Language support

The ten language engines, the frameworks they model, and their known limitations.

Ten languages have their own parser and data-flow engine. The number of rules each reports comes straight from the engine registry; the coverage matrix lists them.

LanguageFilesData-flow rulesFrameworks modelled
TypeScript / JavaScript
Cross-file flows through functions, classes & services
.ts .tsx .js19Express, NestJS, Next.js, browser DOM
Python
Cross-module flows through functions & service classes
.py22Flask, FastAPI, Django
Java
Spring services across files, entry-point aware
.java24Servlet, Spring, JAX-RS
Kotlin
Spring, Ktor, Servlet & Android; shares cross-file evidence with Java
.kt .kts23Spring, Ktor, Android
Go
Cross-package calls, path-sensitive, BOLA dominance
.go27net/http, Gin, Echo, Fiber
C#
Services across files, path-sensitive, BOLA dominance
.cs21ASP.NET Core, System.Web (HttpRequest)
PHP
Functions & classes across includes, path-sensitive
.php23Laravel, plain PHP, WordPress nonces
Ruby
Rails & Sinatra controllers to services and models
.rb20Rails, Sinatra
Rust
Axum, Actix & Rocket handlers, format! traced per argument
.rs26Axum, Actix, Rocket, Warp
C / C++
argv, stdin, sockets & getenv into exec, printf formats, SQL and files
.c .cpp .cc .h .hpp16POSIX, Win32, C++ STL streams

Status definitions

  • Implemented — the language engine reports the rule (it is in the engine's registry, which is checked at compile time).
  • Measured — the language also has a public labelled corpus in the benchmark report, with per-category results.
  • Not covered — the engine does not report the rule; line-level detectors may still flag simple cases.

Measured languages: Go, PHP, Ruby, Rust, Java, Python, C#, C, C++, TypeScript. Kotlin has no public labelled corpus and is tested with unit tests and deliberately vulnerable apps.

Known limitations

  • TypeScript: Dynamic property access with computed keys is followed only for literal keys.
  • TypeScript: Framework middleware defined in another package is not inspected.
  • Python: Values that pass through dynamically imported modules (importlib) are not followed.
  • Java: An algorithm or setting read from a .properties file is not resolved (BenchmarkJava hash cases).
  • Java: Values built by a factory in another file are followed only through its return value, not its internal state.
  • Kotlin: There is no public labelled Kotlin corpus, so Kotlin results are not in the benchmark report; the engine is tested with unit tests and the AndroGoat / Goatlin apps.
  • Go: Code-injection and authentication-failure categories score low on the sast-benchmark corpus.
  • C#: A value returned by a method in another file is not yet used to prove a sink safe (Juliet flow variant 61), which leaves some SQL false positives.
  • PHP: XXE detection follows libxml's real defaults, so many corpus cases labelled vulnerable are not flagged.
  • PHP: CSRF is reported only for session- or cookie-authenticated handlers.
  • Ruby: XXE: we follow Nokogiri's real defaults (external entities off), which disagrees with the corpus labels.
  • Ruby: Authorization-failure and CSRF categories are not detected yet.
  • Rust: Calls to a project's own stub HTTP clients or deserializers are not treated as sinks; only real library calls are.
  • C/C++: Macros defined in headers that are not in the scan are not expanded.
  • C/C++: rand() is not treated as attacker input, so Juliet's rand-sourced cases are not flagged.

Next: Vulnerability coverage →