leanfran 0.6.0__tar.gz → 0.7.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- leanfran-0.7.0/.gitignore +47 -0
- leanfran-0.7.0/PKG-INFO +62 -0
- leanfran-0.7.0/README.md +49 -0
- leanfran-0.7.0/hatch_build.py +14 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/pyproject.toml +5 -3
- leanfran-0.6.0/.gitignore +0 -8
- leanfran-0.6.0/PKG-INFO +0 -388
- leanfran-0.6.0/README.md +0 -374
- {leanfran-0.6.0 → leanfran-0.7.0}/LICENSE +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/__init__.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/check.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/cli.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/embed.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/io.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/io_externs.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/lean/LeanPythonLib/Py.lean +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/marshal.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/models.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/runtime.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/sync.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/translate.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/types.py +0 -0
- {leanfran-0.6.0 → leanfran-0.7.0}/src/leanfran/verify.py +0 -0
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
/result
|
|
2
|
+
/.claude-dev/
|
|
3
|
+
/langs/Rust/result
|
|
4
|
+
/langs/Rust/.lake/
|
|
5
|
+
/langs/Rust/target/
|
|
6
|
+
/langs/Rust/test-out/
|
|
7
|
+
/langs/Rust/**/.leanfran/workspace/
|
|
8
|
+
/langs/Rust/.lake-aeneas/
|
|
9
|
+
/langs/Python/result
|
|
10
|
+
/langs/Python/.lake/
|
|
11
|
+
/langs/Python/test-out/
|
|
12
|
+
/langs/Python/**/.leanfran/
|
|
13
|
+
/langs/Python/**/__pycache__/
|
|
14
|
+
/langs/Python/.venv/
|
|
15
|
+
/langs/Python/dist/
|
|
16
|
+
/langs/TypeScript/.lake
|
|
17
|
+
/langs/TypeScript/**/*.olean
|
|
18
|
+
/langs/TypeScript/result
|
|
19
|
+
/langs/TypeScript/node_modules
|
|
20
|
+
/langs/TypeScript/dist
|
|
21
|
+
/langs/TypeScript/test-out
|
|
22
|
+
/langs/TypeScript/**/.leanfran/workspace/
|
|
23
|
+
/langs/TypeScript/test/embed/app/node_modules
|
|
24
|
+
/langs/TypeScript/test/verify/app/node_modules
|
|
25
|
+
/langs/TypeScript/samples/**/node_modules
|
|
26
|
+
/langs/TypeScript/samples/**/.lake
|
|
27
|
+
/langs/Clojure/result
|
|
28
|
+
/langs/Clojure/.lake/
|
|
29
|
+
/langs/Clojure/**/.leanfran/
|
|
30
|
+
/langs/Clojure/**/.cpcache/
|
|
31
|
+
/langs/Clojure/.nrepl-port
|
|
32
|
+
/langs/Clojure/.lsp/
|
|
33
|
+
/langs/Clojure/.clj-kondo/*
|
|
34
|
+
!/langs/Clojure/.clj-kondo/config.edn
|
|
35
|
+
/langs/Clojure/test-out/
|
|
36
|
+
/langs/Clojure/target/
|
|
37
|
+
!/langs/Clojure/resources/clj-kondo.exports/io.github.conao3/
|
|
38
|
+
/langs/PHP/.lake/
|
|
39
|
+
/langs/PHP/vendor/
|
|
40
|
+
/langs/PHP/test-out/
|
|
41
|
+
/langs/PHP/samples/*/composer.lock
|
|
42
|
+
/langs/PHP/samples/*/lean/.lake/
|
|
43
|
+
/langs/PHP/samples/*/vendor/
|
|
44
|
+
/Rust/**/target/
|
|
45
|
+
/Rust/**/.lake/
|
|
46
|
+
/Rust/**/.lake-aeneas/
|
|
47
|
+
/Rust/**/test-out/
|
leanfran-0.7.0/PKG-INFO
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
1
|
+
Metadata-Version: 2.5
|
|
2
|
+
Name: leanfran
|
|
3
|
+
Version: 0.7.0
|
|
4
|
+
Summary: Run Lean 4 code as Python: a runtime for code generated from Lean's IR, Lean blocks in Python files, and pydantic models for Lean types
|
|
5
|
+
Project-URL: Repository, https://github.com/conao3/lean-leanfran/tree/master/langs/Python
|
|
6
|
+
License-Expression: Apache-2.0
|
|
7
|
+
Classifier: Framework :: Pydantic :: 2
|
|
8
|
+
Classifier: Programming Language :: Python :: 3
|
|
9
|
+
Classifier: Topic :: Software Development :: Compilers
|
|
10
|
+
Requires-Python: >=3.12
|
|
11
|
+
Requires-Dist: pydantic>=2.11
|
|
12
|
+
Description-Content-Type: text/markdown
|
|
13
|
+
|
|
14
|
+
# Python
|
|
15
|
+
|
|
16
|
+
[leanfran](https://github.com/conao3/lean-leanfran/blob/master/README.md) for Python: write Lean blocks with `lean(...)`, expose Lean types as pydantic models, and verify supported Python functions against Lean specifications.
|
|
17
|
+
|
|
18
|
+
## Install
|
|
19
|
+
|
|
20
|
+
```sh
|
|
21
|
+
uv add leanfran
|
|
22
|
+
nix profile install 'github:conao3/lean-leanfran#python'
|
|
23
|
+
```
|
|
24
|
+
|
|
25
|
+
Generation also requires Lean 4.35.0 (`lean` and `lake`) on `PATH`. Generated modules run with the Python package and pydantic.
|
|
26
|
+
|
|
27
|
+
## First Lean block
|
|
28
|
+
|
|
29
|
+
Create `src/myapp/__init__.py` and put this in `src/myapp/arith.py`:
|
|
30
|
+
|
|
31
|
+
```python
|
|
32
|
+
from leanfran import lean
|
|
33
|
+
|
|
34
|
+
arith = lean("arith", r"""
|
|
35
|
+
def double (n : Nat) : Nat := n + n
|
|
36
|
+
|
|
37
|
+
theorem double_eq (n : Nat) : double n = 2 * n := by
|
|
38
|
+
unfold double; omega
|
|
39
|
+
""")
|
|
40
|
+
```
|
|
41
|
+
|
|
42
|
+
Generate the Python modules:
|
|
43
|
+
|
|
44
|
+
```sh
|
|
45
|
+
leanfran sync --package src/myapp
|
|
46
|
+
```
|
|
47
|
+
|
|
48
|
+
Then call the function from your application:
|
|
49
|
+
|
|
50
|
+
```python
|
|
51
|
+
from myapp.arith import arith
|
|
52
|
+
|
|
53
|
+
assert arith.double(21) == 42
|
|
54
|
+
```
|
|
55
|
+
|
|
56
|
+
Commit `src/myapp/_lean/` with your package, and add `.leanfran/` to `.gitignore`. Run sync again when you change a Lean block.
|
|
57
|
+
|
|
58
|
+
## Next steps
|
|
59
|
+
|
|
60
|
+
- [Full Python guide](https://github.com/conao3/lean-leanfran/blob/master/docs/python-guide.md): pydantic models, JSON, `@model`, verification, CLI and limitations.
|
|
61
|
+
- [Samples](https://github.com/conao3/lean-leanfran/blob/master/langs/Python/samples/README.md), including pricing, word count and step-by-step examples.
|
|
62
|
+
- [Development](https://github.com/conao3/lean-leanfran/blob/master/CONTRIBUTING.md) and [implementation notes](https://github.com/conao3/lean-leanfran/blob/master/docs/python-development.md).
|
leanfran-0.7.0/README.md
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
# Python
|
|
2
|
+
|
|
3
|
+
[leanfran](https://github.com/conao3/lean-leanfran/blob/master/README.md) for Python: write Lean blocks with `lean(...)`, expose Lean types as pydantic models, and verify supported Python functions against Lean specifications.
|
|
4
|
+
|
|
5
|
+
## Install
|
|
6
|
+
|
|
7
|
+
```sh
|
|
8
|
+
uv add leanfran
|
|
9
|
+
nix profile install 'github:conao3/lean-leanfran#python'
|
|
10
|
+
```
|
|
11
|
+
|
|
12
|
+
Generation also requires Lean 4.35.0 (`lean` and `lake`) on `PATH`. Generated modules run with the Python package and pydantic.
