Added test for #2548 and address rebase fail.

Fixed test failure in rebased. We were not converting fz exception into C++
exception in src/extra.i:page_get_textpage(). Also fixed other cases where we
leaked fz exception.

src/extra.i

@@ -1659,7 +1659,7 @@ static pdf_obj *lll_JM_pdf_obj_from_str(fz_context *ctx, pdf_document *doc, cons
     }
 
     fz_catch(ctx) {
-        fz_rethrow(ctx);
+        mupdf::internal_throw_exception(ctx);
     }
 
     return result;
@@ -3744,7 +3744,7 @@ fz_stext_page* page_get_textpage(
         fz_drop_device(ctx, dev);
     }
     fz_catch(ctx) {
-        return NULL;
+        mupdf::internal_throw_exception(ctx);
     }
     return tpage;
 }
@@ -3965,7 +3965,7 @@ JM_new_buffer_from_stext_page(fz_stext_page *page)
     }
     fz_catch(ctx) {
         fz_drop_buffer(ctx, buf);
-        fz_rethrow(ctx);
+        mupdf::internal_throw_exception(ctx);
     }
     return buf;
 }
@@ -4209,7 +4209,7 @@ no_more_matches:;
     fz_always(ctx)
         fz_drop_buffer(ctx, buffer);
     fz_catch(ctx)
-        fz_rethrow(ctx);
+        mupdf::internal_throw_exception(ctx);
 
     return quads;
 }

tests/test_2548.py

@@ -0,0 +1,27 @@
+import os
+
+import fitz
+
+root = os.path.abspath(f'{__file__}/../..')
+
+def test_2548():
+    """Text extraction should fail because of PDF structure cycle.
+
+    Old MuPDF version did not detect the loop.
+    """
+    doc = fitz.open(f'{root}/tests/resources/test_2548.pdf')
+    e = False
+    for page in doc:
+        try:
+            _ = page.get_text()
+        except Exception as ee:
+            print(f'test_2548: {ee=}')
+            if hasattr(fitz, 'mupdf'):
+                # Rebased.
+                expected = "RuntimeError('code=2: cycle in structure tree')"
+            else:
+                # Classic.
+                expected = "RuntimeError('cycle in structure tree')"
+            assert repr(ee) == expected, f'Expected {expected=} but got {repr(ee)=}.'
+            e = True
+    assert e