remove python3 typehints and transform source code to older python compatibility

This project has been inspired by 'strip-hints' and 'py-backwards' but it is now based on the Python 3.9+ standard library's 'ast.unparse()'. More specifically it is using 'ast_comments' to keep the comments. The implementation wants to make it very easy so that you can add your own source code transformations to support compatibility with older python versions.

The 'strip-python3' tool can extract typehints to '*.pyi' file containing only the outer object-oriented interface. That allows you to ship modern typed python3 code to older systems, possibly even having only python2 preinstalled. The extra pyi however will allow to run it on older systems with a python3.6 installed.

The default configuration is to strip the *.py file to python2 syntax that can also work in python3 (using __print_function__), and (when using -2 or -3) to put a *.pyi file next to it containing the typehints in a syntax compatible with python3.6. When not using -2 or -3 it just prints the type-stripped and backward-transformed python script to stdout.

references

• https://github.com/t3rn0/ast-comments/blob/master/ast_comments.py (copied)
• https://github.com/python/typed_ast (archived)
• https://github.com/nvbn/py-backwards (stopped)
• https://github.com/abarker/strip-hints (still active)

background

The "typed_ast" project was started in April 2016 showing a way to handle type comments. The "py-backwards" project delivered an "unparse()" function for "typed_ast" in May 2017. It also showed a generic approach to code transformers. However the syntactic sugar for python3 grew over time and with only the standard "ast.parse()" to understand all features. Since Python 3.8 in 2019 the "typed_ast" project was officialy discontinued, and finally archived in 2023. Since Python 3.9 in 2020 the standard "ast" module got its own "unparse()" function, basically taking over the approach from "py-backwards". That's nice. The only missing part was that standard python does not even load comments. But in 2022 the "ast-comments" project showed how to hook into the standard module for that - the standard python3's "ast.parse()" got an "enrich()" phase for adding "Comment" nodes, and the standard python3 Unparser class got a "handle_Comment" method to generate decent source code back from it.

Since then the standard python3 features a NodeTransformer class that is supposedly making it easy to rewrite python code. However we are missing examples - if it is used then it is buried deep in other projects. I would like to assemble some of these python code transformations, specifically for removing newer features so that python code can be shipped and used on older systems with some possibly archaic python version. The "py-backwards" has stopped at a point somehwere around python 3.6, so a new start is required with a code base that makes it easier to get contributions by users.

Specifically for removing typehints, the "strip-hints" project shows how to use the standard python "tokenizer" module which is clearly better than using some grep/sed style text transformations. In both text-based variants however you need to guess what kind of code you have on a specific line, which is where "ast"-based approaches have a much easier time. The standard python3 "unparse()" however produces code that is effectivly reformatted. As it follows the python standard formatting that is alrady good enough for shipping via pypi and other distribution channels. The differences to the output of code beautyfiers have become minimal. Note however that the code removal does sometimes make comments appear on lines where they don't belong to. And it seems 'ast.unparse' has not been used extensivly so far.

split your deployment

Pydantic requires atleast Python 3.8 (v2.10 Jan 2025). Mypy requires atleast Python 3.9 (v1.15 Feb 2025). The main Linux distros ship with an additional Python 3.11 (EPEL v9, Almalinux 9.5, Opensuse 15.4+) or default to atleast Python 3.10 (Ubuntu 22.04 python3 is 3.10, Ubuntu 24.04 python3 is 3.12). Many developers have switched their IDE like Visual Code to atleast Python 3.10 in 2024. That has a major impact why modern Python features are getting used.

With strip-python3 developers can continue to work and test with a modern python version on their own infrastructure. But for the delivery to some customer servers and to publication sites like pypi.org, you can add a split-step in between which extracts the pyi type hints and downgrades some features to be compatible with say atleast Python 3.6.

That's what I am doing. The splitted sources are put into a subdirectory and pip-build and twine-upload are run from there. That is actually easy.

configuration

Configuration occurs in this order - with last overriding first

• environment variables - on PYTHON3_REMOVE_POSITIONAL=true
• setup.cfg settings - in [strip-python3] on remove_positional=true
• pyproject.toml settings - in [tool.strip-python3] on remove_positional=true
• commandline options - using --remove-positional option
• setup.cfg python-version in [strip-python3] on python_version=3.6
• pyproject.toml python-version - in [toml.strip-python3] on python_version=3.6
• commandline python-version - using --python-version=3.6
• setup.cfg no-setting [strip-python3] on no_remove_positional=true
• pyproject.toml no-setting - in [toml.strip-python3] on no_remove_positional=true
• commandline no-setting - using --no-remove-positional option

Some implementation options can be selected only by environment variables to allow for extended testing. They don't show up as commandline options while you can see most configuration options from strip_python3.py --help.

transformations

Transformations are automatically selected by the minimum target python version that is given by --python-version=3.6 (which has a shorthand --py36). The default is to support even python2.