|
|
13
|
+
|
|
14
|
+
## First Lean block
|
|
15
|
+
|
|
16
|
+
Create `src/myapp/__init__.py` and put this in `src/myapp/arith.py`:
|
|
17
|
+
|
|
18
|
+
```python
|
|
19
|
+
from leanfran import lean
|
|
20
|
+
|
|
21
|
+
arith = lean("arith", r"""
|
|
22
|
+
def double (n : Nat) : Nat := n + n
|
|
23
|
+
|
|
24
|
+
theorem double_eq (n : Nat) : double n = 2 * n := by
|
|
25
|
+
unfold double; omega
|
|
26
|
+
""")
|
|
27
|
+
```
|
|
28
|
+
|
|
29
|
+
Generate the Python modules:
|
|
30
|
+
|
|
31
|
+
```sh
|
|
32
|
+
leanfran sync --package src/myapp
|
|
33
|
+
```
|
|
34
|
+
|
|
35
|
+
Then call the function from your application:
|
|
36
|
+
|
|
37
|
+
```python
|
|
38
|
+
from myapp.arith import arith
|
|
39
|
+
|
|
40
|
+
assert arith.double(21) == 42
|
|
41
|
+
```
|
|
42
|
+
|
|
43
|
+
Commit `src/myapp/_lean/` with your package, and add `.leanfran/` to `.gitignore`. Run sync again when you change a Lean block.
|
|
44
|
+
|
|
45
|
+
## Next steps
|
|
46
|
+
|
|
47
|
+
- [Full Python guide](https://github.com/conao3/lean-leanfran/blob/master/docs/python-guide.md): pydantic models, JSON, `@model`, verification, CLI and limitations.
|
|
48
|
+
- [Samples](https://github.com/conao3/lean-leanfran/blob/master/langs/Python/samples/README.md), including pricing, word count and step-by-step examples.
|
|
49
|
+
- [Development](https://github.com/conao3/lean-leanfran/blob/master/CONTRIBUTING.md) and [implementation notes](https://github.com/conao3/lean-leanfran/blob/master/docs/python-development.md).
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
from pathlib import Path
|
|
2
|
+
|
|
3
|
+
from hatchling.builders.hooks.plugin.interface import BuildHookInterface
|
|
4
|
+
|
|
5
|
+
|
|
6
|
+
class CustomBuildHook(BuildHookInterface):
|
|
7
|
+
def initialize(self, version, build_data):
|
|
8
|
+
root = Path(self.root)
|
|
9
|
+
license_file = root / "LICENSE"
|
|
10
|
+
if not license_file.is_file():
|
|
11
|
+
license_file = root.parents[1] / "LICENSE"
|
|
12
|
+
if not license_file.is_file():
|
|
13
|
+
raise FileNotFoundError("The distribution requires the repository LICENSE")
|
|
14
|
+
build_data["force_include"][str(license_file)] = "LICENSE"
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
[project]
|
|
2
2
|
name = "leanfran"
|
|
3
|
-
version = "0.
|
|
3
|
+
version = "0.7.0"
|
|
4
4
|
description = "Run Lean 4 code as Python: a runtime for code generated from Lean's IR, Lean blocks in Python files, and pydantic models for Lean types"
|
|
5
5
|
readme = "README.md"
|
|
6
6
|
license = "Apache-2.0"
|
|
@@ -13,7 +13,7 @@ classifiers = [
|
|
|
13
13
|
]
|
|
14
14
|
|
|
15
15
|
[project.urls]
|
|
16
|
-
Repository = "https://github.com/conao3/lean-leanfran/tree/master/Python"
|
|
16
|
+
Repository = "https://github.com/conao3/lean-leanfran/tree/master/langs/Python"
|
|
17
17
|
|
|
18
18
|
[project.scripts]
|
|
19
19
|
leanfran = "leanfran.cli:main"
|
|
@@ -25,11 +25,13 @@ dev = ["pytest>=8", "pytest-icdiff>=0.9", "hypothesis>=6", "fastapi>=0.115", "ht
|
|
|
25
25
|
requires = ["hatchling"]
|
|
26
26
|
build-backend = "hatchling.build"
|
|
27
27
|
|
|
28
|
+
[tool.hatch.build.hooks.custom]
|
|
29
|
+
|
|
28
30
|
[tool.hatch.build.targets.wheel]
|
|
29
31
|
packages = ["src/leanfran"]
|
|
30
32
|
|
|
31
33
|
[tool.hatch.build.targets.sdist]
|
|
32
|
-
include = ["/src/leanfran", "/README.md", "/
|
|
34
|
+
include = ["/src/leanfran", "/README.md", "/pyproject.toml", "/hatch_build.py"]
|
|
33
35
|
|
|
34
36
|
[tool.ruff]
|
|
35
37
|
line-length = 120
|
leanfran-0.6.0/.gitignore
DELETED
leanfran-0.6.0/PKG-INFO
DELETED
|
@@ -1,388 +0,0 @@
|
|
|
1
|
-
Metadata-Version: 2.5
|
|
2
|
-
Name: leanfran
|
|
3
|
-
Version: 0.6.0
|
|
4
|
-
Summary: Run Lean 4 code as Python: a runtime for code generated from Lean's IR, Lean blocks in Python files, and pydantic models for Lean types
|
|
5
|
-
Project-URL: Repository, https://github.com/conao3/lean-leanfran/tree/master/Python
|
|
6
|
-
License-Expression: Apache-2.0
|
|
7
|
-
License-File: LICENSE
|
|
8
|
-
Classifier: Framework :: Pydantic :: 2
|
|
9
|
-
Classifier: Programming Language :: Python :: 3
|
|
10
|
-
Classifier: Topic :: Software Development :: Compilers
|
|
11
|
-
Requires-Python: >=3.12
|
|
12
|
-
Requires-Dist: pydantic>=2.11
|
|
13
|
-
Description-Content-Type: text/markdown
|
|
14
|
-
|
|
15
|
-
<p align="center">
|
|
16
|
-
<img src="https://raw.githubusercontent.com/conao3/lean-leanfran/master/Python/docs/assets/banner.png" alt="leanfran: write and prove functions in Lean 4, run them as Python" width="880">
|
|
17
|
-
</p>
|
|
18
|
-
|
|
19
|
-
<p align="center"><strong>Write and prove functions in Lean 4, run them as Python</strong></p>
|
|
20
|
-
|
|
21
|
-
<p align="center">
|
|
22
|
-
<a href="https://pypi.org/project/leanfran/"><img src="https://img.shields.io/pypi/v/leanfran" alt="PyPI version"></a>
|
|
23
|
-
<a href="https://github.com/conao3/lean-leanfran/actions/workflows/ci.yml"><img src="https://github.com/conao3/lean-leanfran/actions/workflows/ci.yml/badge.svg?branch=master" alt="CI"></a>
|
|
24
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/LICENSE"><img src="https://img.shields.io/github/license/conao3/lean-leanfran" alt="License: Apache-2.0"></a>
|
|
25
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/pyproject.toml"><img src="https://img.shields.io/badge/Python-%3E%3D3.12-blue" alt="Python >=3.12"></a>
|
|
26
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/lean-toolchain"><img src="https://img.shields.io/badge/Lean-v4.35.0-blue" alt="Lean v4.35.0"></a>
|
|
27
|
-
<a href="https://github.com/conao3/lean-leanfran/stargazers"><img src="https://img.shields.io/github/stars/conao3/lean-leanfran?style=flat" alt="GitHub stars"></a>
|
|
28
|
-
</p>
|
|
29
|
-
|
|
30
|
-
<p align="center">
|
|
31
|
-
<a href="#install">Install</a> ·
|
|
32
|
-
<a href="#lean-blocks">Lean blocks</a> ·
|
|
33
|
-
<a href="#values">Values</a> ·
|
|
34
|
-
<a href="#verifying-python-functions">Verifying Python functions</a> ·
|
|
35
|
-
<a href="#examples">Examples</a> ·
|
|
36
|
-
<a href="#performance">Performance</a> ·
|
|
37
|
-
<a href="https://pypi.org/project/leanfran/">PyPI</a>
|
|
38
|
-
</p>
|
|
39
|
-
|
|
40
|
-
# leanfran
|
|
41
|
-
|
|
42
|
-
leanfran lets you write functions in Lean 4, prove theorems about them, and call them from Python
|
|
43
|
-
with pydantic models as arguments and results. The PyPI distribution is `leanfran`; you import
|
|
44
|
-
it as `leanfran`.
