Incorrect PixMap from page due to cached data from other PDF

[stijnvermeeren-swisstopo]

Description of the bug

Cached data from a one PDF file can cause incorrect colors and shapes when generating a PixMap of a page of a different PDF file.

The issue does not appear when only one PDF file is opened, and explicitly clearing the cache with pymupdf.TOOLS.store_shrink(100) also avoids the problem.

How to reproduce the bug

BOZ1.pdf
BUL1.pdf

Just opening one file and generating a PixMap works correctly:

import pymupdf

bul = pymupdf.open("BUL1.pdf")
bul.load_page(0).get_pixmap().save("correct1.png")
bul.close()

Creating a PixMap of the other PDF first, breaks the PixMap of the second PDF (wrong colors and shapes)

import pymupdf

boz = pymupdf.open("BOZ1.pdf")
boz.load_page(0).get_pixmap()
boz.close()

bul = pymupdf.open("BUL1.pdf")
bul.load_page(0).get_pixmap().save("wrong.png")
bul.close()

Clearing the cache in between serves as a workaround:

import pymupdf

boz = pymupdf.open("BOZ1.pdf")
boz.load_page(0).get_pixmap()
boz.close()

pymupdf.TOOLS.store_shrink(100)
bul = pymupdf.open("BUL1.pdf")
bul.load_page(0).get_pixmap().save("correct2.png")
bul.close()

Fragment of the incorrect PixMap:

Corresponding fragment of the original PDF / correct PixMap:

PyMuPDF version

1.25.4

Operating system

Linux

Python version

3.11

[JorjMcKie]

This seems to be an upstream (MuPDF) problem. Let me make a reproducer in C language, so the MuPDF team has an easier hold of this.

[sebras]

This seems to be an upstream (MuPDF) problem. Let me make a reproducer in C language, so the MuPDF team has an easier hold of this.

I created a mutool run script that appears to reproduce this and reported it upstream over at: https://bugs.ghostscript.com/show_bug.cgi?id=708643

Oh, and I added @JorjMcKie to the CC of that bug, I hope you don't mind. :)

[sebras]

The upstream MuPDF bug has been resolved. 🎉

[julian-smith-artifex-com]

Fixed in PyMuPDF-1.26.4.