Calling pdf.image() might closes passed BytesIO object

[Niwo1403]

If pdf.image() gets called with a BytesIO object this object might be closed while processing (making the bytes in it unavailable after the execution of the method). Whether the object will be closed depends on the image format (e.g. PNG will be closed, JPG not). Since this might break compatibility with previous versions and acts differently depending on the image format, I assume this is not done on purpose, right?

Error details

Raised error:

Traceback (most recent call last):
  File ".../runfile.py", line 15, in <module>
    pdf.image(image, h=80)
  File "...\venv\Lib\site-packages\fpdf\fpdf.py", line 217, in wrapper
    return fn(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "...\venv\Lib\site-packages\fpdf\fpdf.py", line 3766, in image
    if isinstance(name, io.BytesIO) and _is_svg(name.getvalue().strip()):
                                                ^^^^^^^^^^^^^^^
ValueError: I/O operation on closed file.

Reason for error:

1. Here the BytesIO object is used to create a PIL Image object.
2. The Image object gets used...
3. Here the PIL Image object isn't required anymore and gets closed for some image formats (depending on the taken branch & return of the get_img_info method). Closing the PIL Image object then closes the BytesIO object as well (maybe accidentally).

First appearance of error:
I think this error was added in this commit around 5 months ago and is included since version 2.7.0.

Minimal code

# std
from io import BytesIO
# 3rd party
from fpdf import FPDF
from requests import get

# get image from URL and save it as BytesIO
WORKING_LINK = "https://cdn.mos.cms.futurecdn.net/5PMe5hr8tjSS9Nq5d6Cebe.jpg"
NOT_WORKING_LINK = "https://logos-world.net/wp-content/uploads/2021/10/Python-Logo.png"
image = BytesIO(get(NOT_WORKING_LINK).content)
# create a pdf containing two times the same image, passed as BytesIO
pdf = FPDF()
pdf.add_page()
pdf.image(image, h=80)
pdf.image(image, h=80)  # raises ValueError
pdf.output(name="./out.pdf")

Environment

• Operating System: Windows 10
• Python version: 3.11.0
• fpdf2 version used: 2.7.5 & current master
• (requests version: 2.28.2)

[Lucas-C]

Thank you for the detailed report @Niwo1403!

Since this might break compatibility with previous versions and acts differently depending on the image format, I assume this is not done on purpose, right?

Agreed!

I introduced that change in PR #721 as an attempt to avoid memory leaks.
But this was probably not done right...

We should only close BytesIO instances created by fpdf2!

[Lucas-C]

Closing the PIL Image object then closes the BytesIO object as well (maybe accidentally).

This is indeed a side effect of PIL.Image.close():

print(img_bytes.closed)  # False
img = Image.open(img_bytes)
img.close()
print(img_bytes.closed)  # True

It's done there in Pillow source code:
https://github.com/python-pillow/Pillow/blob/10.0.0/src/PIL/Image.py#L551

[Lucas-C]

I opened #887 to fix this

[Lucas-C]

Fixed!

While waiting for a new fpdf2 release including this bugfix,
you can already test it by installing fpdf2 from git:

pip install git+https://github.com/py-pdf/fpdf2.git@master

[Lucas-C]

@allcontributors please add @Niwo1403 for bug

[allcontributors]

@Lucas-C

I've put up a pull request to add @Niwo1403! 🎉