|
|
45
|
-
|
|
46
|
-
```python
|
|
47
|
-
from pydantic import BaseModel, Field
|
|
48
|
-
from leanfran import lean
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
class LineItem(BaseModel):
|
|
52
|
-
sku: str
|
|
53
|
-
qty: int = Field(ge=1)
|
|
54
|
-
unitPrice: int = Field(ge=0)
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
pricing = lean("pricing", r"""
|
|
58
|
-
inductive Discount where
|
|
59
|
-
| none
|
|
60
|
-
| percent (p : Nat) (h : p ≤ 100)
|
|
61
|
-
| fixed (amount : Nat)
|
|
62
|
-
|
|
63
|
-
def subtotal (items : List LineItem) : Nat :=
|
|
64
|
-
items.foldl (fun acc i => acc + i.qty * i.unitPrice) 0
|
|
65
|
-
|
|
66
|
-
def total (items : List LineItem) (d : Discount) : Nat :=
|
|
67
|
-
match d with
|
|
68
|
-
| .none => subtotal items
|
|
69
|
-
| .percent p _ => subtotal items - subtotal items * p / 100
|
|
70
|
-
| .fixed a => subtotal items - a
|
|
71
|
-
|
|
72
|
-
theorem total_le (items : List LineItem) (d : Discount) : total items d ≤ subtotal items := by
|
|
73
|
-
unfold total; split <;> omega
|
|
74
|
-
""", models=[LineItem])
|
|
75
|
-
|
|
76
|
-
items = [LineItem(sku="a", qty=2, unitPrice=100)]
|
|
77
|
-
pricing.total(items, pricing.Discount.percent(p=10)) # 180
|
|
78
|
-
pricing.Discount.percent(p=150) # pydantic.ValidationError: p: Input should be less than or equal to 100
|
|
79
|
-
```
|
|
80
|
-
|
|
81
|
-
- `leanfran sync` checks the proofs with Lake and compiles the Lean code from Lean's IR to plain
|
|
82
|
-
Python. It writes the generated modules to `<package>/_lean/`, and they ship with your package.
|
|
83
|
-
To run them you need the `leanfran` wheel and pydantic. You do not need Lean.
|
|
84
|
-
- The generated Python computes what the C backend computes. A diff test runs the same Lean
|
|
85
|
-
programs natively and as Python and compares the output byte for byte. The generated code keeps
|
|
86
|
-
reference counting, so `Array` updates happen in place as in C.
|
|
87
|
-
- Lean structures and inductive types become pydantic models. A proof field (`h : p ≤ 100`) becomes
|
|
88
|
-
a validator that runs a decision procedure compiled from the proposition, so every value that
|
|
89
|
-
reaches Lean satisfies the invariants the proofs rely on. A comparison with a constant also becomes
|
|
90
|
-
an annotated-types constraint (`Le(100)`) that pydantic-core checks and the JSON Schema shows. The
|
|
91
|
-
JSON of the models is the JSON of Lean's `deriving ToJson, FromJson`.
|
|
92
|
-
- Lean code can use your own pydantic models (`models=[...]`). Their field constraints
|
|
93
|
-
(`ge`, `gt`, `le`, `lt`, `min_length`, `max_length`) become proof fields that Lean proofs may assume.
|
|
94
|
-
- The models work wherever pydantic does: `TypeAdapter`, `model_validate_json`, JSON Schema, FastAPI
|
|
95
|
-
request and response types.
|
|
96
|
-
- `@model` binds a plain Python function to Lean specifications. `leanfran check` translates the
|
|
97
|
-
function to Lean and judges your proofs ([Verifying Python functions](#verifying-python-functions)).
|
|
98
|
-
|
|
99
|
-
## Install
|
|
100
|
-
|
|
101
|
-
The Python package and the Lean CLI:
|
|
102
|
-
|
|
103
|
-
```sh
|
|
104
|
-
uv add leanfran # the runtime, lean(), leanfran sync
|
|
105
|
-
nix profile install 'github:conao3/lean-leanfran?dir=Python' # the Lean CLI `leanfran-python`
|
|
106
|
-
```
|
|
107
|
-
|
|
108
|
-
`leanfran sync` also needs `lake` and the Lean toolchain of `lean-toolchain` (v4.35.0) on `PATH`.
|
|
109
|
-
|
|
110
|
-
## Lean blocks
|
|
111
|
-
|
|
112
|
-
`lean(name, source, *, imports=(), uses=(), models=())` defines a Lean block of the calling module,
|
|
113
|
-
which must belong to a package. The block's definitions live in the namespace
|
|
114
|
-
`LeanPythonEmbed.<module>.<name>`. The returned module exposes them under their names relative to it
|
|
115
|
-
(`Discount.percent`, `total`).
|
|
116
|
-
|
|
117
|
-
```sh
|
|
118
|
-
leanfran sync --package src/myapp
|
|
119
|
-
```
|
|
120
|
-
|
|
121
|
-
`leanfran sync` imports every module of the package, writes each block to a Lake workspace in
|
|
122
|
-
`.leanfran/workspace/`, builds it, and generates `src/myapp/_lean/<module>__<name>.py` and
|
|
123
|
-
`src/myapp/_lean/blocks.json`. Lean errors point at the Python file:
|
|
124
|
-
|
|
125
|
-
```
|
|
126
|
-
leanfran: src/myapp/m.py:6: error: (lean block m, Lean 4:28) Tactic `rfl` failed: ...
|
|
127
|
-
```
|
|
128
|
-
|
|
129
|
-
Add `.leanfran/` to `.gitignore` and commit `<package>/_lean/`.
|
|
130
|
-
|
|
131
|
-
At run time `lean(...)` compares a hash of the block (its source, imports, uses, the Lean structures
|
|
132
|
-
of its models and the generator version) with `blocks.json`. If the block changed since the last
|
|
133
|
-
sync, it raises `LeanBlockStale`. With `LEANFRAN_PYTHON_AUTO_SYNC=1` it runs `leanfran sync` instead.
|
|
134
|
-
|
|
135
|
-
- `imports=["Std.Data.HashMap"]`: Lean modules to import.
|
|
136
|
-
- `uses=["geometry"]`: other blocks of the module (or `"pkg.mod:name"`) to import and open. The
|
|
137
|
-
block shares their types and does not generate them again.
|
|
138
|
-
- `models=[LineItem]`: pydantic models available as Lean structures (below).
|
|
139
|
-
|
|
140
|
-
## Values
|
|
141
|
-
|
|
142
|
-
| Lean | Python |
|
|
143
|
-
|---|---|
|
|
144
|
-
| `Nat` | `int` (`leanfran.types.Nat` = `Annotated[int, Ge(0)]`) |
|
|
145
|
-
| `Int` | `int` |
|
|
146
|
-
| `UInt8` … `UInt64`, `USize`, `Int8` … `Int64` | `int` with the range as `Ge`/`Le` |
|
|
147
|
-
| `Fin n` | `Annotated[int, Ge(0), Lt(n)]` |
|
|
148
|
-
| `Bool` | `bool` |
|
|
149
|
-
| `Char` | `str` of one character |
|
|
150
|
-
| `String` | `str` |
|
|
151
|
-
| `Float`, `Float32` | `float` |
|
|
152
|
-
| `Unit` | `None` |
|
|
153
|
-
| `Option α` | `α \| None` |
|
|
154
|
-
| `List α`, `Array α` | `list[α]` |
|
|
155
|
-
| `α × β` | `tuple[α, β]` |
|
|
156
|
-
| `{x : α // p x}` | `Annotated[α, AfterValidator(p)]` |
|
|
157
|
-
| structure | a frozen, strict pydantic model |
|
|
158
|
-
| inductive type, constructors without arguments | `enum.StrEnum` |
|
|
159
|
-
| other inductive type | a model per constructor, subclasses of the type's class |
|
|
160
|
-
| `IO α` | `α`; an `IO.Error` is raised as `LeanIOError` (an `OSError`) |
|
|
161
|
-
|
|
162
|
-
The generated functions check arguments while they convert them: `pub.double(-1)` raises
|
|
163
|
-
`ValidationError` with the location of the bad value. Constants are functions without arguments
|
|
164
|
-
(`answer()`).
