Skip to content

Latest commit

 

History

History
132 lines (86 loc) · 8.73 KB

README.md

File metadata and controls

132 lines (86 loc) · 8.73 KB
Raw

PyPI version Supported Python Wheel

Important note

image showing deprecation warning

  • Since 2024.11.08:
    • pyright users will see warning if a single string is supplied where collection of string is expected (tuple, set, list etc). In terms of typing, a single str itself is valid as a Sequence, so type checkers normally would not raise alarm when using str in such function parameters, but can induce unexpected runtime behavior. See #64 for more info. mypy does not support this feature (which (ab)uses @deprecated warning).
    • It is possible to verify release files indeed come from GitHub and not maliciously altered. See Release file attestation for detail.
  • Since 2024.08.07:
    • Each release will contain multiple builds; besides normal version, there is an alternative build suitable for multiple XML element subclass. Please check out relevant Installation section for detail (hint: many people don't need to bother with this). Also note that this release requires mypy 1.11 if one chooses to use mypy for type checking. pyright version requirement (1.1.351) is not changed.

Introduction

This repository contains external type annotations for lxml. It can be used by type-checking tools (currently supporting pyright and mypy) to check code that uses lxml, or used within IDEs like VSCode to facilitate development.

Goal ① : Completion

Now the coverage of lxml submodules is complete (unless intentionally rejected, see further below), thus no more considered as partial:

  • lxml.etree
  • lxml.html
    • lxml.html.builder
    • lxml.html.clean (already removed in lxml 5.2.0, this project will follow suite in future)
    • lxml.html.diff
    • lxml.html.html5parser
    • lxml.html.soupparser
  • lxml.isoschematron
  • lxml.objectify
  • lxml.builder
  • lxml.cssselect
  • lxml.sax
  • lxml.ElementInclude

Following submodules will not be implemented due to irrelevance to type checking or other reasons:

  • lxml.etree.Schematron (obsolete and superseded by lxml.isoschematron)
  • lxml.usedoctest
  • lxml.html.usedoctest
  • lxml.html.formfill (shouldn't have existed, this would belong to HTTP libraries like requests or httpx)

Check out project page for future plans and progress.

Goal ② : Support multiple type checkers

Currently the annotations are validated for both pyright and mypy, with pyright recommended because of its greater flexibility and early adoption of newer type checking features.

In the future, there is plan to bring even more type checker support.

Goal ③: Review and test suite

  • All prior lxml-stubs contributions are reviewed thoroughly, bringing coherency of annotation across the whole package
  • Perform runtime check, and compare against static type checker result; this guarantees annotations are indeed working in real world, not just within some cooked up test suite
  • Existing static test suite already vastly expanded, and is under progress of migrating to runtime test
  • Modernize package building infrastructure

Goal ④ : Support for IDEs

Despite having no official PEP, some IDEs support showing docstring from external annotations. This package tries to bring type annotation specific docstrings for some lxml classes and functions, explaining how they can be used. Following screenshots show what would look like in Visual Studio Code, behaving as if docstrings come from real python code:

Stub docstring in VSCode mouseover tooltip

Besides docstring, current annotations are geared towards convenience for code writers instead of absolute logical 'correctness'. The deviation of class inheritance for HtmlComment and friends is one prominent example.

Installation

The normal choice for most people is to fetch package from PyPI, like:

uv pip install -U types-lxml  # using uv
pip install -U types-lxml  # using pip

In the unlikely case PyPI is down, one can directly download wheel from latest release in GitHub, and then perform installation as local file.

As convenience, it is possible to pull type checker directly with extras:

uv pip install -U types-lxml[pyright]
pip install -U types-lxml[mypy]

Choosing the build

Since 2024.08.07 release, there will be two versions of types-lxml. First one is the default one; if there's no problem using it, there's no need to switch.

The second version, types-lxml-multi-subclass, is intended for specific need, namely creation of multiple lxml element subclasses. For example:

  graph TD;
      etree.ElementBase-->MyBaseElement;
      MyBaseElement-->MySubElement1;
      MyBaseElement-->MySubElement2;
Loading

If a parsed or constructed element tree consists of single type of element nodes, it is safe to assume the children or parent of a node are of the same type too. But this assumption does not hold for multiple subclasses. Using diagram above as example, calling .iter() method from MyBaseElement node may produce element of any subelement or even MyBaseElement itself. Therefore output type should be simply MyBaseElement only.

Such scenario is already in effect for lxml.html. <form> element (FormElement) is supposed to contain other form related tags like <input>, <select> etc. But we can't possibly pinpoint single subelement type, so <form> children can only possibly be of type HtmlElement. The multiple subelement scenario is already hardcoded for HtmlElement and ObjectifiedElement within this annotation package, but users may choose to have their own overridden element subclasses (inherit from ElementBase) too.

The 2 paradigms can't coexist within a single type annotation package. See bug #51 that illustrated why multiple build is necessary.

Remember that anybody can only choose one of the 2 builds. It is impossible to install both, as pip just arbitrarily overwrite conflicting files with one another. If in doubt, removing existing package first, then install the one you needed.

Release file attestation

Since 2024.11.08 users can download types-lxml release files and verify that they indeed do originate from GitHub. For those haven't heard of it, this is sort of like gnupg or minisign signatures, but with GitHub backed infrastructure.

After downloading release wheel file (say pip download types-lxml, or browser access to PyPI directly), one can use GitHub cli to verify it comes from this GitHub repository without being altered:

gh at verify types_lxml-2024.11.8-py3-none-any.whl --repo abelcheung/types-lxml

Should generate following result:

Loaded digest sha256:4b4fa7f9e2f1d5f58b98ac9852a75927e4e0f69363249f9cebc78db095c046e0 for file://types_lxml-2024.11.8-py3-none-any.whl
Loaded 1 attestation from GitHub API
✓ Verification succeeded!

sha256:4b4fa7f9e2f1d5f58b98ac9852a75927e4e0f69363249f9cebc78db095c046e0 was attested by:
REPO                   PREDICATE_TYPE                  WORKFLOW
abelcheung/types-lxml  https://slsa.dev/provenance/v1  .github/workflows/release.yml@refs/tags/2024.11.08

History

Type annotations for lxml were initially included in typeshed, but as it was still incomplete at that time, the stubs are ripped out as a separate project. The code was since then under governance of lxml, until 2022 when this fork intended to revamp lxml-stubs completely and emerge into separate project.

types-lxml is a fork of lxml-stubs that strives for the goals described above, so that most people would find it more useful.