From 5948305ce1d302fb9be6144e1c690e08cbfc28fc Mon Sep 17 00:00:00 2001
From: Eapp1e <163811520+Eapp1e@users.noreply.github.com>
Date: Wed, 28 Oct 2026 05:06:33 +0000
Subject: [PATCH] fix(asciidoc): open blocks on four or more delimiters

AsciiDoc treats the delimiter length as a minimum: listing blocks open
with four or more hyphens and literal blocks with four or more dots.
The backend only accepted exactly four, so a block opened with five or
more delimiters was parsed as a paragraph and its content leaked into
running text.

Match four or more and keep the real delimiter line so the closing
delimiter is still found.

Signed-off-by: Eapp1e <163811520+Eapp1e@users.noreply.github.com>
---
 docling/backend/asciidoc_backend.py | 18 ++++++++++++++----
 tests/test_backend_asciidoc.py      | 21 +++++++++++++++++++++
 2 files changed, 35 insertions(+), 4 deletions(-)

diff --git a/docling/backend/asciidoc_backend.py b/docling/backend/asciidoc_backend.py
index 63c984b44b..130a7c7ddb 100644
--- a/docling/backend/asciidoc_backend.py
+++ b/docling/backend/asciidoc_backend.py
@@ -41,6 +41,12 @@
 # to an identical line is a comment.
 _LINE_COMMENT_RE = re.compile(r"^//(?!/)")
 _COMMENT_BLOCK_DELIMITER_RE = re.compile(r"/{4,}")
+# A line of four or more hyphens opens a listing block and a line of four or
+# more dots opens a literal block: the count is a minimum, so "-----" opens a
+# listing just as "----" does.
+_VERBATIM_DELIMITER_RE = re.compile(r"-{4,}|\.{4,}")
+# A listing block takes a "[source,lang]" attribute; a literal block does not.
+_LISTING_DELIMITER_RE = re.compile(r"-{4,}")
 
 # Cell format specifier that may precede a "|" delimiter, e.g. "^.^h" in
 # "^.^h|Header": span (3*, 2+, .2+, 2.3+), alignment (<, ^, >, .^), style
@@ -567,9 +573,9 @@ def _iter_blocks(
                     yield comment
                 continue
 
-            if stripped in {"....", "----"}:
+            if _VERBATIM_DELIMITER_RE.fullmatch(stripped):
                 # "...." is a literal block; "----" a listing block. A
-                # "[source,lang]" attribute consumed just before a "----"
+                # "[source,lang]" attribute consumed just before a listing
                 # labels the listing's language.
                 block_data = []
                 block_delimiter = stripped
@@ -597,9 +603,13 @@ def _iter_blocks(
                 continue
 
             source_attr = cls._SOURCE_ATTR_RE.match(stripped)
-            if source_attr is not None and i + 1 < n and lines[i + 1].strip() == "----":
+            if (
+                source_attr is not None
+                and i + 1 < n
+                and _LISTING_DELIMITER_RE.fullmatch(lines[i + 1].strip())
+            ):
                 block_data = []
-                block_delimiter = "----"
+                block_delimiter = lines[i + 1].strip()
                 block_language = source_attr.group(1)
                 i += 2
                 continue
diff --git a/tests/test_backend_asciidoc.py b/tests/test_backend_asciidoc.py
index 3b227cf375..302ae3b344 100644
--- a/tests/test_backend_asciidoc.py
+++ b/tests/test_backend_asciidoc.py
@@ -247,6 +247,27 @@ def test_source_listing_block_becomes_code_item() -> None:
     assert [item.text for item in doc.texts if item.label != DocItemLabel.CODE] == []
 
 
+def test_delimiter_longer_than_four_characters_opens_the_block() -> None:
+    # A listing block opens on four or more hyphens and a literal block on four
+    # or more dots, so the marker of a valid block is not always four characters
+    # long. Only "----" and "...." opened a block, so the markers of a longer
+    # block stayed in the document as paragraphs and the code became prose.
+    src = b"[source,python]\n-----\nx = 1\n-----\n\n.....\nraw\n.....\n"
+    in_doc = InputDocument(
+        path_or_stream=BytesIO(src),
+        format=InputFormat.ASCIIDOC,
+        backend=AsciiDocBackend,
+        filename="long-delimiter.adoc",
+    )
+    doc = in_doc._backend.convert()
+
+    code_items = [item for item in doc.texts if item.label == DocItemLabel.CODE]
+    assert [item.text for item in code_items] == ["x = 1", "raw"]
+    assert code_items[0].code_language == CodeLanguageLabel.PYTHON
+    # no marker or attribute text may leak into the body
+    assert [item.text for item in doc.texts if item.label != DocItemLabel.CODE] == []
+
+
 def test_listing_block_without_language_stays_code() -> None:
     src = b"----\nplain listing\n----\n"
     in_doc = InputDocument(