ripex

A language-agnostic parser and fact-extraction library with a CLI. ripex has seven hand-written recursive-descent parser families and eight language modes (JavaScript and TypeScript are distinct modes), built around a common pipeline:

Lexer → Parser → AST → Facts extraction.

ripex is not a compiler, linker, runtime, or type checker. The optional check command delegates validation to the language's existing production toolchain; it does not reimplement compiler semantics.

Ripex v0.3.0

Ripex v0.3.0 was published on 2026-08-01: crates.io package · docs.rs API documentation · GitHub release.

Supported languages

Language	Extensions	Structural parser	Optional external validation
JavaScript	.js .jsx .mjs .cjs	yes	Node + TypeScript checkJs
TypeScript	.ts .tsx .mts .cts	yes	TypeScript tsc --noEmit --strict
Python	.py .pyi	yes	CPython compile + mypy --strict
Go	.go	yes	go build
Rust	.rs	yes	rustc / cargo check
C	.c .h	yes	GCC / Clang, strict C17 by default; compile_commands.json projects
C++	.cpp .hpp .cc .cxx .hh .hxx	yes	GCC / Clang, strict C++20 by default; compile_commands.json projects
C#	.cs	yes	dotnet build / csc

The in-process parsers are resilient structural parsers optimized for high-fidelity source facts, not replacement compiler front ends. Name resolution, project loading, linking, SDK behavior, and type semantics remain the responsibility of each language's production toolchain. A missing toolchain is reported as unavailable and never treated as success.

All eight language modes parse the complete checked-in corpus without diagnostics. CI also compiles/type-checks the corresponding real fixture projects and requires every language to extract symbols, imports, calls, and variables, with no parser panics or hangs.

The checked-in corpus and curated evidence report establish structural parsing and fact-extraction behavior only. They do not establish compiler-level semantic equivalence; use ripex check and the production-toolchain conformance gate for that separate question. See docs/BENCHMARKS.md for measured limits and docs/TESTING.md for the evidence and compiler gates.

Each language pipeline produces four fact types:

• Symbols — functions, classes, structs, enums, traits, interfaces, methods, constants, fields
• Imports — use, import, #include, using, from … import, etc.
• Calls — function calls, method calls, constructor calls, path calls
• Variables — local bindings, parameters, statics, constants with type info

JavaScript and TypeScript also retain hashbang, line, and block comments with precise source spans through ParseResult::comments, including comments at end-of-file. Type-only imports/re-exports, dynamic imports, optional calls, awaited calls, generic call arguments, and destructured bindings are emitted as first-class facts.

The public fact payloads retain the detail needed for lossless downstream indexing: symbol parameters/defaults/docstrings/attributes and type metadata; import specifiers and re-export metadata; awaited, optional, receiver, and generic call metadata; plus variable scope, usage-site, and storage metadata. Consumers can serialize these native facts with the serde feature, then build their own cross-file relationships without discarding parser provenance.

Library

use ripex::{detect_language, parser_for_ext};

// Auto-detect language from extension.
let language = detect_language("src/lib.rs").unwrap();
let parser = parser_for_ext(language.id(), "rs").unwrap();
let result = parser.parse("pub fn add(a: i32, b: i32) -> i32 { a + b }");
let facts = parser.extract(&result);

println!("{} symbols", facts.symbols.len());

External toolchain validation is a separate explicit operation:

use ripex::compiler::{check_with_compiler, CompilerCheckOptions};

let report = check_with_compiler(
    "tests/lang-test/rust/Cargo.toml",
    None,
    &CompilerCheckOptions::default(),
)?;
assert!(report.passed(), "{report:#?}");
# Ok::<(), std::io::Error>(())

The crate exposes:

• LanguageParser trait — parse(), extract(), symbols(), imports(), calls(), variables()
• parser_for(lang_id) / parser_for_ext(lang_id, ext) — language registry
• ExtractionResult — symbols, imports, calls, variables vectors
• Fact types: ParsedSymbol, ParsedImport, ParsedCall, ParsedVariable
• check_with_compiler() / plan_compiler_check() — optional bounded production-toolchain validation
• CompilerCheckReport — per-stage commands, status, exit code, output, and normalized diagnostics

All fact types are serializable when the serde feature is enabled. The cli feature enables serde automatically.

serde_json remains a required dependency for library builds because the public ripex::compiler module parses compile_commands.json; the serde feature only controls serialization derives for public fact and diagnostic types.

