Docx reader: preserve non-textbox content when unwrapping textboxes.

Previously, when a w:p paragraph contained runs with textboxes,
the entire paragraph was replaced by just the textbox content,
discarding all other runs (including image-bearing runs).
Now we walk the paragraph's children in order, grouping
non-textbox content into copies of the original w:p and splicing
unwrapped textbox content in place, preserving the original order.

We also treat text inside a textbox containing an image as a
figure caption.  (One often finds captioned images of this
kind in docx files.)

Closes #11510.
Closes #6893.
Closes #11412.
Closes #5394.
Closes #9633.

Co-Authored-By: Claude <noreply@anthropic.com>

src/Text/Pandoc/Readers/Docx/Parse.hs

@@ -160,8 +160,10 @@ unwrapElement ns element
   | isElem ns "w" "smartTag" element
   = concatMap (unwrapElement ns) (elChildren element)
   | isElem ns "w" "p" element
-  , textboxes@(_:_) <- findChildrenByName ns "w" "r" element >>= findTextboxes
-  = concatMap (unwrapElement ns) (concatMap elChildren textboxes) -- handle #9214, #11053
+  , _:_ <- findChildrenByName ns "w" "r" element >>= findTextboxes
+  = let result = splitP (elContent element) [] -- handle #9214, #11053
+    in [element{ elName = QName "textbox-image" Nothing Nothing
+              , elContent = map Elem result }]
   | isElem ns "w" "r" element
   , Just fallback <- findChildByName ns "mc" "AlternateContent" element >>=
                      findChildByName ns "mc" "Fallback"
@@ -169,6 +171,26 @@ unwrapElement ns element
   | otherwise
   = [element{ elContent = concatMap (unwrapContent ns) (elContent element) }]
   where
+    -- Split a w:p's children around textbox-bearing runs, preserving order.
+    -- Non-textbox content is grouped into a copy of the original w:p;
+    -- textbox content is unwrapped into sibling elements in place.
+    splitP [] acc = [wrapP acc | hasContent acc]
+    splitP (c:cs) acc
+      | Elem el <- c
+      , isElem ns "w" "r" el
+      , tbs@(_:_) <- findTextboxes el
+      = [wrapP acc | hasContent acc]
+        ++ concatMap (unwrapElement ns) (concatMap elChildren tbs)
+        ++ splitP cs []
+      | otherwise
+      = splitP cs (acc ++ [c])
+    wrapP cs = element{ elContent = concatMap (unwrapContent ns) cs }
+    hasContent = any (\case Elem el -> isElem ns "w" "r" el
+                                                  && any (isContentElem . elName)
+                                                         (elChildren el)
+                            _ -> False)
+    isContentElem n = qName n `elem`
+      ["t", "drawing", "pict", "br", "tab", "sym", "fldChar", "instrText"]
     -- Search textbox content in the run's effective children.
     -- If AlternateContent is present, use only the fallback branch,
     -- matching the w:r unwrapping logic and avoiding duplicate textbox
@@ -475,9 +497,17 @@ archiveToDocument zf = do
   return $ Document namespaces body
 
 elemToBody :: NameSpaces -> Element -> D Body
-elemToBody ns element | isElem ns "w" "body" element =
-  Body . addCaptioned <$> mapD (elemToBodyPart ns) (elChildren element)
-elemToBody _ _ = throwError WrongElem
+elemToBody ns element
+  | isElem ns "w" "body" element =
+      Body . addCaptioned <$> concatMapM (elemToBodyParts ns) (elChildren element)
+  | otherwise = throwError WrongElem
+  where
+    elemToBodyParts ns' el
+      | qName (elName el) == "textbox-image"
+      = addSplitGroupCaptioned . splitMixedParagraphs
+          <$> mapD (elemToBodyPart ns') (elChildren el)
+      | otherwise
+      = ((:[]) <$> elemToBodyPart ns' el) `catchError` (\_ -> return [])
 
 archiveToStyles :: Archive -> (CharStyleMap, ParStyleMap)
 archiveToStyles = archiveToStyles' getStyleId getStyleId
@@ -803,6 +833,28 @@ addCaptioned (x : Paragraph parstyle parparts : xs)
     = Captioned parstyle parparts x : addCaptioned xs
 addCaptioned (x:xs) = x : addCaptioned xs
 
