panproto 0.45.0

Automatic schema migration and version control for 51 schema languages and 248 programming languages

panproto

Python bindings for panproto, a schematic version-control system that treats every supported schema language (around 50, including ATProto, OpenAPI, AsyncAPI, Avro, Protobuf, JSON Schema, and Kubernetes CRDs) as views over a single graph format.

The bindings are built with PyO3 against the Rust core directly: no WASM runtime, no subprocess, no shelling out. The full panproto surface is available from Python, including the schema/migration/lens/VCS pipelines and the tree-sitter-driven parser for ~250 programming languages.

Status

panproto is pre-1.0. The 0.x series carries arbitrary breaking changes between minor versions; the panproto package version tracks the workspace version on every release. Python 3.13+ is required (the wheels are abi3 and forward-compatible across newer Python releases).

Installation

pip install panproto

Wheels are published on PyPI for Linux x86_64/aarch64, macOS arm64+x86_64, and Windows x86_64. No Rust toolchain is required to install.

Synopsis

import panproto

# Pick a built-in protocol, or define your own with `Protocol.from_theories`.
atproto = panproto.get_builtin_protocol("atproto")

# Build a schema using the fluent builder.
v1 = atproto.schema()
v1.vertex("post", "record", "app.bsky.feed.post")
v1.vertex("post:body", "object")
v1.vertex("post:body.text", "string")
v1.edge("post", "post:body", "record-schema")
v1.edge("post:body", "post:body.text", "prop", "text")
v1.constraint("post:body.text", "maxLength", "3000")
schema_v1 = v1.build()

# (build schema_v2 the same way, with the field renamed to `content` ...)

# Detect breaking changes.
report = panproto.diff_and_classify(schema_v1, schema_v2, atproto)
print(report.compatible)        # True or False
print(report.report_text())     # human-readable summary

# Auto-generate a bidirectional converter.
lens, quality, _ = panproto.auto_generate_lens(schema_v1, schema_v2, atproto)
view, complement = lens.get(instance)
restored = lens.put(view, complement)

# Version-control schemas.
repo = panproto.Repository.init("/path/to/repo")
repo.add(schema_v1)
repo.commit("initial schema")
repo.branch("feature")
repo.merge("feature")

API overview

Module / class	Purpose
Schema, SchemaBuilder	Fluent schema construction; Schema.validate(protocol) checks rules.
Protocol	Schema-language definition. Protocol.from_theories(...) builds one from a Theory.
get_builtin_protocol(name)	Load any of the ~50 builtin protocols by name.
define_protocol(spec)	Define a custom protocol from a dict.
Theory, create_theory	GAT-level theory construction (sorts, ops, equations, directed_eqs).
diff_schemas, diff_and_classify	Structural diff and breaking-change classification.
auto_generate_lens	Generate a bidirectional Lens from two schemas.
Lens	get(instance) -> (view, complement), put(view, complement) -> instance.
MigrationBuilder, compile_migration, compose_migrations	Hand-rolled migration construction.
Instance, IoRegistry	Parse/emit data across the 50+ supported formats.
Repository	Filesystem-backed VCS: init, commit, branch, merge, log, blame, bisect, stash, tag, plus data versioning.
AstParserRegistry, parse_source_file, ParseEmitLens	Full-AST parsing across ~250 languages via tree-sitter.
Expr, parse_expr, pretty_print_expr	Embedded expression language.

Companion grammar packs

The published panproto wheel ships only the 11 group-core tree-sitter grammars (Python, JavaScript, TypeScript, Java, C#, C++, PHP, Bash, C, Go, Rust). Beyond that, grammars are distributed as separately-installable companion wheels, one per language group:

Wheel	Languages
panproto-grammars-web	HTML, CSS, JavaScript, TypeScript, TSX, JSON, Vue, Svelte, Astro, GraphQL
panproto-grammars-systems	C, C++, Rust, Go, Zig, D, Nim, Odin, V, Hare
panproto-grammars-jvm	Java, Kotlin, Scala, Groovy, Clojure
panproto-grammars-scripting	Python, Ruby, Lua, Bash, Perl, R, Julia, Nushell, Fish
panproto-grammars-data	JSON, TOML, XML, YAML, SQL, CSV, GraphQL, Protobuf
panproto-grammars-functional	Haskell, OCaml, Elm, Gleam, Erlang, Elixir, PureScript, F#, Clojure, Scheme, Racket
panproto-grammars-devops	Dockerfile, Terraform, HCL, Nix, Bash, YAML, TOML, Make, CMake
panproto-grammars-mobile	Swift, Kotlin, Dart, Java, Objective-C
panproto-grammars-music	SuperCollider, LilyPond, ABC, Csound, ChucK, Glicol, Tidal mini-notation, Strudel mini-notation
panproto-grammars-all	every grammar in panproto-grammars (around 250 languages)

Install whichever group you need:

pip install panproto-grammars-functional

There is nothing to import from these packages. They register a panproto.grammars entry point on installation; panproto.AstParserRegistry() walks every such entry point and threads the discovered grammar metadata into the native registry on construction. The native class is reachable as panproto._native.AstParserRegistry() for callers who want only the group-core baseline.

How it works under the hood:

• Each companion is its own pyo3 cdylib depending on panproto-grammars with one group-* feature flag.
• On import, the companion's grammars_metadata() returns a list of dicts containing the tree-sitter Language pointer plus byte-slice pointer/length pairs (cast to integers for transport across cdylib boundaries).
• The trust boundary lives on the panproto side, in register_external_from_metadata (crates/panproto-py/src/parse.rs), which decodes the integers back into &'static references and registers a full LanguageParser. The companion's grammar bytes live in its .rodata and stay valid for the process lifetime.
• A single broken grammar (e.g. an upstream node-types.json with a malformed entry) is skipped with a RuntimeWarning rather than aborting registration; mirrors the resilience of the built-in path.

Performance notes

• The _native extension talks to the Rust core through PyO3's zero-copy pyclass slabs. Schemas, theories, and lenses are reference-counted Rust objects on the Python side; mutations go through dedicated builder types (SchemaBuilder, MigrationBuilder) that consume on build(), so you can't accidentally observe partial state.
• Cross-thread sharing of these objects requires the GIL; for parallel work, fan out at the data layer (e.g. parallelise lens.get calls with concurrent.futures) and keep the schema/lens objects per worker.
• Wheel-load cost is one-time; the import sets up the protocol registry lazily so cold-start is fast.

Contributing

Source: bindings/python. Issues and pull requests at github.com/panproto/panproto/issues.

The native extension lives at crates/panproto-py on the Rust side; bindings/python/src/panproto/__init__.py is the pure-Python re-export layer that maturin ships alongside the compiled extension.

License

MIT © 2026 Aaron Steven White.