Each transformation can be selectivly enabled or disabled. This can also be done in a pyproject.toml [tool.strip-python3] or setup.cfg [strip-python3] section. The names are the same as the commandline options, just without a leading "--".

define-print-function

If print() is used then from __future__ import print_function is added if python2 compatibility is requested.

define-float-division

If / is used then from __future__ import division is added if python2 compatibility is requested.

define-absolute-import

If .localtypes is used then from __future__ import absolute_import is added if python2 compatibility is requested.

define-basestring

If isinstance(x, str) is used then each is replaced by isinstance(x, basestring) if python2 compatibility is requested. It does also import basestring = str for python3 versions.

define-range

If range(x) is used then range = xrange is defined for python2 versions if python2 compatibility is required.

define-callable

If callable(x) is used then a def callable boilerplate is added for python3 versions where it was not predefined (3.0 and 3.1)

datetime-fromisoformat

If datetime.datetime.fromisoformat is used then it is replaced by datetime_fromisodate with boilerplate code if python2 or python3 older than 3.7.

subprocess-run

If subprocess.run is used then it is replaced by subprocess_run with boilerplate code if python2 or python3 older than 3.5.

time-monotonic

If time.monotonic is used then it is replaced by time_monotonic with boilerplate code based on time.time. This is needed for python2 or python3 older than 3.3.

import-pathlib2

If import pathlib is used then it is replaced by import pathlib2 if python2 compatiblity is requested. Pathlib exists since python 3.3 but there is a backport available (to be installed seperately)

import-backports-zoneinfo

If import zoneinfo is used then it is replaced by from backports import zoneinfo if python3 compatiblity is requested before python 3.9 (backport requires atleast 3.6)

import-toml

If import tomllib is used then it is replaced by import toml if compatiblity with python older than 3.11 is requested. The external toml package (to be installed seperately) did exist before and was integrated into the standard lib.

replace-fstring

If F"string {part}" is used then it is replaced by "string {}".format(part) if python2 compatibility requested, as well as python3 older than 3.5.

The transformation scheme works nicely with any expression. So that you can write F"string {len(part):4n}: {part}" which gets expanded to:

"string {:4n}: {}".format(len(part), part)

replace-walrus-operator

If if x := fu(): pass is used then it is replaced by x = fu() followed by if x: pass if python2 compatibility requested, as well as python3 older than 3.8.

Currently only if-walrus and while-walrus are supported. Only a direct assignment or compare/binop is supported, i.e if (x:= fu()) > 0 works as well as while (x:= fu()) != "end": print(x). That one gets expanded to:

while True:
    x = fu()
    if x != 'end':
        print(x)
    else:
        break

replace-annotated-typing

If var: Annotated[int, Field(gt=0)] is used then it is replaced by var: int if python2 compatibility requested, as well as python3 older than 3.9.

replace-builtin-typing

If var: list[int] is used then it is replaced by var: List[int] if python2 compatibility requested, as well as python3 older than 3.9.

replace-union-typing

If var: int|str is used then it is replaced by var: Union[int, str] if python2 compatibility requested, as well as python3 older than 3.10. It does also replace var: int|None by var: Optional[int]

replace-self-typing

If param: Self is used then it is replaced by param: SelfB if python2 compatibility requested, as well as python3 older than 3.11. It does declare a SelfB = TypeVar("SelfB", bound="B") as suggested in PEP 673, and it works for the return type of a class method as well.

remove-keywordsonly

The keywordsonly syntax became available in python 3.0, so it need to be removed for python2

remove-positionalonly

The postionalonly syntax became available in python 3.8, so it need to be removed for python2 and python3 older than 3.8

remove-var-typehints

The var typehints became available in python 3.6, so they need to be removed for older python3 (or python2)

remove-typehints

The function annotations became available in python 3.0, so they need to be removed for python2. Note that the typehints syntax became available in python 3.5 - before that the "typing" module did not exist.

...

pyi typehint include files generation

The "*.pyi" file generation needs to be requested with the "--pyi" commandline option.

The transformations are automatically selected by the minimum target python version that is given by --pyi-version=3.6 (which is the default).

remove-pyi-positionalonly

The postionalonly syntax became available in python 3.8, so it need to be removed for python3 compatiblity older than 3.8.

outer interface only

Note that not all variable annotations and function typehints are being exported.

Only the global variables and classes, and the direct class methods and member annotations are reproduced in the "*.pyi" typehints file. That's good enough for type checkers that want to know the type of imported elements in the "*.py" file.

Be aware that there is no type inference done for global variables without a type annotation. They will simply not exist in the pyi typehints file.

option -2 and option -3

If you only provide one input file on the commandline then you can select the output file as "-o output.py". If no output is selected then all input files are converted and printed to standard output.

If multiple files are provided on the command line then the output file is selected based on "-1" or "-2" or "-3", where "-1" means to modify the file in place (i.e. overwrite it).

If all your files are named like "myfile3.py" then "-3" will remove the 3 and each output file will be named like "myfile.py" and "myfile.pyi". Otherwise let "-2" append a "_2" so that "myfile.py" becomes "myfile_2.py" and "myfile_2.pyi" after transformations.

If no *.pyi file is wanted then disable it with "--no-pyi" again.

Development

DEVGUIDE.MD for more infos.

I take patches!

https://github.com/gdraheim/strip_python3/issues