GitHub - grandimam/cpython: Experimental Python

⚡ Quick start

Zero install — open in Codespaces

One click gives you a cloud VS Code with the interpreter already compiled. When the terminal opens, just run:

./python -c 'print(None?.upper(), "hi"?.upper())'   # None HI

Codespaces runs on your GitHub account's free quota (120 core-hours/month on personal accounts — no paid plan required). The first build takes a few minutes; the interpreter is ready when the welcome message appears.

Build from source

Takes about five minutes on macOS / Linux:

git clone https://github.com/grandimam/cpython.git
cd cpython
./configure --enable-optimizations
make -j"$(nproc 2>/dev/null || sysctl -n hw.ncpu)"
./python -c 'print("hello"?.upper(), None?.upper())'   # HELLO None

On macOS the built binary is ./python.exe (case-insensitive filesystem); on Linux it's ./python.

Prefer a throwaway, side-by-side install so your system Python is untouched:

./configure --prefix="$HOME/.local/cpython-experimental"
make -j8 && make altinstall
"$HOME/.local/cpython-experimental/bin/python3.16" --version
# Python 3.16.0a0+grandimam.0.1

✨ What's new in this fork

?. — None-aware attribute access

Stop writing x.y if x is not None else None chains. The new ?. operator short-circuits the entire attribute / subscript / call chain to None the moment it hits a None:

class User:
    name: str | None

class Response:
    customer: User | None

# Lowercase name or None
name = response?.customer?.name.lower()

Semantics at a glance

• Only None short-circuits — ""?.upper() is "", 0?.real is 0.
• Short-circuiting covers the whole chain: in x?.f(expensive()), expensive() is never called when x is None.
• It stops at the end of the chain: x?.a + 1 still raises TypeError for x = None.
• Load-only: x?.a = 1, del x?.a and x?.a += 1 are syntax errors.
• Fully integrated: new QUESTIONDOT token, ast.NoneAwareAttribute node, ast.unparse round-trips, dis shows the short-circuit, feature-version gated (ast.parse(src, feature_version=(3, 15)) rejects it).

🧪 Try it

>>> import ast, dis
>>> ast.dump(ast.parse("a?.b.c", mode="eval").body)
"Attribute(value=NoneAwareAttribute(value=Name(id='a'), attr='b'), attr='c')"

>>> def f(r): return r?.customer?.name.lower()
>>> dis.dis(f)
  1           RESUME                   0
              LOAD_FAST                0 (r)
              COPY                     1
              POP_JUMP_IF_NONE        40 (to L1)
              NOT_TAKEN
              LOAD_ATTR                0 (customer)
              COPY                     1
              POP_JUMP_IF_NONE        26 (to L1)
              NOT_TAKEN
              LOAD_ATTR                2 (name)
              LOAD_ATTR                4 (lower)
              PUSH_NULL
              CALL                     0
      L1:     RETURN_VALUE

Run the feature's test-suite:

./python -m test test_grammar test_ast test_tokenize test_unparse test_syntax

🔬 How it works

Each feature is implemented end-to-end, the same way a real CPython change would be:

Regenerate derived files after touching the grammar or ASDL:

make regen-token regen-ast regen-pegen

🗺️ Roadmap

• ?. None-aware attribute access
• ?[] None-aware subscription
• ?? None-coalescing operator
• ??= None-coalescing augmented assignment

Have an idea? Open an issue or ping @grandimam.

⚠️ Caveats

• Experimental. Syntax, AST shapes and bytecode may change or disappear without notice.
• Not for production. Code written against this fork will not run on official CPython.
• Tracks main. Expect rebases; pin a commit if you depend on it.
• Upstream policy. This fork follows CPython's AI tools policy and coding conventions so that features could, in principle, be proposed upstream.

📄 License

Copyright © 2001 Python Software Foundation. All rights reserved. Distributed under the PSF License. See README.rst for the full upstream copyright and license information.

_{Built with ❤️ by @grandimam on top of the amazing work of the CPython core team.}