Skip to main content

static type checking for Python (but based)

Project description

basedpyright

pypi visual studio marketplace open VSX nvim-lspconfig sublime text pycharm Discord basedpyright - checked

Basedpyright is a fork of pyright with various type checking improvements, improved vscode support and pylance features built into the language server.

📚 Documentation | 🛝 Playground

why?

the main motivation behind this fork was the fact that pyright has several serious issues that the maintainers didn't want to address, and many bugs that they consider to be intentional behavior. here is a list of some of its major problems that basedpyright resolves:

ability to pin the version used by vscode

in pyright, if the vscode extension gets updated, you may see errors in your project that don't appear in the CI, or vice-versa. see this issue.

basedpyright fixes this problem by adding an importStrategy option to the extension, which defaults to looking in your project for the basedpyright pypi package.

published as a pypi package - no nodejs required

pyright is only published as an npm package, which requires you to install nodejs. the version on pypi is just an unofficial wrapper that installs node and the npm package the first time you invoke the cli, which is quite flaky.

python developers should not be expected to have to install nodejs in order to typecheck their python code. it should just be a regular pypi package like mypy, ruff, and pretty much all other python tooling. this is why basedpyright is officially published on pypi, which comes bundled with the npm package.

new diagnostic rules

reportUnreachable - report errors on code that would otherwise be completely unchecked

pyright often incorrectly marks code as unreachable. in most cases, unreachable code is a mistake and therefore should be an error, but pyright does not have an option to report unreachable code. in fact, unreachable code is not even type-checked at all:

if sys.platform == "win32":
  1 + "" # no error

by default, pyright will treat the body in the code above as unreachable if pyright itself was run on an operating system other than windows. this is bad of course, because chances are if you write such a check, you intend for your code to be executed on multiple platforms.

to make things worse, unreachable code is not even type-checked, so the obviously invalid 1 + "" above will go completely unnoticed by the type checker.

basedpyright solves this issue with a reportUnreachable option, which will report an error on such unchecked code. in this example, you can update your pyright config to specify more platforms using the pythonPlatform option if you intend for the code to be reachable.

reportAny - fully ban the Any type

pyright has a few options to ban "Unknown" types such as reportUnknownVariableType, reportUnknownParameterType, etc. but "Unknown" is not a real type, rather a distinction pyright uses used to represent Anys that come from untyped code or unfollowed imports. if you want to ban all kinds of Any, pyright has no way to do that:

def foo(bar, baz: Any) -> Any:
    print(bar) # error: unknown type
    print(baz) # no error

basedpyright introduces the reportAny option, which will report an error on usages of anything typed as Any.

reportIgnoreCommentWithoutRule - enforce that all ignore comments specify an error code

it's good practice to specify an error code in your pyright: ignore comments:

# pyright: ignore[reportUnreachable]

this way, if the error changes or a new error appears on the same line in the future, you'll get a new error because the comment doesn't account for the other error. unfortunately there are many rules in pyright that do not have error codes, so you can't always do this.

basedpyright resolves this by reporting those errors under the reportGeneralTypeIssues diagnostic rule. this isn't a perfect solution, but there were over 100 errors that didn't have diagnostic rules. i intend to split them into their own rules in the future, but this will do for now.

[!WARNING]
type: ignore comments are unsafe and will soon be disabled by default (see #330 and #55). we recommend using pyright: ignore comments instead.

reportPrivateLocalImportUsage - prevent implicit re-exports in local code

pyright's reportPrivateImportUsage rule only checks for private imports of third party modules inside py.typed packages. but there's no reason your own code shouldn't be subject to the same restrictions. to explicitly re-export something, give it a redundant alias as described in the "Stub Files" section of PEP484 (although it only mentions stub files, other type checkers like mypy have also extended this behavior to source files as well):

# foo.py

from .some_module import a  # private import
from .some_module import b as b  # explicit re-export
# bar.py

# reportPrivateLocalImportUsage error, because `a` is not explicitly re-exported by the `foo` module:
from foo import a

# no error, because `b` is explicitly re-exported:
from foo import b

reportImplicitRelativeImport - reporting errors on invalid "relative" imports

pyright allows invalid imports such as this:

# ./module_name/foo.py:
# ./module_name/bar.py:
import foo # wrong! should be `import module_name.foo` or `from module_name import foo`

this may look correct at first glance, and will work when running bar.py directly as a script, but when it's imported as a module, it will crash:

# ./main.py:
import module_name.bar  # ModuleNotFoundError: No module named 'foo'

the new reportImplicitRelativeImport rule bans imports like this. if you want to do a relative import, the correct way to do it is by importing it from . (the current package):

