More fine-grained error management than ImageFile.LOAD_TRUNCATED_IMAGES = True

[NHOrus]

What did you do?

I am working with existing large archive of images, and need to open them to make hashes.
Some images are malformed from the source, some are truncated due to download errors. Whole processing pipeline should be read-only in respect to the archive.

What did you expect to happen?

Being able to selectively tell Pillow what Color profile 'iCCP' and image metadata 'eXIf' chunks in PNG images are not exception-worthy and can be safely replaced with None, while all other problems should have exception raised, so the script crashes and alerts me.

What actually happened?

https://gitlab.com/NHOrus/phash_indexer/-/blob/8c8925413463fe4f892456d72de3fe34cdf2bdae/utils.py#L43-L58
I can only disable verification globally, which induces possibility of silently skipping over unwanted errors.

What are your OS, Python and Pillow versions?

• OS: openSUSE Tumbleweed 20251225
• Python: Python 3.13.9
• Pillow: 12.0.0

python3 -m PIL.report                                                                
--------------------------------------------------------------------
Pillow 12.0.0
Python 3.13.9 (main, Nov 23 2025, 01:32:31) [GCC]
--------------------------------------------------------------------
Python executable is /usr/bin/python3
System Python files loaded from /usr
--------------------------------------------------------------------
Python Pillow modules loaded from /home/nho/.local/lib/python3.13/site-packages/PIL
Binary Pillow modules loaded from /home/nho/.local/lib/python3.13/site-packages/PIL
--------------------------------------------------------------------
--- PIL CORE support ok, compiled for 12.0.0
--- TKINTER support ok, loaded 8.6
--- FREETYPE2 support ok, loaded 2.14.1
--- LITTLECMS2 support ok, loaded 2.17
--- WEBP support ok, loaded 1.6.0
--- AVIF support ok, loaded 1.3.0
--- JPEG support ok, compiled for libjpeg-turbo 3.1.2
--- OPENJPEG (JPEG2000) support ok, loaded 2.5.4
--- ZLIB (PNG/ZIP) support ok, loaded 1.3.1, compiled for zlib-ng 2.2.5
--- LIBTIFF support ok, loaded 4.7.1
--- RAQM (Bidirectional Text) support ok, loaded 0.10.3, fribidi 1.0.15, harfbuzz 12.1.0
*** LIBIMAGEQUANT (Quantization method) support not installed
--- XCB (X protocol) support ok
--------------------------------------------------------------------

def hash_image(image: io.BufferedReader) -> int:
    img: Image.Image
    try:
        img = Image.open(image)
    except UnidentifiedImageError:
        image.seek(0)
        if PngImagePlugin._accept(image.read(8)):
            image.seek(0)
            try:
                img = PngImagePlugin.PngImageFile(image)
            except SyntaxError as ex:
                permissible_msgs = ["bad header checksum in b'iCCP'", "CRC error in chunk eXIf", "bad header checksum in b'eXIf'"]
                if any(msg in str(ex) for msg in permissible_msgs):
                    store = ImageFile.LOAD_TRUNCATED_IMAGES
                    ImageFile.LOAD_TRUNCATED_IMAGES = True
                    image.seek(0)
                    img = PngImagePlugin.PngImageFile(image)
                    ImageFile.LOAD_TRUNCATED_IMAGES = store
                else:
                    raise
        else:
            raise
    imgHash = str(imagehash.dhash(img))
    return twos_complement(imgHash, 64) #convert from hexadecimal to 64 bit signed integer

[radarhere]

Hi. Are you aware of Image.WARN_POSSIBLE_FORMATS from #8063? I think if you read #7993, it will become apparent that both you and that user had the same problem, but your suggested solution is different. With this setting, you could write

import io
import warnings
from PIL import Image, ImageFile, UnidentifiedImageError
Image.WARN_POSSIBLE_FORMATS = True

def hash_image(image: io.BufferedReader) -> int:
    img: Image.Image
    store = ImageFile.LOAD_TRUNCATED_IMAGES
    with warnings.catch_warnings(record=True) as w:
        try:
            img = Image.open(image)
        except UnidentifiedImageError:
            permissible_msgs = ["bad header checksum in b'iCCP'", "CRC error in chunk eXIf", "bad header checksum in b'eXIf'"]
            if any(msg in str(x.message) for x in w for msg in permissible_msgs):
                ImageFile.LOAD_TRUNCATED_IMAGES = True
                img = Image.open(image)
                ImageFile.LOAD_TRUNCATED_IMAGES = store
            else:
                raise
    imgHash = str(imagehash.dhash(img))
    return twos_complement(imgHash, 64) #convert from hexadecimal to 64 bit signed integer

I think you're really asking for a callback function to be checked for each error raised in open(), but it feels like that's stepping on the toes of Image.WARN_POSSIBLE_FORMATS. An alternate solution might be to loop over each format in turn. Here's how I would implement that. It's not any simpler than your solution, but at least it doesn't use any private APIs.

import io
import struct
from PIL import Image, ImageFile, UnidentifiedImageError

def hash_image(image: io.BufferedReader) -> int:
    img: Image.Image
    store = ImageFile.LOAD_TRUNCATED_IMAGES
    prefix = image.read(16)
    Image.init()
    for factory, accept in Image.OPEN.values():
        if accept and accept(prefix):
            image.seek(0)
            try:
                try:
                    img = factory(image)
                    break
                except SyntaxError as ex:
                    permissible_msgs = ["bad header checksum in b'iCCP'", "CRC error in chunk eXIf", "bad header checksum in b'eXIf'"]
                    if any(msg in str(ex) for msg in permissible_msgs):
                        ImageFile.LOAD_TRUNCATED_IMAGES = True
                        image.seek(0)
                        img = factory(image)
                        ImageFile.LOAD_TRUNCATED_IMAGES = store
                        break
                    else:
                        raise
            except (SyntaxError, IndexError, TypeError, struct.error):
                continue
    else:
        raise UnidentifiedImageError("cannot identify image file")
    imgHash = str(imagehash.dhash(img))
    return twos_complement(imgHash, 64) #convert from hexadecimal to 64 bit signed integer

[NHOrus]

Thank you. Will try to implement when my current solution breaks.