|
|
165
|
-
|
|
166
|
-
Constructors and structures take keyword arguments (`Shape.rect(width=1, height=2)`); unnamed
|
|
167
|
-
constructor arguments are `a0`, `a1`, …. Inductive values work with `match`:
|
|
168
|
-
|
|
169
|
-
```python
|
|
170
|
-
match shape:
|
|
171
|
-
case geometry.Shape.rect(width=w, height=h):
|
|
172
|
-
...
|
|
173
|
-
```
|
|
174
|
-
|
|
175
|
-
### JSON
|
|
176
|
-
|
|
177
|
-
The JSON of a generated model is the JSON of Lean's derived `ToJson`: a structure is an object, a
|
|
178
|
-
constructor without arguments is its name (`"dot"`), a constructor with named arguments is
|
|
179
|
-
`{"rect": {"width": 1, "height": 2}}`, with unnamed arguments `{"num": 7}` or `{"pair": [1, "x"]}`.
|
|
180
|
-
`tests/test_json.py` compares the generated models with a native Lean program that prints `toJson`.
|
|
181
|
-
|
|
182
|
-
### Your pydantic models in Lean
|
|
183
|
-
|
|
184
|
-
`models=[LineItem]` writes a Lean structure for each model (and the models its fields use) to the
|
|
185
|
-
namespace `LeanPythonModels.<python module>`, and the block opens that namespace:
|
|
186
|
-
|
|
187
|
-
```lean
|
|
188
|
-
structure LineItem where
|
|
189
|
-
sku : String
|
|
190
|
-
qty : Nat
|
|
191
|
-
qty_ge : 1 ≤ qty
|
|
192
|
-
unitPrice : Nat
|
|
193
|
-
```
|
|
194
|
-
|
|
195
|
-
`int` with a lower bound of at least 0 becomes `Nat`; the other bounds and `min_length`/`max_length`
|
|
196
|
-
become proof fields. Field types:
|
|
197
|
-
|
|
198
|
-
- `int`, `bool`, `str`, `float`, `list[T]`, `tuple[...]`, `T | None`, other models
|
|
199
|
-
- enums: an inductive type `LeanPythonModels.<module>.<Enum>` with a constructor per member
|
|
200
|
-
- `Literal["a", "b"]`: `LeanPythonLiterals.<module>.<Model>_<field>`
|
|
201
|
-
- unions of models: `LeanPythonUnions.<module>.<Model>_<field>`, a constructor per model holding it
|
|
202
|
-
as `val`
|
|
203
|
-
|
|
204
|
-
Lean proofs cannot assume `multiple_of`, `pattern` or validators; the structure's docstring lists
|
|
205
|
-
them. Generated functions take and return your classes. Values that skipped validation
|
|
206
|
-
(`model_construct`) go through validation again on the way to Lean.
|
|
207
|
-
|
|
208
|
-
## Verifying Python functions
|
|
209
|
-
|
|
210
|
-
`@model` binds a typed, pure Python function to specifications written in a Lean block. The function
|
|
211
|
-
stays ordinary Python. `leanfran check` translates it to Lean and checks your proofs, and
|
|
212
|
-
`check_model` runs the same specifications as a Hypothesis property test.
|
|
213
|
-
|
|
214
|
-
```python
|
|
215
|
-
from leanfran import lean
|
|
216
|
-
from leanfran.verify import model
|
|
217
|
-
|
|
218
|
-
spec = lean("spec", r"""
|
|
219
|
-
def clampSpec (x lo hi : Int) : Int := max lo (min hi x)
|
|
220
|
-
def ordered (_x lo hi : Int) : Bool := lo ≤ hi
|
|
221
|
-
def inRange (_x lo hi : Int) (r : Int) : Bool := lo ≤ r && r ≤ hi
|
|
222
|
-
""")
|
|
223
|
-
|
|
224
|
-
@model(refines=spec.clampSpec, requires=spec.ordered, ensures={"in_range": spec.inRange})
|
|
225
|
-
def clamp(x: int, lo: int, hi: int) -> int:
|
|
226
|
-
if x < lo:
|
|
227
|
-
return lo
|
|
228
|
-
if x > hi:
|
|
229
|
-
return hi
|
|
230
|
-
return x
|
|
231
|
-
```
|
|
232
|
-
|
|
233
|
-
`leanfran check --package calc` writes the translation and one obligation per specification:
|
|
234
|
-
|
|
235
|
-
```lean
|
|
236
|
-
def LeanPythonModel.calc.calc.clamp (x : Int) (lo : Int) (hi : Int) : Int := Id.run do
|
|
237
|
-
if ((decide (x < lo))) then
|
|
238
|
-
return lo
|
|
239
|
-
if ((decide (x > hi))) then
|
|
240
|
-
return hi
|
|
241
|
-
return x
|
|
242
|
-
|
|
243
|
-
abbrev LeanPythonModel.calc.calc.clamp.obligation.refines : Prop :=
|
|
244
|
-
∀ (x : Int) (lo : Int) (hi : Int), ordered x lo hi = true → clamp x lo hi = clampSpec x lo hi
|
|
245
|
-
```
|
|
246
|
-
|
|
247
|
-
Then it judges the theorems `<function>.proof.<label>` of `lean/Proofs/*.lean` next to
|
|
248
|
-
`pyproject.toml`:
|
|
249
|
-
|
|
250
|
-
```lean
|
|
251
|
-
import LeanPythonModel.calc.calc
|
|
252
|
-
open LeanPythonEmbed.calc.spec.spec
|
|
253
|
-
namespace LeanPythonModel.«calc».«calc»
|
|
254
|
-
|
|
255
|
-
theorem clamp.proof.refines : clamp.obligation.refines := by
|
|
256
|
-
intro x lo hi h
|
|
257
|
-
simp only [ordered, decide_eq_true_eq] at h
|
|
258
|
-
simp only [clamp, clampSpec]
|
|
259
|
-
split <;> (try split) <;> simp_all <;> omega
|
|
260
|
-
```
|
|
261
|
-
|
|
262
|
-
```
|
|
263
|
-
proved calc.calc:clamp refines
|
|
264
|
-
proved calc.calc:clamp in_range
|
|
265
|
-
sorry calc.calc:sum_to refines
|
|
266
|
-
missing calc.calc:sum_abs nonneg
|
|
267
|
-
```
|
|
268
|
-
|
|
269
|
-
Under `simp`, a `for` loop over a list becomes `List.foldl` (`forIn_pure_yield_eq_foldl`), so you
|
|
270
|
-
prove a loop by induction on the list (`tests/verify/lean/Proofs/Calc.lean` proves `sum_abs`).
|
|
271
|
-
|
|
272
|
-
The states:
|
|
273
|
-
|
|
274
|
-
| state | meaning |
|
|
275
|
-
|---|---|
|
|
276
|
-
| `proved` | the theorem holds |
|
|
277
|
-
| `sorry` | the proof uses `sorry` |
|
|
278
|
-
| `axioms` | the proof uses non-standard axioms |
|
|
279
|
-
| `failed` | the proof file does not build |
|
|
280
|
-
| `missing` | no theorem for the obligation |
|
|
281
|
-
| `mismatch` | the theorem states something else |
|
|
282
|
-
| `unextracted` | the obligation's statement is not defined |
|
|
283
|
-
|
|
284
|
-
With `--require proved`, `leanfran check` exits with 254 unless every obligation is proved. It
|
|
285
|
-
writes the result to `leanfran.lock` with a hash of each function's source. `leanfran status`
|
|
286
|
-
reads the lock file without Lean and marks changed functions `stale`.
|
|
287
|
-
|
|
288
|
-
- `refines=s`: `f x = s x` (`Except.ok (s x)` for a function that may fail); `requires=p`: `p x = true →`
|
|
289
|
-
before each obligation; `ensures={"label": q}`: `q x (f x) = true`. A function annotated `-> Nat`
|
|
290
|
-
also gets `returns_nat` (`0 ≤ f x`).
|
|
291
|
-
- The translation: `int` is `Int` (a `Nat` parameter is a `Nat` converted to `Int` on entry);
|
|
292
|
-
statements become Lean `do` notation (`let mut`, `for`, `while`, `break`, `continue`, early
|
|
293
|
-
`return`); indexing, `//`, `%`, `assert` and `raise` make the function return
|
|
294
|
-
`Py.M α = Except Py.Exc α`, with Python's negative indices and floor division (`LeanPythonLib.Py`).
