Close file pointer copy in the libtiff encoder if still open

[fcarron]

Resolves #6985

Problem:
When saving many tiff files with compression="group4" I get the error OSError: [Errno 24] Too many open files (after ~8000 images)

def test_pil():
    im_pil = Image.open("T:/test.jpg")
    im_pil = im_pil.reduce(5)
    im_pil = im_pil.convert("1")
    for i in range(10000)
        file = f"{i:03}"
        out_name = os.path.join(
            "T:/trash", os.path.splitext(os.path.basename(file))[0] + ".tif")
        )
        im_pil.save(out_name, format="TIFF", compression="group4")

Without compression="group4" it works fine.

This also happens when a new image is created on each iteration and with explicit close()

def test_pil():
    for i in range(10000):
        file = f"{i:03}"
        im_pil = Image.open("T:/test.jpg")
        im_pil = im_pil.reduce(5)
        im_pil = im_pil.convert("1")
        out_name = os.path.join(
            "T:/trash", os.path.splitext(os.path.basename(file))[0] + ".tif")
        )
        im_pil.save(out_name, format="TIFF", compression="group4")
        im_pil.close()

Changes proposed in this pull request:

The file pointer seems to be cloned by the libtiff library, but not freed at the end.
Closing to file pointer solve the problem for me.

[geoawd]

Hi @fcarron - I opened #6985 (comment) 2 minutes before you posted this merge request. I've made the same changes and this seems to have resolved the issue in #6985 - in the test files I had provided as well. I can test this properly next week but thanks for submitting this.

[radarhere]

I presume, like the other reports we've received about this, that you're a Windows user?

[fcarron]

@radarhere Yes, exactly (sorry I forgot to mention that),
My setup:
Microsoft Windows 11 Pro, 10.0.22621 Build 22621
Python 3.11.2
Pillow version 9.4.0
But I first had this issue on an old Windows 7 machine with Python 3.8.6 and Pillow 9.4.0.

[radarhere]

Thanks.

I've created fcarron#1 as a suggestion for a test to add, to help ensure that this change doesn't get broken again in the future.

[geoawd]

Thanks @fcarron for this pull request which has also resolved #6985. I’ve now tested this with a couple of sets of ~40k files and they have completed without issue.

[fcarron]

@radarhere Thank you for adding this test.

[radarhere]

Just to link, this PR is similar to #5946