# ./module_name/bar.py:
from . import foo

reportInvalidCast - prevent non-overlapping casts

most of the time when casting, you want to either cast to a narrower or wider type:

foo: int | None
cast(int, foo) #  narrower type
cast(object, foo) #  wider type

but pyright doesn't prevent casts to a type that doesn't overlap with the original:

foo: int
cast(str, foo)

in this example, it's impossible to be foo to be a str if it's also an int, because the int and str types do not overlap. the reportInvalidCast rule will report invalid casts like these.

note about casting with TypedDicts

a common use case of cast is to convert a regular dict into a TypedDict:

foo: dict[str, int | str]
bar = cast(dict[{"foo": int, "bar": str}], foo)

unfortunately, this will cause a reportInvalidCast error when this rule is enabled, because although at runtime TypedDict is a dict, type checkers treat it as an unrelated subtype of Mapping that doesn't have a clear method, which would break its type-safety if it were to be called on a TypedDict.

this means that although casting between them is a common use case, TypedDicts and dicts technically do not overlap.

reportUnsafeMultipleInheritance - ban inheriting from multiple different base classes with constructors

multiple inheritance in python is awful:

class Foo:
    def __init__(self):
        super().__init__()
class Bar:
    def __init__(self):
        ...

class Baz(Foo, Bar):
    ...

Baz()

in this example, Baz() calls Foo.__init__, and the super().__init__() in Foo now calls to Bar.__init__ even though Foo does not extend Bar.

this is complete nonsense and very unsafe, because there's no way to statically know what the super class will be.

pyright has the reportMissingSuperCall rule which, for this reason, complains even when your class doesn't have a base class. but that sucks because there's no way to know what arguments the unknown __init__ takes. so this rule is super annoying when it's enabled, and has very little benefit because it barely makes a difference in terms of type safety.

reportUnsafeMultipleInheritance bans multiple inheritance when any base class except for the first one has an __init__ or __new__ method, as there's no way to guarantee that they will get called with the correct arguments (or at all). this allows reportMissingSuperCall to be more lenient, ie. when reportUnsafeMultipleInheritance is enabled, missing super() calls will only be reported on classes that actually have a base class.

re-implementing pylance-exclusive features

basedpyright re-implements some of the features that microsoft made exclusive to pylance, which is microsoft's closed-source vscode extension built on top of the pyright language server with some additional exclusive functionality (see the pylance FAQ for more information).

the following features have been re-implemented in basedpyright's language server, meaning they are no longer exclusive to vscode. you can use any editor that supports the language server protocol. for more information on installing pyright in your editor of choice, see the installation instructions.

import suggestion code actions

pyright only supports import suggestions as autocomplete suggestions, but not as quick fixes (see this issue).

basedpyright re-implements pylance's import suggestion code actions:

image

semantic highlighting

before after
image image

basedpyright re-implements pylance's semantic highlighting along with some additional improvements:

initial implementation of the semantic highlighting provider was adapted from the pyright-inlay-hints project.

inlay hints

image

basedpyright contains several improvements and bug fixes to the original implementation adapted from pyright-inlay-hints.

errors on invalid configuration

in pyright, if you have any invalid config, it may or may not print a warning to the console, then it will continue type-checking and the exit code will be 0 as long as there were no type errors:

[tool.pyright]
mode = "strict"  # wrong! the setting you're looking for is called `typeCheckingMode`

in this example, it's very easy for errors to go undetected because you thought you were on strict mode, but in reality pyright just ignored the setting and silently continued type-checking on "basic" mode.

to solve this problem, basedpyright will exit with code 3 on any invalid config.

fixes for the reportRedeclaration and reportDuplicateImport rules

pyright does not report redeclarations if the redeclaration has the same type:

foo: int = 1
foo: int = 2  # no error

nor does it care if you have a duplicated import in multiple different import statements, or in aliases:

from foo import bar
from bar import bar  # no error
from baz import foo as baz, bar as baz  # no error

basedpyright solves both of these problems by always reporting an error on a redeclaration or an import with the same name as an existing import.

better defaults

we believe that type checkers and linters should be as strict as possible by default, making the user aware of all the available rules so they can more easily make informed decisions about which rules they don't want enabled in their project. that's why the following defaults have been changed in basedpyright

typeCheckingMode

used to be basic, but now defaults to all. in the future we intend to add baseline to allow for easy adoption of more strict rules in existing codebases.

pythonPlatform

used to assume that the operating system pyright is being run on is the only operating system your code will run on, which is rarely the case. in basedpyright, pythonPlatform defaults to All, which assumes your code can run on any operating system.

improved integration with CI platforms

regular pyright has third party integrations for github actions and gitlab, but they are difficult to install/set up. these integrations are built into basedpyright, which makes them much easier to use.

github actions

basedpyright automatically detects when it's running in a github action, and modifies its output to use github workflow commands. this means errors will be displayed on the affected lines of code in your pull requests automatically:

image

this is an improvement to regular pyright, which requires you to use a third party action that requires boilerplate to get working. basedpyright just does it automatically without you having to do anything special:

