Repository navigation
Use pymupdf-lite: pymupdf 2.0 pulls heavy dependencies (onnxruntime, numpy, ...) #372
Description
pymupdf 2.0 now hard-depends on many heavy packages, while 1.28.x had no dependency at all:
networkx, numpy, onnxruntime, pyyaml, pymupdf-layout==2.0, psutil, tabulate, pymupdf4llm==2.0
zimscraperlib only uses pymupdf to open PDFs and extract their text / metadata (zim/indexing.py), so pulling onnxruntime and numpy (hundreds of MB) into every scraper image is not acceptable.
Additionally, onnxruntime has no Python 3.15 wheel yet, so on Python 3.15 pip silently falls back to pymupdf 1.28.2.
For now, pymupdf is going to be pinned to <2.0, in 5.5.1 (for Python 3.14) and in the upcoming Python 3.15 / libzim 4 version.
To revisit when:
- pymupdf makes these dependencies optional (extras) or ships a lean variant of 2.x
- or 1.x stops receiving fixes, especially security ones (2.0 fixed CVE-2026-82035, in its CLI font extraction which we do not use)
Activity
Support only Python 3.15, with libzim 4
benoit74 commentedon Oct 29, 2026
Artifex publishes an official lean variant: pymupdf-lite, built from the same pymupdf/PyMuPDF repository, which is "a cut-down version of PyMuPDF without the PyMuPDF Layout engine".
Findings:
pymupdf2.0 plus itsonnxruntime,numpy,networkx,pymupdf4llm,pymupdf-layout, ... dependenciespymupdf2.0 whoseonnxruntimedependency has no 3.15 wheel)pymupdf(andfitz) package, the fullpymupdfwheel only addspymupdf/layout/*,_features.so,_tgif.soandpymupdf4llm, none of which we use (we only open documents and extract their text and metadata)pymupdf-lite>=2.0,<3.0on both Python 3.14 (main) and Python 3.15 (Support only Python 3.15, with libzim 4 #374), with no other package pulledCaveat: since both distributions install the same
pymupdfpackage, a scraper also depending on the fullpymupdfmust keep both at the same version, and uninstalling one of them removes the files of the other (import pymupdfthen fails whilepymupdf-liteis still listed as installed). Fresh installs (e.g. scrapers Docker images) are not affected.Plan: depend on
pymupdf-lite>=2.0,<3.0instead of pinningpymupdf<2.0, in #373 (5.5.1, Python 3.14) and #374 (6.0.0, Python 3.15). This issue will be closed by #373.[-]pymupdf 2.0 pulls heavy dependencies (onnxruntime, numpy, ...)[/-][+]Use pymupdf-lite: pymupdf 2.0 pulls heavy dependencies (onnxruntime, numpy, ...)[/+]Depend on pymupdf-lite instead of pymupdf
Support only Python 3.15, with libzim 4