|
|
295
|
-
Supported: `int`, `bool`, `str`, `float`, `list`, `tuple`, `T | None`, pydantic models (as the
|
|
296
|
-
Lean structures of `models=`), `len`, `abs`, `min`, `max`, `sum`, list comprehensions, `append` on
|
|
297
|
-
local lists, and calls of other `@model` functions of the module.
|
|
298
|
-
- `check_model(f)` / `check_module(m)` generate arguments from the annotations (pydantic constraints
|
|
299
|
-
included), skip those failing `requires`, and compare with `refines` and `ensures`.
|
|
300
|
-
|
|
301
|
-
## Examples
|
|
302
|
-
|
|
303
|
-
[`samples/`](https://github.com/conao3/lean-leanfran/tree/master/Python/samples) holds Python packages with Lean
|
|
304
|
-
blocks. `make samples` syncs each one, runs it, and compares the output with its `expected.txt`.
|
|
305
|
-
|
|
306
|
-
| sample | shows |
|
|
307
|
-
|---|---|
|
|
308
|
-
| [word-count](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/word-count) | strings and lists, a proof by induction |
|
|
309
|
-
| [roman-numerals](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/roman-numerals) | a round-trip theorem checked for every number below 1000 (`decide +kernel`) |
|
|
310
|
-
| [todo-cli](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/todo-cli) | an invariant over a list kept by every operation, JSON storage, tampered JSON rejected |
|
|
311
|
-
| [pricing-api](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/pricing-api) | FastAPI with a pydantic model and a Lean inductive type as request body, 422 for invalid input |
|
|
312
|
-
|
|
313
|
-
[`samples/steps/`](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/steps) introduces one
|
|
314
|
-
feature at a time, from functions on `Nat` (01) to types shared between blocks with `uses=` (09).
|
|
315
|
-
|
|
316
|
-
## The CLI
|
|
317
|
-
|
|
318
|
-
The Python package installs `leanfran`:
|
|
319
|
-
|
|
320
|
-
```
|
|
321
|
-
leanfran sync --package <dir> [--leanfran-python <path>]
|
|
322
|
-
leanfran check --package <dir> [--leanfran-python <path>] [--require proved]
|
|
323
|
-
leanfran status --package <dir>
|
|
324
|
-
```
|
|
325
|
-
|
|
326
|
-
`--leanfran-python` defaults to `$LEANFRAN_PYTHON_CLI`, then `leanfran-python` on `PATH`. The Lean CLI
|
|
327
|
-
`leanfran-python` does the compilation:
|
|
328
|
-
|
|
329
|
-
```
|
|
330
|
-
leanfran-python module generate --module-name <Name> [--roots <Name> ...] [--namespace <Name>]
|
|
331
|
-
[--import <Name> ...] [--search-path <dir> ...] [--output-file <path>] [--internal-only]
|
|
332
|
-
[--output json|text] [--type-module <Name>=<module> ...] [--model-module <Name>=<module> ...]
|
|
333
|
-
[--batch <file>]
|
|
334
|
-
leanfran-python obligations check --modules <Name> ... --obligations <file> [--search-path <dir> ...]
|
|
335
|
-
leanfran-python --version | --generator-version | help
|
|
336
|
-
```
|
|
337
|
-
|
|
338
|
-
`module generate` writes one Python module with the IR of the roots and of everything they call
|
|
339
|
-
(Init and Std included), and public functions and types for the roots. It reports the definitions it
|
|
340
|
-
cannot expose (polymorphic functions, functions over functions) as skipped. Both CLIs use the exit
|
|
341
|
-
codes 0 success, 252 invalid parameters, 254 input that cannot be converted, 255 other errors.
|
|
342
|
-
|
|
343
|
-
## Performance
|
|
344
|
-
|
|
345
|
-
`make bench` (CPython 3.14, x86_64; the native column is the C backend):
|
|
346
|
-
|
|
347
|
-
| workload | n | Python (s) | native C (s) |
|
|
348
|
-
|---|---|---|---|
|
|
349
|
-
| `Array.push`, `set!`, sum | 10⁴ | 0.017 | 0.005 |
|
|
350
|
-
| `Array.push`, `set!`, sum | 10⁵ | 0.179 | 0.006 |
|
|
351
|
-
| `Array.push`, `set!`, sum | 10⁶ | 1.796 | 0.017 |
|
|
352
|
-
| `fib` (loop) | 10⁵ | 0.176 | 0.074 |
|
|
353
|
-
| `naiveFib` | 25 | 0.149 | 0.005 |
|
|
354
|
-
| `String.push` | 10⁴ | 0.012 | 0.005 |
|
|
355
|
-
| `String.push` | 10⁵ | 0.227 | 0.006 |
|
|
356
|
-
| `Array.qsort` | 10⁵ | 1.929 | 0.020 |
|
|
357
|
-
|
|
358
|
-
Passing 10⁴ models to a Lean function takes 0.036 s (pydantic's `TypeAdapter(list[Model])` validates
|
|
359
|
-
10⁴ dicts in 0.009 s).
|
|
360
|
-
|
|
361
|
-
## Development
|
|
362
|
-
|
|
363
|
-
```sh
|
|
364
|
-
nix develop -c make check # treefmt, the diff tests against the C backend, pytest, the samples
|
|
365
|
-
nix develop -c make bench
|
|
366
|
-
```
|
|
367
|
-
|
|
368
|
-
- `LeanPython/EmitPy.lean`: IR → Python. `LeanPython/Marshal.lean`, `LeanPython/Generate.lean`: public
|
|
369
|
-
types and functions. `LeanPython/Check.lean`: `obligations check`. `Main.lean`: the CLI.
|
|
370
|
-
- `src/leanfran/runtime.py`, `io_externs.py`: the runtime (`lean_*` are the C names of Lean's
|
|
371
|
-
externs). `marshal.py`, `types.py`: pydantic types. `embed.py`, `sync.py`, `models.py`, `cli.py`:
|
|
372
|
-
`lean(...)` and `leanfran sync`. `verify.py`, `translate.py`, `check.py`: `@model` and
|
|
373
|
-
`leanfran check`; `lean/LeanPythonLib/Py.lean` holds Python's semantics in Lean.
|
|
374
|
-
- `LeanPythonTest/`: Lean programs of the tests; `test-diff.sh` runs `Vectors` and `Io` natively and
|
|
375
|
-
as Python.
|
|
376
|
-
|
|
377
|
-
## Limitations
|
|
378
|
-
|
|
379
|
-
- leanfran does not expose polymorphic functions, functions taking or returning functions, or
|
|
380
|
-
`Thunk`/`Task`/`IO.Ref`/`ByteArray` in public signatures. They still run inside Lean code.
|
|
381
|
-
- `Task`s run synchronously.
|
|
382
|
-
- Building a `String` with `String.push` copies the string each time (Python `bytes`), so it grows
|
|
383
|
-
quadratically.
|
|
384
|
-
- The translation of Python functions covers the subset above; the guarantee is partial correctness.
|
|
385
|
-
|
|
386
|
-
## License
|
|
387
|
-
|
|
388
|
-
[Apache-2.0](https://github.com/conao3/lean-leanfran/blob/master/Python/LICENSE)
|
leanfran-0.6.0/README.md
DELETED
|
@@ -1,374 +0,0 @@
|
|
|
1
|
-
<p align="center">
|
|
2
|
-
<img src="https://raw.githubusercontent.com/conao3/lean-leanfran/master/Python/docs/assets/banner.png" alt="leanfran: write and prove functions in Lean 4, run them as Python" width="880">
|
|
3
|
-
</p>
|
|
4
|
-
|
|
5
|
-
<p align="center"><strong>Write and prove functions in Lean 4, run them as Python</strong></p>
|
|
6
|
-
|
|
7
|
-
<p align="center">
|
|
8
|
-
<a href="https://pypi.org/project/leanfran/"><img src="https://img.shields.io/pypi/v/leanfran" alt="PyPI version"></a>
|
|
9
|
-
<a href="https://github.com/conao3/lean-leanfran/actions/workflows/ci.yml"><img src="https://github.com/conao3/lean-leanfran/actions/workflows/ci.yml/badge.svg?branch=master" alt="CI"></a>
|
|
10
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/LICENSE"><img src="https://img.shields.io/github/license/conao3/lean-leanfran" alt="License: Apache-2.0"></a>
|
|
11
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/pyproject.toml"><img src="https://img.shields.io/badge/Python-%3E%3D3.12-blue" alt="Python >=3.12"></a>
|
|
12
|
-
<a href="https://github.com/conao3/lean-leanfran/blob/master/Python/lean-toolchain"><img src="https://img.shields.io/badge/Lean-v4.35.0-blue" alt="Lean v4.35.0"></a>
|
|
13
|
-
<a href="https://github.com/conao3/lean-leanfran/stargazers"><img src="https://img.shields.io/github/stars/conao3/lean-leanfran?style=flat" alt="GitHub stars"></a>
|
|
14
|
-
</p>
|
|
15
|
-
|
|
16
|
-
<p align="center">
|
|
17
|
-
<a href="#install">Install</a> ·
|
|
18
|
-
<a href="#lean-blocks">Lean blocks</a> ·
|
|
19
|
-
<a href="#values">Values</a> ·
|
|
20
|
-
<a href="#verifying-python-functions">Verifying Python functions</a> ·
|
|
21
|
-
<a href="#examples">Examples</a> ·
|
|
22
|
-
<a href="#performance">Performance</a> ·
|
|
23
|
-
<a href="https://pypi.org/project/leanfran/">PyPI</a>
|
|
24
|
-
</p>
|
|
25
|
-
|
|
26
|
-
# leanfran
|
|
27
|
-
|
|
28
|
-
leanfran lets you write functions in Lean 4, prove theorems about them, and call them from Python
|
|
29
|
-
with pydantic models as arguments and results. The PyPI distribution is `leanfran`; you import
|
|
30
|
-
it as `leanfran`.
