ROB: Cope with some issues in pillow (#2595)

Closes #2265.
Closes #2266.

pypdf/_xobj_image_helpers.py

@@ -73,9 +73,7 @@ def _get_imagemode(
         color_components = cast(int, icc_profile["/N"])
         color_space = icc_profile.get("/Alternate", "")
     elif color_space[0] == "/Indexed":
-        color_space = color_space[1]
-        if isinstance(color_space, IndirectObject):
-            color_space = color_space.get_object()
+        color_space = color_space[1].get_object()
         mode2, invert_color = _get_imagemode(
             color_space, color_components, prev_mode, depth + 1
         )
@@ -291,13 +289,15 @@ def _handle_jpx(
     if img1.mode == "RGBA" and mode == "RGB":
         mode = "RGBA"
     # we need to convert to the good mode
-    try:
-        if img1.mode != mode:
-            img = Image.frombytes(mode, img1.size, img1.tobytes())
-        else:
-            img = img1
-    except OSError:
+    if img1.mode == mode or {img1.mode, mode} == {"L", "P"}:  # compare (unordered) sets
+        # L,P are indexed modes which should not be changed.
+        img = img1
+    elif {img1.mode, mode} == {"RGBA", "CMYK"}:
+        # RGBA / CMYK are 4bytes encoding where
+        # the encoding should be corrected
         img = Image.frombytes(mode, img1.size, img1.tobytes())
+    else:  # pragma: no cover
+        img = img1.convert(mode)
     # for CMYK conversion :
     # https://stcom/questions/38855022/conversion-from-cmyk-to-rgb-with-pillow-is-different-from-that-of-photoshop
     # not implemented for the moment as I need to get properly the ICC

pypdf/filters.py

@@ -894,10 +894,13 @@ def _xobj_to_image(x_object_obj: Dict[str, Any]) -> Tuple[Optional[str], bytes,
     img_byte_arr = BytesIO()
     try:
         img.save(img_byte_arr, format=image_format)
-    except OSError:  # pragma: no cover
-        # odd error
+    except OSError:  # pragma: no cover  # covered with pillow 10.3
+        # in case of we convert to RGBA and then to PNG
+        img1 = img.convert("RGBA")
+        image_format = "PNG"
+        extension = ".png"
         img_byte_arr = BytesIO()
-        img.save(img_byte_arr, format=image_format)
+        img1.save(img_byte_arr, format=image_format)
     data = img_byte_arr.getvalue()
 
     try:  # temporary try/except until other fixes of images

tests/test_images.py

@@ -8,6 +8,7 @@
 from io import BytesIO
 from pathlib import Path
 from typing import Union
+from zipfile import ZipFile
 
 import pytest
 from PIL import Image, ImageChops, ImageDraw
@@ -283,3 +284,65 @@ def test_data_with_lf():
     name = "iss2343b0.png"
     img = Image.open(BytesIO(get_data_from_url(url, name=name)))
     assert image_similarity(reader.pages[8].images[9].image, img) == 1.0
+
+
+@pytest.mark.enable_socket()
+def test_oserror():
+    """Cf #2265"""
+    url = "https://github.com/py-pdf/pypdf/files/13127130/Binance.discovery.responses.2.gov.uscourts.dcd.256060.140.1.pdf"
+    name = "iss2265.pdf"
+    reader = PdfReader(BytesIO(get_data_from_url(url, name=name)))
+    reader.pages[2].images[1]
+    # Due to errors in translation in pillow we may not get
+    # the correct image. Therefore we cannot use `image_similarity`.
+
+
+@pytest.mark.parametrize(
+    ("pdf", "pdf_name", "images", "images_name", "filtr"),
+    [
+        (
+            "https://github.com/py-pdf/pypdf/files/13127197/FTX.Claim.SC30.01072023101624File595287144.pdf",
+            "iss2266a.pdf",
+            "https://github.com/py-pdf/pypdf/files/14967061/iss2266a_images.zip",
+            "iss2266a_images.zip",
+            ((0, 0), (1, 0), (4, 0), (9, 0)),  # random pick-up to speed up test
+        ),
+        (
+            "https://github.com/py-pdf/pypdf/files/13127242/FTX.Claim.Skybridge.Capital.30062023113350File971325116.pdf",
+            "iss2266b.pdf",
+            "https://github.com/py-pdf/pypdf/files/14967099/iss2266b_images.zip",
+            "iss2266b_images.zip",
+            ((0, 0), (1, 0), (4, 0), (9, 0)),  # random pick-up to speed up test
+        ),
+    ],
+)
+@pytest.mark.enable_socket()
+def test_corrupted_jpeg_iss2266(pdf, pdf_name, images, images_name, filtr):
+    """
+    Code to create zipfile:
+    import pypdf;zipfile
+
+    with pypdf.PdfReader("____inputfile___") as r:
+     with zipfile.ZipFile("__outputzip___","w") as z:
+      for p in r.pages:
+       for ii,i in enumerate(p.images):
+        print(i.name)
+        b=BytesIO()
+        i.image.save(b,"JPEG")
+        z.writestr(f"image_{p.page_number}_{ii}_{i.name}",b.getbuffer())
+    """
+    url = pdf
+    name = pdf_name
+    reader = PdfReader(BytesIO(get_data_from_url(url, name=name)))
+    url = images
+    name = images_name
+    print(pdf_name, images_name)  # noqa: T201
+    with ZipFile(BytesIO(get_data_from_url(url, name=name)), "r") as zf:
+        for fn in zf.namelist():
+            sp = fn.split("_")
+            p, i = int(sp[1]), int(sp[2])
+            if filtr is not None and (p, i) not in filtr:
+                continue
+            print(fn)  # noqa: T201
+            img = Image.open(BytesIO(zf.read(fn)))
+            assert image_similarity(reader.pages[p].images[i].image, img) >= 0.99