ROB: Fix corrupted (wrongly) linear PDF (#1008)

Fix: Rescan the whole PDF and update/rebuild the trailer

Closes #989

PyPDF2/_reader.py

@@ -1448,17 +1448,12 @@ def _rebuild_xref_table(self, stream: StreamType) -> None:
             if generation not in self.xref:
                 self.xref[generation] = {}
             self.xref[generation][idnum] = m.start(1)
-        trailer_pos = f_.rfind(b"trailer") - len(f_) + 7
-        stream.seek(trailer_pos, 2)
-        # code below duplicated
-        read_non_whitespace(stream)
-        stream.seek(-1, 1)
-
-        # there might be something that is not a dict (see #856)
-        new_trailer = cast(Dict[Any, Any], read_object(stream, self))
-
-        for key, value in list(new_trailer.items()):
-            if key not in self.trailer:
+        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)
+            new_trailer = cast(Dict[Any, Any], read_object(stream, self))
+            # Here, we are parsing the file from start to end, the new data have to erase the existing.
+            for key, value in list(new_trailer.items()):
                 self.trailer[key] = value
 
     def _read_xref_subsections(

tests/test_workflows.py

@@ -7,6 +7,7 @@
 
 from PyPDF2 import PdfReader
 from PyPDF2.constants import PageAttributes as PG
+from PyPDF2.errors import PdfReadWarning
 
 from . import get_pdf_from_url
 
@@ -167,15 +168,23 @@ def test_rotate_45():
             "https://corpora.tika.apache.org/base/docs/govdocs1/971/971703.pdf",
             [0, 1, 5, 8, 14],
         ),
+        (  # faulty PDF, wrongly linearized and with 2 trailer, second with /Root
+            True,
+            "https://corpora.tika.apache.org/base/docs/govdocs1/989/989691.pdf",
+            [0],
+        ),
     ],
 )
 def test_extract_textbench(enable, url, pages, print_result=False):
     if not enable:
         return
-    reader = PdfReader(BytesIO(get_pdf_from_url(url, url.split("/")[-1])))
-    for page_number in pages:
-        if print_result:
-            print(f"**************** {url} / page {page_number} ****************")
-        rst = reader.pages[page_number].extract_text()
-        if print_result:
-            print(f"{rst}\n*****************************\n")
+    try:
+        reader = PdfReader(BytesIO(get_pdf_from_url(url, url.split("/")[-1])))
+        for page_number in pages:
+            if print_result:
+                print(f"**************** {url} / page {page_number} ****************")
+            rst = reader.pages[page_number].extract_text()
+            if print_result:
+                print(f"{rst}\n*****************************\n")
+    except PdfReadWarning:
+        pass