Linear-time rendering of deep nesting; html5ever 0.40.1; Python 3.15t wheels

.github/pytest.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+# Installs the wheel built into dist/ into a fresh virtualenv for each Python
+# version given (or the default Python, with none) and runs the tests.
+set -euo pipefail
+
+# Only uv's own Pythons: otherwise a free-threaded 3.15t that the runner has
+# for building would do for "3.15".
+export UV_PYTHON_PREFERENCE=only-managed
+
+for py in "${@:-}"; do
+  rm -rf .venv-test
+  uv venv -q ${py:+-p "$py"} .venv-test
+  if [ -d .venv-test/Scripts ]; then python=.venv-test/Scripts/python.exe; else python=.venv-test/bin/python; fi
+  uv pip install -q -p "$python" pyhtml2text --no-index --no-deps --find-links dist --reinstall
+  uv pip install -q -p "$python" pyhtml2text pytest markdown-it-py html5lib
+  "$python" -c 'import sys; print(sys.version)'
+  "$python" -m pytest
+done

.github/workflows/CI.yml

@@ -59,7 +59,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist -i python3.14t
+          args: --release --out dist -i python3.14t python3.15t
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
           manylinux: auto
           rust-toolchain: ${{ env.PGO_RUST_TOOLCHAIN }}
@@ -68,18 +68,11 @@ jobs:
         with:
           name: wheels-linux-${{ matrix.platform.target }}
           path: dist
-      - uses: astral-sh/setup-uv@v10.3.2
+      - uses: astral-sh/setup-uv@v10.3.3
         if: ${{ matrix.platform.target == 'x86_64' || matrix.platform.target == 'aarch64' }}
       - name: pytest
         if: ${{ matrix.platform.target == 'x86_64' || matrix.platform.target == 'aarch64' }}
-        shell: bash
-        run: |
-          set -e
-          uv venv .venv
-          source .venv/bin/activate
-          uv pip install pyhtml2text --no-index --no-deps --find-links dist --reinstall
-          uv pip install pyhtml2text pytest markdown-it-py html5lib
-          pytest
+        run: .github/pytest.sh 3.10 3.14t 3.15 3.15t
       - name: pytest
         if: ${{ contains(fromJSON('["armv7", "s390x", "ppc64le"]'), matrix.platform.target) }}
         uses: uraimo/run-on-arch-action@v3
@@ -129,7 +122,7 @@ jobs:
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist -i python3.14t
+          args: --release --out dist -i python3.14t python3.15t
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
           manylinux: musllinux_1_2
       - name: Upload wheels
@@ -176,12 +169,18 @@ jobs:
             target: x64
             python_arch: x64
             pgo: true
+            pythons: 3.10 3.14t 3.15 3.15t
           - runner: windows-latest
             target: x86
             python_arch: x86
+            pythons: >-
+              cpython-3.10-windows-x86 cpython-3.14t-windows-x86
+              cpython-3.15-windows-x86 cpython-3.15t-windows-x86
           - runner: windows-11-arm
             target: aarch64
             python_arch: arm64
+            # There is no uv-managed Python 3.10 for arm64 Windows.
+            pythons: 3.11 3.14t 3.15 3.15t
     steps:
       - uses: actions/checkout@v7
       - uses: actions/setup-python@v7
@@ -198,34 +197,34 @@ jobs:
         with:
           target: ${{ matrix.platform.target }}
           args: --release --out dist
-          sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
+          # sccache before 0.19 can't parse the `-Tp <path>` that cc 1.9 and
+          # later pass cl.exe for mimalloc's C++ source (rust-lang/cc-rs#2063).
+          sccache: false
           rust-toolchain: ${{ env.PGO_RUST_TOOLCHAIN }}
       - uses: actions/setup-python@v7
         with:
-          python-version: "3.14t"
+          python-version: |
+            3.14t
+            3.15t
           architecture: ${{ matrix.platform.python_arch }}
       - name: Build free-threaded wheels
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist -i python3.14t
-          sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
+          args: --release --out dist -i python3.14t python3.15t
+          # sccache before 0.19 can't parse the `-Tp <path>` that cc 1.9 and
+          # later pass cl.exe for mimalloc's C++ source (rust-lang/cc-rs#2063).
+          sccache: false
           rust-toolchain: ${{ env.PGO_RUST_TOOLCHAIN }}
       - name: Upload wheels
         uses: actions/upload-artifact@v7
         with:
           name: wheels-windows-${{ matrix.platform.target }}
           path: dist
-      - uses: astral-sh/setup-uv@v10.3.2
+      - uses: astral-sh/setup-uv@v10.3.3
       - name: pytest
         shell: bash
-        run: |
-          set -e
-          uv venv .venv
-          source .venv/Scripts/activate
-          uv pip install pyhtml2text --no-index --no-deps --find-links dist --reinstall
-          uv pip install pyhtml2text pytest markdown-it-py html5lib
-          pytest
+        run: .github/pytest.sh ${{ matrix.platform.pythons }}
 
