zimscraperlib.image.probing.get_colors relies on colorthief (pinned to 0.2.1), which calls Image.Image.getdata(). Pillow 13 deprecates this method and will remove it in Pillow 14 (2027-10-15):

colorthief.py:65: DeprecationWarning: Image.Image.getdata is deprecated and will be removed in Pillow 14 (2027-10-15). Use get_flattened_data instead.
    pixels = image.getdata()

This is currently the only warning raised by our test suite (tests/image/test_image.py::test_colors_*).

Our pillow<14.0 upper bound protects us for now, but it means we will be stuck on Pillow 13 until this is solved.

colorthief is unmaintained

• last release (0.2.1) is from February 2017, last commit from 2019
• a fengsp/color-thief-py#36 issue about this exact deprecation got no answer from the maintainer
• an upstream fix and release is hence very unlikely

Options

1. Vendor colorthief into zimscraperlib (BSD-licensed, ~420 lines including docstrings, pure Python, only depends on Pillow), replacing getdata() by get_flattened_data().
   • pros: no new dependency, exact same colors as today (no change for scrapers), we can type it properly and drop the many pyright: ignore in get_colors
   • cons: we own the maintenance of the code (but the MMCQ algorithm is stable, only Pillow API evolutions are expected to matter)
2. Switch to modern_colorthief, a Rust re-implementation created precisely because of this deprecation (MIT, actively maintained, last release 0.3.0 in June 2026).
   • pros: maintained, fast, no Python dependency
   • cons: single maintainer and low adoption (supply-chain risk), compiled wheels needed for all platforms our scrapers run on (amd64/arm64, glibc/musl), results are slightly different from colorthief (see their DIFFERENCES.md) so our expected colors in tests (and scrapers output) would change slightly; to be assessed on our test images
3. Reimplement with Pillow only (Image.quantize(..., method=Quantize.MEDIANCUT) + getcolors()).
   • pros: no dependency, very little code
   • cons: different algorithm, so different colors; colorthief also ignores white and transparent pixels, which we would need to replicate

Other packages considered and discarded: fast-colorthief (needs numpy, no release since 2023, no wheels for recent Python), Pylette (way too heavy: OpenCV, scikit-learn, ...), extcolors and haishoku (unmaintained since 2019/2020).

Proposal

Option 1 seems the most conservative one: it removes an unmaintained dependency without changing any output nor adding a new third-party dependency. Option 2 is worth considering if we prefer not to own this code.