|
|
31
|
-
|
|
32
|
-
```python
|
|
33
|
-
from pydantic import BaseModel, Field
|
|
34
|
-
from leanfran import lean
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
class LineItem(BaseModel):
|
|
38
|
-
sku: str
|
|
39
|
-
qty: int = Field(ge=1)
|
|
40
|
-
unitPrice: int = Field(ge=0)
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
pricing = lean("pricing", r"""
|
|
44
|
-
inductive Discount where
|
|
45
|
-
| none
|
|
46
|
-
| percent (p : Nat) (h : p ≤ 100)
|
|
47
|
-
| fixed (amount : Nat)
|
|
48
|
-
|
|
49
|
-
def subtotal (items : List LineItem) : Nat :=
|
|
50
|
-
items.foldl (fun acc i => acc + i.qty * i.unitPrice) 0
|
|
51
|
-
|
|
52
|
-
def total (items : List LineItem) (d : Discount) : Nat :=
|
|
53
|
-
match d with
|
|
54
|
-
| .none => subtotal items
|
|
55
|
-
| .percent p _ => subtotal items - subtotal items * p / 100
|
|
56
|
-
| .fixed a => subtotal items - a
|
|
57
|
-
|
|
58
|
-
theorem total_le (items : List LineItem) (d : Discount) : total items d ≤ subtotal items := by
|
|
59
|
-
unfold total; split <;> omega
|
|
60
|
-
""", models=[LineItem])
|
|
61
|
-
|
|
62
|
-
items = [LineItem(sku="a", qty=2, unitPrice=100)]
|
|
63
|
-
pricing.total(items, pricing.Discount.percent(p=10)) # 180
|
|
64
|
-
pricing.Discount.percent(p=150) # pydantic.ValidationError: p: Input should be less than or equal to 100
|
|
65
|
-
```
|
|
66
|
-
|
|
67
|
-
- `leanfran sync` checks the proofs with Lake and compiles the Lean code from Lean's IR to plain
|
|
68
|
-
Python. It writes the generated modules to `<package>/_lean/`, and they ship with your package.
|
|
69
|
-
To run them you need the `leanfran` wheel and pydantic. You do not need Lean.
|
|
70
|
-
- The generated Python computes what the C backend computes. A diff test runs the same Lean
|
|
71
|
-
programs natively and as Python and compares the output byte for byte. The generated code keeps
|
|
72
|
-
reference counting, so `Array` updates happen in place as in C.
|
|
73
|
-
- Lean structures and inductive types become pydantic models. A proof field (`h : p ≤ 100`) becomes
|
|
74
|
-
a validator that runs a decision procedure compiled from the proposition, so every value that
|
|
75
|
-
reaches Lean satisfies the invariants the proofs rely on. A comparison with a constant also becomes
|
|
76
|
-
an annotated-types constraint (`Le(100)`) that pydantic-core checks and the JSON Schema shows. The
|
|
77
|
-
JSON of the models is the JSON of Lean's `deriving ToJson, FromJson`.
|
|
78
|
-
- Lean code can use your own pydantic models (`models=[...]`). Their field constraints
|
|
79
|
-
(`ge`, `gt`, `le`, `lt`, `min_length`, `max_length`) become proof fields that Lean proofs may assume.
|
|
80
|
-
- The models work wherever pydantic does: `TypeAdapter`, `model_validate_json`, JSON Schema, FastAPI
|
|
81
|
-
request and response types.
|
|
82
|
-
- `@model` binds a plain Python function to Lean specifications. `leanfran check` translates the
|
|
83
|
-
function to Lean and judges your proofs ([Verifying Python functions](#verifying-python-functions)).
|
|
84
|
-
|
|
85
|
-
## Install
|
|
86
|
-
|
|
87
|
-
The Python package and the Lean CLI:
|
|
88
|
-
|
|
89
|
-
```sh
|
|
90
|
-
uv add leanfran # the runtime, lean(), leanfran sync
|
|
91
|
-
nix profile install 'github:conao3/lean-leanfran?dir=Python' # the Lean CLI `leanfran-python`
|
|
92
|
-
```
|
|
93
|
-
|
|
94
|
-
`leanfran sync` also needs `lake` and the Lean toolchain of `lean-toolchain` (v4.35.0) on `PATH`.
|
|
95
|
-
|
|
96
|
-
## Lean blocks
|
|
97
|
-
|
|
98
|
-
`lean(name, source, *, imports=(), uses=(), models=())` defines a Lean block of the calling module,
|
|
99
|
-
which must belong to a package. The block's definitions live in the namespace
|
|
100
|
-
`LeanPythonEmbed.<module>.<name>`. The returned module exposes them under their names relative to it
|
|
101
|
-
(`Discount.percent`, `total`).
|
|
102
|
-
|
|
103
|
-
```sh
|
|
104
|
-
leanfran sync --package src/myapp
|
|
105
|
-
```
|
|
106
|
-
|
|
107
|
-
`leanfran sync` imports every module of the package, writes each block to a Lake workspace in
|
|
108
|
-
`.leanfran/workspace/`, builds it, and generates `src/myapp/_lean/<module>__<name>.py` and
|
|
109
|
-
`src/myapp/_lean/blocks.json`. Lean errors point at the Python file:
|
|
110
|
-
|
|
111
|
-
```
|
|
112
|
-
leanfran: src/myapp/m.py:6: error: (lean block m, Lean 4:28) Tactic `rfl` failed: ...
|
|
113
|
-
```
|
|
114
|
-
|
|
115
|
-
Add `.leanfran/` to `.gitignore` and commit `<package>/_lean/`.
|
|
116
|
-
|
|
117
|
-
At run time `lean(...)` compares a hash of the block (its source, imports, uses, the Lean structures
|
|
118
|
-
of its models and the generator version) with `blocks.json`. If the block changed since the last
|
|
119
|
-
sync, it raises `LeanBlockStale`. With `LEANFRAN_PYTHON_AUTO_SYNC=1` it runs `leanfran sync` instead.
|
|
120
|
-
|
|
121
|
-
- `imports=["Std.Data.HashMap"]`: Lean modules to import.
|
|
122
|
-
- `uses=["geometry"]`: other blocks of the module (or `"pkg.mod:name"`) to import and open. The
|
|
123
|
-
block shares their types and does not generate them again.
|
|
124
|
-
- `models=[LineItem]`: pydantic models available as Lean structures (below).
