TST: Issue with merging pdfkit

[MartinThoma]

No description provided.

[MartinThoma]

@Lucas-C This one is pretty weird. With #2086 the merged PDF is now rotated by 180° and mirrored. I have no idea how that can happen.

[MartinThoma]

Also, I'm very open to better similarity metrics. It is over 99% although there are very important differences (just the amount of pixels being different is low).

[codecov]

Codecov Report

Patch and project coverage have no change.

Comparison is base (fb35485) 94.34% compared to head (7d918f5) 94.34%.
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2191   +/-   ##
=======================================
  Coverage   94.34%   94.34%           
=======================================
  Files          43       43           
  Lines        7572     7572           
  Branches     1488     1488           
=======================================
  Hits         7144     7144           
  Misses        263      263           
  Partials      165      165           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[stefan6419846]

I can indeed confirm that this relates to the changes in #2086 which fixed the general watermarking behavior for me. As documented in #2113, page.transfer_rotation_to_content() before the watermarking fixes this, although I do not know why as page.rotation is 0.

[Lucas-C]

@Lucas-C This one is pretty weird. With #2086 the merged PDF is now rotated by 180° and mirrored. I have no idea how that can happen.

Sorry, I miss some context there...
This PR pipeline seems to be passing.
What is the problem exactly?
How could I help?

[stefan6419846]

The PR pipeline is passing due to the xfail = expected failure decorator. The problem is that the watermark is inserted in rotated/mirrored/shifted way and not as you would expect. The relevant change apparently comes from #2086, which you are the author of.

[martin-thoma]

Thanks for explaining @stefan6419846

Just to clarify: I pinged you @Lucas-C because you might have an idea or be interested in the topic. No other reason :-)

I'm adding this xfail test (and later an issue) so that it is easier to work on this topic. I'll investigate the PDF generated by PDFkit. Maybe that one is weird. Or maybe the one created by reportlab.

[stefan6419846]

Just for the record as I have been testing this anyway: This is how it currently renders when running the test as-is: pdfkit_watermarked.png

[MartinThoma]

When I run

from pypdf import PdfReader, PdfWriter
from pathlib import Path
import shutil
import subprocess

GHOSTSCRIPT_BINARY = shutil.which("gs")
SAMPLE_ROOT = Path(".")
RESOURCE_ROOT = Path("../resources")
base_path = SAMPLE_ROOT / "022-pdfkit/pdfkit.pdf"
watermark_path = SAMPLE_ROOT / "013-reportlab-overlay/reportlab-overlay.pdf"

reader = PdfReader(base_path)
base_page = reader.pages[0]
watermark = PdfReader(watermark_path).pages[0]

writer = PdfWriter()
base_page.merge_page(watermark)
writer.add_page(base_page)

for page in writer.pages:
    page.compress_content_streams()

tmp_path = Path(".")

target_png_path = RESOURCE_ROOT / "test_watermarking_reportlab_rendering.png"
pdf_path = tmp_path / "out.pdf"
png_path = tmp_path / "test_watermarking_reportlab_rendering.png"

writer.write(pdf_path)
# False positive: https://github.com/PyCQA/bandit/issues/333
subprocess.run(
    [  # noqa: S603
        GHOSTSCRIPT_BINARY,
        "-r120",
        "-sDEVICE=pngalpha",
        "-o",
        png_path,
        pdf_path,
    ]
)
assert png_path.is_file()

I get:

[MartinThoma]

Please note especially that part:

[Lucas-C]

So what was the origin of this problem in the end?

[stefan6419846]

This still has not yet been fixed and is still marked as an expected failure, thus there is no known origin for now.

[martin-thoma]

I should probably open an issue (@MartinThoma pinging so that I don't forget :-) )

[Lucas-C]

I opened #2203 to solve this