+-- Split paragraphs that contain both a Drawing and text into two body parts,
+-- so that the captioning logic can combine them.
+splitMixedParagraphs :: [BodyPart] -> [BodyPart]
+splitMixedParagraphs [] = []
+splitMixedParagraphs (Paragraph ps parts : rest)
+  | (drawings@(_:_), texts@(_:_)) <- partition isDrawingPart parts
+  = Paragraph ps drawings : Paragraph ps texts : splitMixedParagraphs rest
+  where isDrawingPart Drawing{} = True
+        isDrawingPart _         = False
+splitMixedParagraphs (x:xs) = x : splitMixedParagraphs xs
+
+-- For textbox-image groups: combine image + any adjacent text into Captioned
+addSplitGroupCaptioned :: [BodyPart] -> [BodyPart]
+addSplitGroupCaptioned [] = []
+addSplitGroupCaptioned (Paragraph parstyle parparts : x : xs)
+  | isCaptionable x
+  = Captioned parstyle parparts x : addSplitGroupCaptioned xs
+addSplitGroupCaptioned (x : Paragraph parstyle parparts : xs)
+  | isCaptionable x
+  = Captioned parstyle parparts x : addSplitGroupCaptioned xs
+addSplitGroupCaptioned (x:xs) = x : addSplitGroupCaptioned xs
+
 isCaptionable :: BodyPart -> Bool
 isCaptionable (Paragraph _ [Drawing{}]) = True
 isCaptionable (Tbl{}) = True

test/Tests/Readers/Docx.hs

@@ -223,6 +223,10 @@ tests = [ testGroup "document"
             "image inside textbox content with duplicate encoding"
             "docx/textbox_image_duplicate_encoding.docx"
             "docx/textbox_image_duplicate_encoding.native"
+          , testCompare
+            "image with textbox caption in same paragraph"
+            "docx/image_with_textbox_caption.docx"
+            "docx/image_with_textbox_caption.native"
           ]
         , testGroup "blocks"
           [ testCompare
@@ -487,6 +491,9 @@ tests = [ testGroup "document"
           , testMediaBag
             "image inside textbox content with duplicate encoding populates media bag"
             "docx/textbox_image_duplicate_encoding.docx"
+          , testMediaBag
+            "image with textbox caption in same paragraph populates media bag"
+            "docx/image_with_textbox_caption.docx"
           ]
         , testGroup "custom styles"
           [ testCompare

test/docx/image_with_textbox_caption.native

@@ -0,0 +1,44 @@
+[ Figure
+    ( "" , [] , [] )
+    (Caption
+       Nothing
+       [ Para
+           [ Str "1"
+           , Space
+           , Str "Daniel"
+           , Space
+           , Str "Schliebner:"
+           , Space
+           , Str "Cantor'sches"
+           , Space
+           , Str "Diagonalverfahren."
+           , Space
+           , Str "Von"
+           , Space
+           , Str "Mengen,"
+           , Space
+           , Str "Unendlichkeiten"
+           , Space
+           , Str "und"
+           , Space
+           , Str "Wahnsinn,"
+           , Space
+           , Str "Pdf:"
+           , Space
+           , Str
+               "https://www2.informatik.hu-berlin.de/~kossahl/Uni/Ma1/Cantor.pdf"
+           ]
+       ])
+    [ Plain
+        [ Image
+            ( ""
+            , []
+            , [ ( "width" , "2.3680555555555554in" )
+              , ( "height" , "0.9340277777777778in" )
+              ]
+            )
+            []
+            ( "media/image1.emf" , "" )
+        ]
+    ]
+]