Generated PDF is corrupted when using Python 3.13.7 but not with 3.12.10

[TheGiolly]

I have a script that does the following:

1. Reads an input.pdf file that contains a 4x10 grid of labels.
2. Extracts the texts that are present under every bercode.
   Each extracted text matches the name of a pdf file that is contained in the labels folder.
3. A new pdf is generated by arranging the contents of the pdf files contained in the labels folder in a 3x5 grid.

Example:
The input.pdf file contains 25 barcodes/texts of which 10 are ABCD and 15 are EFGH.
The output.pdf file will have 2 pages in which the will be 10 contents of labels/ABCD.pdf and 15 contents of labels/EFGH.pdf.

When using Python 3.13.7, the output.pdf file is corrupted.
When using Python 3.12.10, the output.pdf file is correct.

Environment

fpdf2 version is 2.8.4

Which environment were you using when you encountered the problem?

$ python --version 
Python 3.13.7

$ python -m platform
Windows-11-10.0.26100-SP0

$ python -c "import pypdf;print(pypdf._debug_versions)"
pypdf==6.0.0, crypt_provider=('local_crypt_fallback', '0.0.0'), PIL=11.3.0

The following environment doesn't generate the problem

$ python --version 
Python 3.12.10

$ python -m platform
Windows-11-10.0.26100-SP0

$ python -c "import pypdf;print(pypdf._debug_versions)"
pypdf==6.0.0, crypt_provider=('local_crypt_fallback', '0.0.0'), PIL=11.3.0

Code + PDF

This is a minimal, complete example that shows the issue:
main.zip

These are the output files, generated with Python 3.13.7 and 3.12.10.
Look for example from page 38 onwards.
output_3-12-10.pdf
output_3-13-7.pdf

[stefan6419846]

Thanks for the report. This sounds odd, as pypdf should not be specific to a Python version.

Having a quick look at it, page 38 looks correct for me in both cases. Due to the complexity, I do not have any plans to analyze all of your code (which uses fpdf as well) and compare all of the 197 pages of your documents.

Please provide a real minimal example which does not rely on fpdf and which we can run and analyze properly.

[TheGiolly]

Thanks for your reply and sorry for the "complexity" of my code.
I simplified the code to the minimum where I can still generate the issue.
It doesn't rely on fpdf anymore, so it's definitely an issue with pypdf itself.
I also reduced the size of the input.pdf file, so the processing time is faster.

You can analyze the code from line 32 onwards.
The code before that is just reading the input.pdf file and, while it's necessary, it's not related to the issue.
I also added a few comments to hopefully ease the understanding of the code.

Please let me know if you need any more information.

Thank you!

Code and supporting files: main.zip

Output files, generated with Python 3.13.7 and 3.12.10:
output_3-13-7.pdf
output_3-12-10.pdf

Page 78 from output_3-13-7.pdf (corrupted):

Page 78 from output_3-12-10.pdf (good):

The source of the corrupted part of page 78 is labels/X0021C1XLN.pdf, that in its original form looks like this:

[stefan6419846]

I am sorry, but the output files still have 134 pages and more than 5 MB with highly compressed streams. Ideally, we could reduce this to one specific function call and page.

What happens without compressing identical objects? Does removing another function call seem to solve this? What happens if you just do the page merging for the faulty entry? I do not think that we need the current complexity for showing the general issue, thus we are explicitly asking for a minimal example showing this issue. All "side quests" just complicate analysis.

[TheGiolly]

I've managed to reduce the code even further and got it to the point where it seems to be a repeatability issue.
The code is very simple now: it reads 4 pdf files (X001SJC8X7, X001SEHMOH, X001ZISMG3, X0021C1XLN) and generates an output pdf of one page that contains the 4 files arranged in a single column.
It does this 4 times, generating output0.pdf to output3.pdf files that should be identical.

When using Python 3.13.7, output0.pdf to output2.pdf are correct, while output3.pdf is corrupted.
output3_3-13-7.pdf
When using Python 3.12.10, all the output files are correct.
output3_3-12-10.pdf

Code and supporting files:
main.zip

Please let me know if you need any more information.

Thank you!

[bellini666]

Hi folks,

I was investigating the same issue today, and I think I found a way to fix the reproduction from @TheGiolly

In short, added the PdfWriter in a list to avoid it from being garbage collected until the final PDF gets saved, like:

for n in range(0, 5):
    readers = []  # THIS IS THE LIST

    fp = open(f"output{n}.pdf", "wb")

    writer = PdfWriter()
    output = writer.add_blank_page(width=PaperSize.A4.width, height=PaperSize.A4.height)

    for idx, FNSKU in enumerate(FNSKUs):
        reader = PdfReader(f"labels/{FNSKU}.pdf")

        readers.append(reader)  # ADDING THE READER HERE

        label = reader.pages[0]
        output.merge_transformed_page(
            label, Transformation().translate(0, idx * label.mediabox.height)
        )

    writer.write(fp)

With this, the issue is gone for me on 3.13