# .github/workflows/your_workflow.yaml

jobs:
  check:
    steps:
      - run: ...  # checkout repo, install dependencies, etc
      - run: basedpyright  # no additional arguments required. it automatically detects if it's running in a github action 

gitlab code quality reports

the --gitlabcodequality argument will output a gitlab code quality report which shows up on merge requests:

image

to enable this in your gitlab CI, just specify a file path to output the report to, and in the artifacts.reports.codequality section of your .gitlab-ci.yml file:

basedpyright:
  script: basedpyright --gitlabcodequality report.json
  artifacts:
    reports:
      codequality: report.json

basedmypy feature parity

basedmypy is a fork of mypy with a similar goal in mind: to fix some of the serious problems in mypy that do not seem to be a priority for the maintainers. it also adds many new features which may not be standardized but greatly improve the developer experience when working with python's far-from-perfect type system.

we aim to port most of basedmypy's features to basedpyright, however as mentioned above our priority is to first fix the critical problems with pyright.

note that any non-standard features we add will be optional, as we intend to support library developmers who can't control what type checker their library is used with.

pypi package

basedpyright differs from pyright by publishing the command line tool as a pypi package instead of an npm package. this makes it far more convenient for python developers to use, since there's no need to install any additional tools.

for more information, see the installation instructions.

vscode extension

install

install the extension from the vscode extension marketplace or the open VSX registry

usage

the basedpyright vscode extension will automatically look for the pypi package in your python environment.

if you're adding basedpyright as a development dependency in your project, we recommend addibg it to the recommended extensions list in your workspace to prompt others working on your repo to install it:

// .vscode/extensions.json

{
  "recommendations": ["detachhead.basedpyright"]
}

in .vscode/settings.json, remove any settings starting with python.analysis, as they are not used by basedpyright. you should instead set these settings using the tool.basedpyright (or tool.pyright) section in pyroject.toml (see below)

you should also disable the built in language server support from the python extension, as it conflicts with basedpyright's language server. the basedpyright extension will detect this problem and suggest fixing it automatically.

using basedpyright with pylance (not recommended)

unless you depend on any pylance-exclusive features that haven't yet been re-implemented in basedpyright, it's recommended to disable/uninstall the pylance extension.

if you do want to continue using pylance, all of the options and commands in basedpyright have been renamed to avoid any conflicts with the pylance extension, and the restriction that prevents both extensions from being enabled at the same time has been removed. for an optimal experience you should change the following settings in your .vscode/settings.json file:

  • disable pylance's type-checking by setting "python.analysis.typeCheckingMode" to "off". this will prevent pylance from displaying duplicated errors from its bundled pyright version alongside the errors already displayed by the basedpyright extension.
  • disable basedpyright's LSP features by setting "basedpyright.disableLanguageServices" to true. this will prevent duplicated hover text and other potential issues with pylance's LSP. keep in mind that this may result in some inconsistent behavior since pylance uses its own version of the pyright LSP.
{
    "python.analysis.typeCheckingMode": "off",
    "basedpyright.disableLanguageServices": true
}

(the basedpyright extension will detect this problem and suggest fixing it automatically)

playground

you can try basedpyright in your browser using the basedpyright playground

pre-commit hook

integration with pre-commit is also supported.

# .pre-commit-config.yaml

repos:
  - repo: https://github.com/DetachHead/basedpyright
    rev: v1.8.0
    hooks:
    - id: basedpyright

to ensure that basedpyright is able to find all of the dependencies in your virtual env, add the following to your pyproject.toml:

[tool.basedpyright]
# ...
venvPath = "."

recommended setup

it's recommended to use both the basedpyright cli and vscode extension in your project. the vscode extension is for local development and the cli is for your CI.

below are the changes i recommend making to your project when adopting basedpyright

pyproject.toml

we recommend using pdm with pyprojectx (click the "inside project" tab) to manage your dependencies.

[tool.pyprojectx]
main = ["pdm==2.12.4"]  # installs pdm to your project instead of globally

[tool.pdm.dev-dependencies]  # or the poetry equivalent
dev = [
    "basedpyright", # you can pin the version here if you want, or just rely on the lockfile
]

[tool.basedpyright]
# many settings are not enabled even in strict mode, which is why basedpyright includes an "all" option
# you can then decide which rules you want to disable
typeCheckingMode = "all"

pinning your dependencies is important because it allows your CI builds to be reproducible (ie. two runs on the same commit will always produce the same result). basedpyright ensures that the version of pyright used by vscode always matches this pinned version.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

basedpyright-1.12.1.tar.gz (8.5 MB view hashes)

Uploaded Source

Built Distribution

basedpyright-1.12.1-py3-none-any.whl (10.4 MB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page