Highlighting issue distorted on recent versions

[sadaqabdo]

Edited:

Description of the bug

When rendering the same content, I observed a noticeable difference in text highlighting appearance between a system using the Rocky Linux 9 base image and systems running on macOS or Linux Ubuntu.
The issue is not related to the platform but rather to changes in the PyMuPDF version. Specifically, the problem arises due to how recent MuPDF versions handle font override values in the PDF.

On Rocky Linux 9, the text appears slightly distorted, with misaligned or inconsistent lines.
On macOS/Ubuntu, the text renders correctly and appears as expected.

macOS/Ubuntu

Steps to Reproduce:

• Use the following Python script to highlight text in a PDF.

1. Run the script in pymupdf==1.24.x and pymupdf==1.25.x
2. Compare the resulting PDFs to observe the difference

import fitz


def highlight_text_in_pdf(input_pdf, output_pdf, text_to_highlight):
    pdf_document = fitz.open(input_pdf)
    for page_number in range(len(pdf_document)):
        page = pdf_document[page_number]
        text_instances = page.search_for(text_to_highlight)
        for inst in text_instances:
            highlight = page.add_highlight_annot(inst)
            if highlight:
                highlight.update()

    pdf_document.save(output_pdf)
    pdf_document.close()


input_pdf_path = "Bart-Simpson.pdf"
output_pdf_path = "Bart-Simpson-highlighted.pdf"

text_to_search = "Bart Simpson"

highlight_text_in_pdf(input_pdf_path, output_pdf_path, text_to_search)

the file:
Bart-Simpson.pdf

Bart-Simpson.pdf

PyMuPDF version

1.25.2

Operating system

Other

Python version

3.12

[JorjMcKie]

I updated the issue title because it has nothing to do with the platform.
If you had upgraded your PyMuPDF to a current version on your normal platforms MAC/Ubuntu before running the comparison, the same awkward behavior would have happened.

The problem was introduced since MuPDF recently started using font override values provided in the PDF instead of solely relying on values given in the embedded font binary. There exist good reasons for doing so!

The error in fact is buried in the PDF when font descender overrides are given without an appropriate sign:
Descender values should be negative (in 99+ % of all cases), but they are not sometimes:

The value should read /Descent -211.91406!
Hacking the file in this way will again lead to the desired result.

[sadaqabdo]

Thank you for the reply, indeed the bug is only on the 1.25.x version. I downgraded to 1.24.x and the issue is resolved.

PS: I edited the issue description

[JorjMcKie]

The MuPDF team has developed a fix which will intercept and correct wrong descender specifications as we have it in this file.

[JorjMcKie]

This issue is a duplicate of #4180 and #4182.

[julian-smith-artifex-com]

Fixed in PyMuPDF-1.25.3.