FlatPPL — Rust implementation

Rust ecosystem for FlatPPL, the Flat Portable Probabilistic Language.

About FlatPPL

FlatPPL is a minimal, inference-agnostic stochastic language for specifying probabilistic models.

Components

This monorepo is a Cargo workspace; member crates live under crates/:

• flatppl-core — the in-memory IR (extended-FlatPIR model)
• flatppl-syntax — canonical FlatPPL surface syntax: parse + print
• flatppl-flatpir — FlatPIR S-expression reader + writer
• flatppl-infer — type, shape, and phase inference
• flatppl-hs3 — HS3 / pyhf → FlatPPL importer
• flatppl-fileaccess — resolve a source (local path or http/https URL) to a local file, with the shared remote-content cache (native host layer)
• flatppl-cli — the flatppl command-line driver
• flatppl-lsp — FlatPPL language server (diagnostics, hover, go-to-definition, completion)

See ARCHITECTURE.md for the design and the road ahead.

Installation (early users)

Requires a Rust toolchain (≥ 1.85, e.g. via rustup). The crates are not published to crates.io yet; install the CLI straight from the repository:

cargo install --git https://github.com/flatppl/flatppl-rust flatppl-cli

This places a flatppl binary on your Cargo bin path:

flatppl convert model.flatppl model.flatpir   # FlatPPL → FlatPIR
flatppl convert model.flatpir model.flatppl   # FlatPIR → FlatPPL
flatppl convert messy.flatppl tidy.flatppl    # canonicalize (same format)
flatppl infer model.flatppl typed.flatpir    # emit type/phase-annotated FlatPIR
flatppl infer --level=phase m.flatppl m.flatpir  # or: type, valueset, normalization, shape
flatppl prepare model.flatppl                  # fetch the model's remote deps into the cache

Formats are inferred from the file extensions. FlatPPL output uses the full surface syntax (operators, indexing, lambdas, :=); pass --syntax minimal for the lowered function-call form instead.

Inputs are local files. A model may load_module remote (http/https) dependencies, but the command input itself is always a local path — like cargo build on a local crate. convert and infer are local and offline: they resolve dependencies from a shared local cache only and never touch the network. To populate that cache, run flatppl prepare <model> — the one command that downloads a model's transitive remote dependencies (recursively, relative URLs resolved against their importing file). flatppl prepare --update refreshes already-cached deps. Cache location and trust are env-controlled ($FLATPPL_CACHEDIR; an untrusted URL prompts interactively and is refused in non-interactive use unless FLATPPL_TRUST=1).

Building and testing (developers)

git clone https://github.com/flatppl/flatppl-rust
cd flatppl-rust
cargo build --workspace        # build all crates
cargo test --workspace         # run all test suites
cargo clippy --workspace --all-targets -- -D warnings
cargo fmt --all
cargo run -p flatppl-cli -- convert model.flatppl model.flatpir

License

MIT