity and NOM/NPA/NPM/WMC do not apply to declarative SQL. The dialect is inferred from syntax hints (postgres, T-SQL, snowflake, bigquery, mysql, sqlite, oracle, clickhouse, redshift, sparksql, hive, athena, db2) with a reported confidence, defaulting to ANSI. See SQL metrics overview for the full set.

​
Polyglot monorepos

mehen runs per-file language detection over any directory tree. Pass multiple paths to limit which trees are walked and let mehen pick supported languages from each:
The GitHub Action accepts the same multi-path input.