Feature flags

Each language is behind a feature flag, allowing minimal builds. Add Ripex as a library dependency and choose only the language features you need:

[dependencies]
ripex = { version = "0.3.0", default-features = false, features = ["lang-rust", "lang-js"] }

Feature	Languages
lang-js	JavaScript + TypeScript + JSX
lang-python	Python
lang-go	Go
lang-rust	Rust
lang-c	C
lang-cpp	C++
lang-csharp	C#
cli	CLI binary (serde, clap, anyhow)
serde	Serialization for public facts, spans, and diagnostics
lang-all (default)	All language features

CLI

To install the CLI binary, opt in explicitly:

cargo install ripex --version 0.3.0 --features cli

cargo build --features cli

# Parse a file; language auto-detected from extension.
cargo run --features cli -- parse path/to/file.rs

# Force a language and emit facts as JSON.
cargo run --features cli -- parse src/models/product.js --lang javascript --json

# List supported parsers.
cargo run --features cli -- ls

# Optionally validate a file with its production compiler.
cargo run --features cli -- check src/lib.rs

# Discover the nearest project manifest and check the whole project.
cargo run --features cli -- check src/lib.rs --project --json

# Select a standard and pass include/configuration flags through to the compiler.
cargo run --features cli -- check native/main.cpp --standard c++23 --arg=-Iinclude

Flags: --json (versioned machine output), --ast (AST shape summary), --facts (symbol/import/call/variable listing). The process exits non-zero when parse errors are present. JSON includes schema_version, the selected language, structured diagnostics, an AST summary, and all extracted facts.

check exits 0 when every compiler stage passes, 1 when source is rejected, and 2 when a toolchain is unavailable, times out, or cannot be invoked. Each stage has a 120-second default timeout. Project checks can execute compiler plugins, procedural macros, and build scripts; only check trusted projects.

Compiler requirements

ripex discovers these commands on PATH: gcc/clang, g++/clang++, rustc/cargo, go, dotnet/csc, node, tsc, python, and mypy. TypeScript can also be supplied as a local project dependency, discovered via npx --no-install tsc; the checked-in JavaScript fixture pins this dependency. Use --toolchain PATH to override the primary compiler. C/C++ include paths, defines, target triples, SDKs, and other build-specific inputs can be supplied with repeated --arg options. Language-standard conformance is therefore the conformance of the selected production compiler, standard, target, and project configuration rather than a claim that ripex reimplements those compilers.

For a C/C++ directory containing compile_commands.json, check validates each configured translation unit using its recorded compiler, include paths, defines, target flags, and generated-header settings. Output and dependency generation flags are removed and replaced with a no-output semantic check, so the project is not rebuilt or modified. Pass --project when checking one source file and the database is in a parent directory.

Architecture

Each language follows the same module structure:

src/<lang>/
  lexer/        Scanner, token definitions, keyword tables
  parser/       Recursive-descent parser with error recovery
  ast/          Typed AST nodes (statements, expressions, declarations)
  facts.rs      AST → ExtractionResult (the consumer-facing fact extractor)
  semantic/     Scope tracking, symbol tables, binding resolution
  syntax/       Language feature flags, operator precedence
  codegen/      Experimental canonical printers (not source preserving)
  transform/    AST transformation passes (JS pipeline + plugins)
  visit/        Visitor, walker, fold patterns for AST traversal
  diagnostics/  Error types and reporter
  config/       Parser options and configuration
  tests/        Per-language test suites

Safety limits

Built-in guards prevent pathological inputs from causing runaway allocation:

• MAX_INPUT_SIZE: 1 MB
• MAX_TOKENS: 200,000
• MAX_RECURSION: 512 depth

All guards are enforced in the lexer and parser state machines.

Testing

The test suite covers lexer tokenization, parser correctness, feature-isolated builds, every fact category, generator reuse, corpus parse-generate-parse round trips, safety limits, and zero-diagnostic fixture parsing.

cargo test --all-targets --all-features

Verification and Safety

Resource Limits and Boundaries

To prevent denial of service and resource exhaustion, ripex enforces strict limits on all parsed inputs:

• Maximum Input Size: 1 MB (MAX_INPUT_SIZE = 1,048,576 bytes). Enforced at CLI ingestion before allocations occur.
• Maximum Token Count: 200,000 tokens (MAX_TOKENS). Enforced during lexing.
• Maximum Recursion Depth: 512 frames (MAX_RECURSION). Enforced in all recursive-descent parsers to prevent stack overflow.

Breaching any limit produces a structured diagnostic of code limit_exceeded and sets status to LimitExceeded.

Trust Boundary and Compiler Safety

The ripex check command executes external compilers (such as gcc, clang, rustc/cargo, go, dotnet/csc, node, tsc, python, and mypy) on target source code or projects.

• Project Execution Risk: Running compilation check on projects (e.g., Cargo workspace, Go module, C# csproj) can execute arbitrary code via build scripts (build.rs), compiler plugins, or procedural macros.
• Explicit Trust Required: To prevent unauthorized code execution, project-level checks and raw compiler argument passthrough require the --trusted-project flag. If this flag is omitted, the planner immediately returns a PermissionDenied error.
• Sandbox Recommendation: We recommend running ripex check in an OS-level sandbox or containerized environment when validating untrusted third-party code.

License

MIT