fix: properly exclude content from unparsed pages

pdfplumber/structure.py

@@ -236,10 +236,10 @@ def _parse_struct_tree(self) -> None:
         root = resolve1(self.root["K"])
 
         def on_parsed_page(obj: Dict[str, Any]) -> bool:
-            if self.page_dict is not None and "Pg" in obj:
+            if self.page_dict is not None and "Pg" in obj:  # pragma: nocover
                 page_objid = obj["Pg"].objid
                 return page_objid in self.page_dict
-            return True  # pragma: nocover
+            return True
 
         # It could just be a single object ... it's in the spec (argh)
         if isinstance(root, dict):
@@ -251,8 +251,7 @@ def on_parsed_page(obj: Dict[str, Any]) -> bool:
             if repr(ref) in s:
                 continue  # pragma: nocover
             obj = resolve1(ref)
-            # Dereference OBJR right away, skipping objects on
-            # unparsed pages
+            # Deref top-level OBJR skipping refs to unparsed pages
             if isinstance(obj, dict) and "Obj" in obj:  # pragma: nocover
                 if not on_parsed_page(obj):
                     continue
@@ -261,12 +260,12 @@ def on_parsed_page(obj: Dict[str, Any]) -> bool:
             element, children = self._make_element(obj)
             # Similar to above, delay resolving the children to avoid
             # tree-recursion.
-            s[repr(ref)] = self._make_element(obj)
+            s[repr(ref)] = element, children
             for child in children:
                 if isinstance(child, PDFObjRef):
                     d.append(child)
                 elif isinstance(child, dict) and "Obj" in child:
-                    if on_parsed_page(obj):
+                    if on_parsed_page(child):
                         d.append(child["Obj"])
 
         # Traverse depth-first, removing empty elements (unsure how to
@@ -277,12 +276,20 @@ def prune(elements: List[Any]) -> List[Any]:
                 if isinstance(ref, int):
                     next_elements.append(ref)
                     continue
-                if isinstance(ref, dict):
-                    if "MCID" in ref:  # type MCR
+                elif isinstance(ref, dict):
+                    if not on_parsed_page(ref):  # pragma: nocover
+                        continue
+                    if "MCID" in ref:  # pragma: nocover
                         next_elements.append(ref["MCID"])
                         continue
-                    elif "Obj" in ref:  # type OBJR
+                    elif "Obj" in ref:
                         ref = ref["Obj"]
+                elif isinstance(ref, PDFObjRef):
+                    obj = resolve1(ref)
+                    if isinstance(obj, dict) and "Obj" in obj:  # pragma: nocover
+                        if not on_parsed_page(obj):
+                            continue
+                        ref = obj["Obj"]
                 element, children = s[repr(ref)]
                 children = prune(children)
                 # See assertions below
@@ -315,19 +322,19 @@ def _resolve_children(self, seen: Dict[str, Any]) -> None:
             element, children = seen[repr(ref)]
             assert element is not None, "Unparsed element"
             for child in children:
-                if isinstance(child, dict):  # Ugh...
-                    # Deal with MCR and OBJR dictionaries
-                    if "MCID" in child:  # pragma: nocover
-                        child = child["MCID"]
-                    elif "Obj" in child:  # type OBJR
-                        # If we are a single page, skip out-of-page refs
-                        if "Pg" in child and self.page is not None:  # pragma: nocover
-                            if child["Pg"].objid != self.page.pageid:
-                                continue
-                        child = child["Obj"]
                 if isinstance(child, int):
                     element.mcids.append(child)
-                elif isinstance(child, PDFObjRef):
+                elif isinstance(child, dict):
+                    # Skip out-of-page MCIDS (which are obviously wrong!) and OBJRs
+                    if "Pg" in child and self.page is not None:  # pragma: nocover
+                        if child["Pg"].objid != self.page.pageid:
+                            continue
+                    if "MCID" in child:  # pragma: nocover
+                        element.mcids.append(child["MCID"])
+                    elif "Obj" in child:
+                        child = child["Obj"]
+                # NOTE: if, not elif, in case of OBJR above
+                if isinstance(child, PDFObjRef):
                     child_element, _ = seen.get(repr(child), (None, None))
                     if child_element is not None:
                         element.children.append(child_element)

tests/test_structure.py

@@ -882,6 +882,9 @@ def test_chelsea_pdta(self):
         # This page has no structure tree (really!)
         tree8 = pdf.pages[7].structure_tree
         assert tree8 == []
+        # We should also have no structure tree here
+        with pdfplumber.open(path, pages=[8]) as pdf8:
+            assert pdf8.structure_tree == []
         # This page is empty
         tree3 = pdf.pages[3].structure_tree
         assert tree3 == []
@@ -891,12 +894,13 @@ def test_chelsea_pdta(self):
         pdf = pdfplumber.open(path, pages=[3])
         tree2 = pdf.structure_tree
         assert tree2
-        # Compare (mostly). FIXME: ignores weird extra MCID 0 in tree2
+        # Compare modulo page_number
         d = deque(zip(tree1, tree2))
         while d:
             el1, el2 = d.popleft()
             if "page_number" in el1:
                 assert el1["page_number"] == 3
                 assert el1 == el2
             if "children" in el1:
+                assert len(el1["children"]) == len(el2["children"])
                 d.extend(zip(el1["children"], el2["children"]))