|
|
125
|
-
|
|
126
|
-
## Values
|
|
127
|
-
|
|
128
|
-
| Lean | Python |
|
|
129
|
-
|---|---|
|
|
130
|
-
| `Nat` | `int` (`leanfran.types.Nat` = `Annotated[int, Ge(0)]`) |
|
|
131
|
-
| `Int` | `int` |
|
|
132
|
-
| `UInt8` … `UInt64`, `USize`, `Int8` … `Int64` | `int` with the range as `Ge`/`Le` |
|
|
133
|
-
| `Fin n` | `Annotated[int, Ge(0), Lt(n)]` |
|
|
134
|
-
| `Bool` | `bool` |
|
|
135
|
-
| `Char` | `str` of one character |
|
|
136
|
-
| `String` | `str` |
|
|
137
|
-
| `Float`, `Float32` | `float` |
|
|
138
|
-
| `Unit` | `None` |
|
|
139
|
-
| `Option α` | `α \| None` |
|
|
140
|
-
| `List α`, `Array α` | `list[α]` |
|
|
141
|
-
| `α × β` | `tuple[α, β]` |
|
|
142
|
-
| `{x : α // p x}` | `Annotated[α, AfterValidator(p)]` |
|
|
143
|
-
| structure | a frozen, strict pydantic model |
|
|
144
|
-
| inductive type, constructors without arguments | `enum.StrEnum` |
|
|
145
|
-
| other inductive type | a model per constructor, subclasses of the type's class |
|
|
146
|
-
| `IO α` | `α`; an `IO.Error` is raised as `LeanIOError` (an `OSError`) |
|
|
147
|
-
|
|
148
|
-
The generated functions check arguments while they convert them: `pub.double(-1)` raises
|
|
149
|
-
`ValidationError` with the location of the bad value. Constants are functions without arguments
|
|
150
|
-
(`answer()`).
|
|
151
|
-
|
|
152
|
-
Constructors and structures take keyword arguments (`Shape.rect(width=1, height=2)`); unnamed
|
|
153
|
-
constructor arguments are `a0`, `a1`, …. Inductive values work with `match`:
|
|
154
|
-
|
|
155
|
-
```python
|
|
156
|
-
match shape:
|
|
157
|
-
case geometry.Shape.rect(width=w, height=h):
|
|
158
|
-
...
|
|
159
|
-
```
|
|
160
|
-
|
|
161
|
-
### JSON
|
|
162
|
-
|
|
163
|
-
The JSON of a generated model is the JSON of Lean's derived `ToJson`: a structure is an object, a
|
|
164
|
-
constructor without arguments is its name (`"dot"`), a constructor with named arguments is
|
|
165
|
-
`{"rect": {"width": 1, "height": 2}}`, with unnamed arguments `{"num": 7}` or `{"pair": [1, "x"]}`.
|
|
166
|
-
`tests/test_json.py` compares the generated models with a native Lean program that prints `toJson`.
|
|
167
|
-
|
|
168
|
-
### Your pydantic models in Lean
|
|
169
|
-
|
|
170
|
-
`models=[LineItem]` writes a Lean structure for each model (and the models its fields use) to the
|
|
171
|
-
namespace `LeanPythonModels.<python module>`, and the block opens that namespace:
|
|
172
|
-
|
|
173
|
-
```lean
|
|
174
|
-
structure LineItem where
|
|
175
|
-
sku : String
|
|
176
|
-
qty : Nat
|
|
177
|
-
qty_ge : 1 ≤ qty
|
|
178
|
-
unitPrice : Nat
|
|
179
|
-
```
|
|
180
|
-
|
|
181
|
-
`int` with a lower bound of at least 0 becomes `Nat`; the other bounds and `min_length`/`max_length`
|
|
182
|
-
become proof fields. Field types:
|
|
183
|
-
|
|
184
|
-
- `int`, `bool`, `str`, `float`, `list[T]`, `tuple[...]`, `T | None`, other models
|
|
185
|
-
- enums: an inductive type `LeanPythonModels.<module>.<Enum>` with a constructor per member
|
|
186
|
-
- `Literal["a", "b"]`: `LeanPythonLiterals.<module>.<Model>_<field>`
|
|
187
|
-
- unions of models: `LeanPythonUnions.<module>.<Model>_<field>`, a constructor per model holding it
|
|
188
|
-
as `val`
|
|
189
|
-
|
|
190
|
-
Lean proofs cannot assume `multiple_of`, `pattern` or validators; the structure's docstring lists
|
|
191
|
-
them. Generated functions take and return your classes. Values that skipped validation
|
|
192
|
-
(`model_construct`) go through validation again on the way to Lean.
|
|
193
|
-
|
|
194
|
-
## Verifying Python functions
|
|
195
|
-
|
|
196
|
-
`@model` binds a typed, pure Python function to specifications written in a Lean block. The function
|
|
197
|
-
stays ordinary Python. `leanfran check` translates it to Lean and checks your proofs, and
|
|
198
|
-
`check_model` runs the same specifications as a Hypothesis property test.
|
|
199
|
-
|
|
200
|
-
```python
|
|
201
|
-
from leanfran import lean
|
|
202
|
-
from leanfran.verify import model
|
|
203
|
-
|
|
204
|
-
spec = lean("spec", r"""
|
|
205
|
-
def clampSpec (x lo hi : Int) : Int := max lo (min hi x)
|
|
206
|
-
def ordered (_x lo hi : Int) : Bool := lo ≤ hi
|
|
207
|
-
def inRange (_x lo hi : Int) (r : Int) : Bool := lo ≤ r && r ≤ hi
|
|
208
|
-
""")
|
|
209
|
-
|
|
210
|
-
@model(refines=spec.clampSpec, requires=spec.ordered, ensures={"in_range": spec.inRange})
|
|
211
|
-
def clamp(x: int, lo: int, hi: int) -> int:
|
|
212
|
-
if x < lo:
|
|
213
|
-
return lo
|
|
214
|
-
if x > hi:
|
|
215
|
-
return hi
|
|
216
|
-
return x
|
|
217
|
-
```
|
|
218
|
-
|
|
219
|
-
`leanfran check --package calc` writes the translation and one obligation per specification:
|
|
220
|
-
|
|
221
|
-
```lean
|
|
222
|
-
def LeanPythonModel.calc.calc.clamp (x : Int) (lo : Int) (hi : Int) : Int := Id.run do
|
|
223
|
-
if ((decide (x < lo))) then
|
|
224
|
-
return lo
|
|
225
|
-
if ((decide (x > hi))) then
|
|
226
|
-
return hi
|
|
227
|
-
return x
|
|
228
|
-
|
|
229
|
-
abbrev LeanPythonModel.calc.calc.clamp.obligation.refines : Prop :=
|
|
230
|
-
∀ (x : Int) (lo : Int) (hi : Int), ordered x lo hi = true → clamp x lo hi = clampSpec x lo hi
|
|
231
|
-
```
|
|
232
|
-
|
|
233
|
-
Then it judges the theorems `<function>.proof.<label>` of `lean/Proofs/*.lean` next to
|
|
234
|
-
`pyproject.toml`:
|
|
235
|
-
|
|
236
|
-
```lean
|
|
237
|
-
import LeanPythonModel.calc.calc
|
|
238
|
-
open LeanPythonEmbed.calc.spec.spec
|
|
239
|
-
namespace LeanPythonModel.«calc».«calc»
|
|
240
|
-
|
|
241
|
-
theorem clamp.proof.refines : clamp.obligation.refines := by
|
|
242
|
-
intro x lo hi h
|
|
243
|
-
simp only [ordered, decide_eq_true_eq] at h
|
|
244
|
-
simp only [clamp, clampSpec]
|
|
245
|
-
split <;> (try split) <;> simp_all <;> omega
|
|
246
|
-
```
|
|
247
|
-
|
|
248
|
-
```
|
|
249
|
-
proved calc.calc:clamp refines
|
|
250
|
-
proved calc.calc:clamp in_range
|
|
251
|
-
sorry calc.calc:sum_to refines
|
|
252
|
-
missing calc.calc:sum_abs nonneg
|
|
253
|
-
```
|
|
254
|
-
|
|
255
|
-
Under `simp`, a `for` loop over a list becomes `List.foldl` (`forIn_pure_yield_eq_foldl`), so you
|
|
256
|
-
prove a loop by induction on the list (`tests/verify/lean/Proofs/Calc.lean` proves `sum_abs`).
|
|
257
|
-
|
|
258
|
-
The states:
|
|
259
|
-
|
|
260
|
-
| state | meaning |
|
|
261
|
-
|---|---|
|
|
262
|
-
| `proved` | the theorem holds |
|
|
263
|
-
| `sorry` | the proof uses `sorry` |
|
|
264
|
-
| `axioms` | the proof uses non-standard axioms |
|
|
265
|
-
| `failed` | the proof file does not build |
|
|
266
|
-
| `missing` | no theorem for the obligation |
|
|
267
|
-
| `mismatch` | the theorem states something else |
|
|
268
|
-
| `unextracted` | the obligation's statement is not defined |
|
|
269
|
-
|
|
270
|
-
With `--require proved`, `leanfran check` exits with 254 unless every obligation is proved. It
|
|
271
|
-
writes the result to `leanfran.lock` with a hash of each function's source. `leanfran status`
|
|
272
|
-
reads the lock file without Lean and marks changed functions `stale`.