   macos:
     runs-on: ${{ matrix.platform.runner }}
@@ -258,28 +257,24 @@ jobs:
           rust-toolchain: ${{ env.PGO_RUST_TOOLCHAIN }}
       - uses: actions/setup-python@v7
         with:
-          python-version: "3.14t"
+          python-version: |
+            3.14t
+            3.15t
       - name: Build free-threaded wheels
         uses: PyO3/maturin-action@v1
         with:
           target: ${{ matrix.platform.target }}
-          args: --release --out dist -i python3.14t
+          args: --release --out dist -i python3.14t python3.15t
           sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
           rust-toolchain: ${{ env.PGO_RUST_TOOLCHAIN }}
       - name: Upload wheels
         uses: actions/upload-artifact@v7
         with:
           name: wheels-macos-${{ matrix.platform.target }}
           path: dist
-      - uses: astral-sh/setup-uv@v10.3.2
+      - uses: astral-sh/setup-uv@v10.3.3
       - name: pytest
-        run: |
-          set -e
-          uv venv .venv
-          source .venv/bin/activate
-          uv pip install pyhtml2text --no-index --no-deps --find-links dist --reinstall
-          uv pip install pyhtml2text pytest markdown-it-py html5lib
-          pytest
+        run: .github/pytest.sh 3.10 3.14t 3.15 3.15t
 
   sdist:
     runs-on: ubuntu-latest
@@ -305,10 +300,9 @@ jobs:
       name: pypi
       url: https://pypi.org/p/pyhtml2text
     permissions:
-      # Use to sign the release artifacts
+      # Trusted publishing to PyPI, and signing the attestations
       id-token: write
-      # Used to upload release artifacts
-      contents: write
+      contents: read
       # Used to generate artifact attestation
       attestations: write
     steps:
@@ -319,7 +313,7 @@ jobs:
           subject-path: 'wheels-*/*'
       - name: Install uv
         if: ${{ startsWith(github.ref, 'refs/tags/') }}
-        uses: astral-sh/setup-uv@v10.3.2
+        uses: astral-sh/setup-uv@v10.3.3
       - name: Publish to PyPI
         if: ${{ startsWith(github.ref, 'refs/tags/') }}
         # Trusted publishing (OIDC) - no token stored in the repo.

.gitignore

@@ -1,6 +1,6 @@
 target/
 dist/
-.venv/
+.venv*/
 __pycache__/
 *.so
 .DS_Store

CHANGELOG.md

@@ -1,5 +1,23 @@
 # Changelog
 
+## Unreleased
+
+- Deeply nested lists, definitions, paragraphs and emphasis are rendered in
+  linear time again. In 0.4.0 each level cost time for every level around
+  it: 400,000 nested lists took 49 s, and now take 0.5 s. Output is
+  unchanged.
+- Wheels for free-threaded Python 3.15 (3.15t), alongside 3.14t. The abi3
+  wheels already cover Python 3.15. CI now tests on 3.10, 3.14t, 3.15 and
+  3.15t.
+- Character widths from Unicode 18 (unicode-width 0.2.3). The halfwidth
+  katakana sound marks ﾞ and ﾟ count one column, as terminals show them,
+  rather than none, and characters new in Unicode 18 (emoji, kana,
+  ideographic characters and combining marks) get their own widths. Other
+  text wraps as before.
+- html5ever 0.40.1, still with
+  [servo/html5ever#800](https://github.com/servo/html5ever/pull/800).
+  Output is unchanged. Building from source needs Rust 1.85 or later.
+
 ## 0.4.0 (2026-10-22)
 
 - `html2text()` is rendered by this package's own code. The library it used

Cargo.toml

@@ -12,9 +12,10 @@ crate-type = ["cdylib"]
 [dependencies]
 pyo3 = { version = "0.30", features = ["abi3-py310"] }
 # The HTML parser; patched below.
-html5ever = "0.39"
-# Display width of characters, for wrapping.
-unicode-width = "=0.2.2"
+html5ever = "0.40"
+# Display width of characters, for wrapping. Pinned, since newer Unicode data
+# can change the output.
+unicode-width = "=0.2.3"
 # Reads the Markdown back to check that emphasis parses as intended.
 pulldown-cmark = { version = "0.13", default-features = false }
 

README.md

@@ -18,6 +18,8 @@ html = """<h2>Before you start</h2>
 print(html2markdown(html, base_url="https://example.com/docs/intro.html"))
 ```
 
+Output:
+
 ```markdown
 ## Before you start
 
@@ -100,12 +102,13 @@ To plain text:
 | **pyhtml2text** | **7.0 ms** | **1.3 ms** |
 
 - The GIL is released while converting, so threads scale across cores. On
-  free-threaded Python (3.14t) the module runs without re-enabling the GIL.
+  free-threaded Python (3.14t and 3.15t) the module runs without re-enabling
+  the GIL.
 - Wheels for Linux (x86_64, aarch64), macOS and Windows x64 are built with
   profile-guided optimisation, which makes them about 25% faster. Other
   platforms get the same code without it.
 - Deeply nested HTML is parsed in linear time, so hostile input can't stall a
-  worker: 40,000 nested lists take 0.14 s, against 6 s with stock html5ever,
+  worker: 40,000 nested lists take 0.02 s, against 18 s with stock html5ever,
   and 100,000 levels of nesting followed by unmatched end tags take under
   0.1 s.