ENH: Robustify parsing for Object streams in XRef rebuilding (#2818)

Closes #2817.

Author: pubpub-zz

pypdf/_reader.py

@@ -77,6 +77,7 @@
     NullObject,
     NumberObject,
     PdfObject,
+    StreamObject,
     TextStringObject,
     read_object,
 )
@@ -316,8 +317,6 @@ def _get_object_from_stream(
         obj_stm: EncodedStreamObject = IndirectObject(stmnum, 0, self).get_object()  # type: ignore
         # This is an xref to a stream, so its type better be a stream
         assert cast(str, obj_stm["/Type"]) == "/ObjStm"
-        # /N is the number of indirect objects in the stream
-        assert idx < obj_stm["/N"]
         stream_data = BytesIO(obj_stm.get_data())
         for i in range(obj_stm["/N"]):  # type: ignore
             read_non_whitespace(stream_data)
@@ -999,6 +998,41 @@ def _rebuild_xref_table(self, stream: StreamType) -> None:
             if generation not in self.xref:
                 self.xref[generation] = {}
             self.xref[generation][idnum] = m.start(1)
+
+        logger_warning("parsing for Object Streams", __name__)
+        for g in self.xref:
+            for i in self.xref[g]:
+                # get_object in manual
+                stream.seek(self.xref[g][i], 0)
+                try:
+                    _ = self.read_object_header(stream)
+                    o = cast(StreamObject, read_object(stream, self))
+                    if o.get("/Type", "") != "/ObjStm":
+                        continue
+                    strm = BytesIO(o.get_data())
+                    cpt = 0
+                    while True:
+                        s = read_until_whitespace(strm)
+                        if not s.isdigit():
+                            break
+                        _i = int(s)
+                        skip_over_whitespace(strm)
+                        strm.seek(-1, 1)
+                        s = read_until_whitespace(strm)
+                        if not s.isdigit():  # pragma: no cover
+                            break  # pragma: no cover
+                        _o = int(s)
+                        self.xref_objStm[_i] = (i, _o)
+                        cpt += 1
+                    if cpt != o.get("/N"):  # pragma: no cover
+                        logger_warning(  # pragma: no cover
+                            f"found {cpt} objects within Object({i},{g})"
+                            f" whereas {o.get('/N')} expected",
+                            __name__,
+                        )
+                except Exception:  # could be of many cause
+                    pass
+
         stream.seek(0, 0)
         for m in re.finditer(rb"[\r\n \t][ \t]*trailer[\r\n \t]*(<<)", f_):
             stream.seek(m.start(1), 0)

tests/test_filters.py

@@ -5,7 +5,6 @@
 from io import BytesIO
 from itertools import product as cartesian_product
 from pathlib import Path
-from unittest.mock import patch
 
 import pytest
 from PIL import Image
@@ -225,14 +224,11 @@ def test_ccitt_fax_decode():
 
 
 @pytest.mark.enable_socket()
-@patch("pypdf._reader.logger_warning")
-def test_decompress_zlib_error(mock_logger_warning):
+def test_decompress_zlib_error(caplog):
     reader = PdfReader(BytesIO(get_data_from_url(name="tika-952445.pdf")))
     for page in reader.pages:
         page.extract_text()
-    mock_logger_warning.assert_called_with(
-        "incorrect startxref pointer(3)", "pypdf._reader"
-    )
+    assert "incorrect startxref pointer(3)" in caplog.text
 
 
 @pytest.mark.enable_socket()

tests/test_reader.py

@@ -276,14 +276,22 @@ def test_get_images(src, expected_images):
             False,
             0,
             False,
-            ["startxref on same line as offset", "incorrect startxref pointer(1)"],
+            [
+                "startxref on same line as offset",
+                "incorrect startxref pointer(1)",
+                "parsing for Object Streams",
+            ],
         ),  # error on startxref, but no strict => xref rebuilt,no fail
         (
             False,
             True,
             0,
             False,
-            ["startxref on same line as offset", "incorrect startxref pointer(1)"],
+            [
+                "startxref on same line as offset",
+                "incorrect startxref pointer(1)",
+                "parsing for Object Streams",
+            ],
         ),
     ],
 )
@@ -344,7 +352,10 @@ def test_issue297(caplog):
     assert caplog.text == ""
     assert "Broken xref table" in exc.value.args[0]
     reader = PdfReader(path, strict=False)
-    assert normalize_warnings(caplog.text) == ["incorrect startxref pointer(1)"]
+    assert normalize_warnings(caplog.text) == [
+        "incorrect startxref pointer(1)",
+        "parsing for Object Streams",
+    ]
     reader.pages[0]
 
 
@@ -898,23 +909,28 @@ def test_form_topname_with_and_without_acroform(caplog):
 def test_extract_text_xref_issue_2(caplog):
     # pdf/0264cf510015b2a4b395a15cb23c001e.pdf
     url = "https://corpora.tika.apache.org/base/docs/govdocs1/981/981961.pdf"
-    msg = "incorrect startxref pointer(2)"
+    msg = [
+        "incorrect startxref pointer(2)",
+        "parsing for Object Streams",
+    ]
     reader = PdfReader(BytesIO(get_data_from_url(url, name="tika-981961.pdf")))
     for page in reader.pages:
         page.extract_text()
-    assert normalize_warnings(caplog.text) == [msg]
+    assert normalize_warnings(caplog.text) == msg
 
 
 @pytest.mark.enable_socket()
 @pytest.mark.slow()
 def test_extract_text_xref_issue_3(caplog):
     # pdf/0264cf510015b2a4b395a15cb23c001e.pdf
     url = "https://corpora.tika.apache.org/base/docs/govdocs1/977/977774.pdf"
-    msg = "incorrect startxref pointer(3)"
+    msg = [
+        "incorrect startxref pointer(3)",
+    ]
     reader = PdfReader(BytesIO(get_data_from_url(url, name="tika-977774.pdf")))
     for page in reader.pages:
         page.extract_text()
-    assert normalize_warnings(caplog.text) == [msg]
+    assert normalize_warnings(caplog.text) == msg
 
 
 @pytest.mark.enable_socket()
@@ -1589,3 +1605,15 @@ def test_iss2761():
     reader = PdfReader(BytesIO(get_data_from_url(url, name=name)), strict=False)
     with pytest.raises(PdfReadError):
         reader.pages[0].extract_text()
+
+
+@pytest.mark.enable_socket()
+def test_iss2817():
+    """Test for rebuiling Xref_ObjStm"""
+    url = "https://github.com/user-attachments/files/16764070/crash-7e1356f1179b4198337f282304cb611aea26a199.pdf"
+    name = "iss2817.pdf"
+    reader = PdfReader(BytesIO(get_data_from_url(url, name=name)))
+    assert (
+        reader.pages[0]["/Annots"][0].get_object()["/Contents"]
+        == "A\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0\xa0 B"
+    )