|
|
273
|
-
|
|
274
|
-
- `refines=s`: `f x = s x` (`Except.ok (s x)` for a function that may fail); `requires=p`: `p x = true →`
|
|
275
|
-
before each obligation; `ensures={"label": q}`: `q x (f x) = true`. A function annotated `-> Nat`
|
|
276
|
-
also gets `returns_nat` (`0 ≤ f x`).
|
|
277
|
-
- The translation: `int` is `Int` (a `Nat` parameter is a `Nat` converted to `Int` on entry);
|
|
278
|
-
statements become Lean `do` notation (`let mut`, `for`, `while`, `break`, `continue`, early
|
|
279
|
-
`return`); indexing, `//`, `%`, `assert` and `raise` make the function return
|
|
280
|
-
`Py.M α = Except Py.Exc α`, with Python's negative indices and floor division (`LeanPythonLib.Py`).
|
|
281
|
-
Supported: `int`, `bool`, `str`, `float`, `list`, `tuple`, `T | None`, pydantic models (as the
|
|
282
|
-
Lean structures of `models=`), `len`, `abs`, `min`, `max`, `sum`, list comprehensions, `append` on
|
|
283
|
-
local lists, and calls of other `@model` functions of the module.
|
|
284
|
-
- `check_model(f)` / `check_module(m)` generate arguments from the annotations (pydantic constraints
|
|
285
|
-
included), skip those failing `requires`, and compare with `refines` and `ensures`.
|
|
286
|
-
|
|
287
|
-
## Examples
|
|
288
|
-
|
|
289
|
-
[`samples/`](https://github.com/conao3/lean-leanfran/tree/master/Python/samples) holds Python packages with Lean
|
|
290
|
-
blocks. `make samples` syncs each one, runs it, and compares the output with its `expected.txt`.
|
|
291
|
-
|
|
292
|
-
| sample | shows |
|
|
293
|
-
|---|---|
|
|
294
|
-
| [word-count](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/word-count) | strings and lists, a proof by induction |
|
|
295
|
-
| [roman-numerals](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/roman-numerals) | a round-trip theorem checked for every number below 1000 (`decide +kernel`) |
|
|
296
|
-
| [todo-cli](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/todo-cli) | an invariant over a list kept by every operation, JSON storage, tampered JSON rejected |
|
|
297
|
-
| [pricing-api](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/pricing-api) | FastAPI with a pydantic model and a Lean inductive type as request body, 422 for invalid input |
|
|
298
|
-
|
|
299
|
-
[`samples/steps/`](https://github.com/conao3/lean-leanfran/tree/master/Python/samples/steps) introduces one
|
|
300
|
-
feature at a time, from functions on `Nat` (01) to types shared between blocks with `uses=` (09).
|
|
301
|
-
|
|
302
|
-
## The CLI
|
|
303
|
-
|
|
304
|
-
The Python package installs `leanfran`:
|
|
305
|
-
|
|
306
|
-
```
|
|
307
|
-
leanfran sync --package <dir> [--leanfran-python <path>]
|
|
308
|
-
leanfran check --package <dir> [--leanfran-python <path>] [--require proved]
|
|
309
|
-
leanfran status --package <dir>
|
|
310
|
-
```
|
|
311
|
-
|
|
312
|
-
`--leanfran-python` defaults to `$LEANFRAN_PYTHON_CLI`, then `leanfran-python` on `PATH`. The Lean CLI
|
|
313
|
-
`leanfran-python` does the compilation:
|
|
314
|
-
|
|
315
|
-
```
|
|
316
|
-
leanfran-python module generate --module-name <Name> [--roots <Name> ...] [--namespace <Name>]
|
|
317
|
-
[--import <Name> ...] [--search-path <dir> ...] [--output-file <path>] [--internal-only]
|
|
318
|
-
[--output json|text] [--type-module <Name>=<module> ...] [--model-module <Name>=<module> ...]
|
|
319
|
-
[--batch <file>]
|
|
320
|
-
leanfran-python obligations check --modules <Name> ... --obligations <file> [--search-path <dir> ...]
|
|
321
|
-
leanfran-python --version | --generator-version | help
|
|
322
|
-
```
|
|
323
|
-
|
|
324
|
-
`module generate` writes one Python module with the IR of the roots and of everything they call
|
|
325
|
-
(Init and Std included), and public functions and types for the roots. It reports the definitions it
|
|
326
|
-
cannot expose (polymorphic functions, functions over functions) as skipped. Both CLIs use the exit
|
|
327
|
-
codes 0 success, 252 invalid parameters, 254 input that cannot be converted, 255 other errors.
|
|
328
|
-
|
|
329
|
-
## Performance
|
|
330
|
-
|
|
331
|
-
`make bench` (CPython 3.14, x86_64; the native column is the C backend):
|
|
332
|
-
|
|
333
|
-
| workload | n | Python (s) | native C (s) |
|
|
334
|
-
|---|---|---|---|
|
|
335
|
-
| `Array.push`, `set!`, sum | 10⁴ | 0.017 | 0.005 |
|
|
336
|
-
| `Array.push`, `set!`, sum | 10⁵ | 0.179 | 0.006 |
|
|
337
|
-
| `Array.push`, `set!`, sum | 10⁶ | 1.796 | 0.017 |
|
|
338
|
-
| `fib` (loop) | 10⁵ | 0.176 | 0.074 |
|
|
339
|
-
| `naiveFib` | 25 | 0.149 | 0.005 |
|
|
340
|
-
| `String.push` | 10⁴ | 0.012 | 0.005 |
|
|
341
|
-
| `String.push` | 10⁵ | 0.227 | 0.006 |
|
|
342
|
-
| `Array.qsort` | 10⁵ | 1.929 | 0.020 |
|
|
343
|
-
|
|
344
|
-
Passing 10⁴ models to a Lean function takes 0.036 s (pydantic's `TypeAdapter(list[Model])` validates
|
|
345
|
-
10⁴ dicts in 0.009 s).
|
|
346
|
-
|
|
347
|
-
## Development
|
|
348
|
-
|
|
349
|
-
```sh
|
|
350
|
-
nix develop -c make check # treefmt, the diff tests against the C backend, pytest, the samples
|
|
351
|
-
nix develop -c make bench
|
|
352
|
-
```
|
|
353
|
-
|
|
354
|
-
- `LeanPython/EmitPy.lean`: IR → Python. `LeanPython/Marshal.lean`, `LeanPython/Generate.lean`: public
|
|
355
|
-
types and functions. `LeanPython/Check.lean`: `obligations check`. `Main.lean`: the CLI.
|
|
356
|
-
- `src/leanfran/runtime.py`, `io_externs.py`: the runtime (`lean_*` are the C names of Lean's
|
|
357
|
-
externs). `marshal.py`, `types.py`: pydantic types. `embed.py`, `sync.py`, `models.py`, `cli.py`:
|
|
358
|
-
`lean(...)` and `leanfran sync`. `verify.py`, `translate.py`, `check.py`: `@model` and
|
|
359
|
-
`leanfran check`; `lean/LeanPythonLib/Py.lean` holds Python's semantics in Lean.
|
|
360
|
-
- `LeanPythonTest/`: Lean programs of the tests; `test-diff.sh` runs `Vectors` and `Io` natively and
|
|
361
|
-
as Python.
|
|
362
|
-
|
|
363
|
-
## Limitations
|
|
364
|
-
|
|
365
|
-
- leanfran does not expose polymorphic functions, functions taking or returning functions, or
|
|
366
|
-
`Thunk`/`Task`/`IO.Ref`/`ByteArray` in public signatures. They still run inside Lean code.
|
|
367
|
-
- `Task`s run synchronously.
|
|
368
|
-
- Building a `String` with `String.push` copies the string each time (Python `bytes`), so it grows
|
|
369
|
-
quadratically.
|
|
370
|
-
- The translation of Python functions covers the subset above; the guarantee is partial correctness.
|
|
371
|
-
|
|
372
|
-
## License
|
|
373
|
-
|
|
374
|
-
[Apache-2.0](https://github.com/conao3/lean-leanfran/blob/master/Python/LICENSE)
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|