Segfault in fitz.py

[rikfaith]

Describe the bug (mandatory)

When iterating over fitz.Document(), a segfault occurs:

Fatal Python error: Segmentation fault

Current thread 0x00007ff9fbecd040 (most recent call first):
File "/usr/lib/python3/dist-packages/fitz/fitz.py", line 4241 in page_count
File "/usr/lib/python3/dist-packages/fitz/fitz.py", line 5418 in contains
File "/usr/lib/python3/dist-packages/fitz/fitz.py", line 5440 in getitem
File "/home/faith/git/lustro/lustro/epub.py", line 51 in _extract_text
File "/home/faith/git/lustro/lustro/epub.py", line 167 in read
File "/home/faith/git/lustro/lustro/identify.py", line 161 in identify
File "/home/faith/git/lustro/lustro/main.py", line 135 in main
File "/home/faith/git/lustro/bin/lustro", line 11 in

Extension modules: fitz._fitz, unrardll.unrar, PIL._imaging, tesserocr, psutil._psutil_linux, psutil._psutil_posix (total: 6)

To Reproduce (mandatory)

Try to read an epub that causes https://bugs.ghostscript.com/show_bug.cgi?id=706093

Sorry, I cannot provide the epub.

Expected behavior (optional)

No segfault.

Screenshots (optional)

None.

Your configuration (mandatory)

python 3.11.0+ (main, Nov 4 2022, 09:23:33) [GCC 12.2.0]
PyMuPDF 1.21.0 (20221108000001), from debian package "python3-fitz"
Linux Linux-5.19.0-2-amd64-x86_64-with-glibc2.36
Debian GNU/Linux bookworm/sid

For example, the output of print(sys.version, "\n", sys.platform, "\n", fitz.__doc__) would be sufficient (for the first two bullets).

3.10.8 (main, Nov 4 2022, 09:21:25) [GCC 12.2.0]
linux
PyMuPDF 1.21.0: Python bindings for the MuPDF 1.21.0 library.
Version date: 2022-11-08 00:00:01.
Built for Python 3.10 on linux (64-bit).

Additional context (optional)

Mostly I wanted to call your attention to https://bugs.ghostscript.com/show_bug.cgi?id=706093

When PyMuPDF builds, does it depend on a fully built mupdf tree to access the libmupdf.a static library? I.e., vs. using a dynamic linkage that I could patch around until upstream fixes the problem.

[julian-smith-artifex-com]

Default builds of PyMuPDF (e.g. the wheels on pypi.org) statically link with an internally-built libmupdf.a library. So you can't use dynamic linkage to use a patched MuPDF.

Instead, you will need to build PyMuPDF yourself, setting PYMUPDF_SETUP_MUPDF_BUILD to the path of your patched mupdf/ directory. Also see: https://pymupdf.readthedocs.io/en/latest/installation.html#install-from-source-without-using-an-sdist

[JorjMcKie]

@rikfaith - it looks like you already have identified this as a MuPDF problem, haven't you?
At least I couldn't see what went wrong in PyMuPDF specifically.
It this is true, we would like to appropriately flag this issue as "upstream" problem.

[rikfaith]

@JorjMcKie Yes, I believe this is an upstream problem, and can be flagged as such.

[rikfaith]

@julian-smith-artifex-com Thanks for the name of the environment variable and the pointer to the build instructions. I did a "make shared" in the patched mupdf tree, then a "python3 setup.py build" in the PyMuPDF tree, and successfully extracted text from the epub without getting a segfault.

[julian-smith-artifex-com]

Great, i'm glad it worked for you.

[You didn't actually need to do make shared in mupdf - PyMuPDF/setup.py does its own mupdf build, typically in mupdf/build/pymupdf-x86_64-release.]

[julian-smith-artifex-com]

Fixed in PyMuPDF-1.21.1.