from __future__ import annotations import logging import re import warnings from copy import deepcopy from enum import Enum from html import unescape from io import BytesIO from pathlib import Path from typing import Literal, Optional, Union, cast from docling_core.types.doc import ( DocItemLabel, DoclingDocument, DocumentOrigin, Formatting, ImageRef, ListItem, NodeItem, TableCell, TableData, TextItem, ) from pydantic import AnyUrl, BaseModel, Field, TypeAdapter from typing_extensions import Annotated, override from docling.backend.abstract_backend import ( DeclarativeDocumentBackend, ) from docling.backend.html_backend import HTMLDocumentBackend from docling.backend.utils.image_resource_loader import ImageResourceLoader from docling.datamodel.backend_options import ( HTMLBackendOptions, MarkdownBackendOptions, ) from docling.datamodel.base_models import InputFormat from docling.datamodel.document import InputDocument from docling.exceptions import DocumentLoadError # marko is only installed by the `format-markdown` extra, but DocumentConverter # imports every backend eagerly. Importing it at module load would therefore # break `import docling` on installs that omit the extra (the slim packages in # particular). Guard the imports like the opendocument and xbrl backends do, and # surface the failure only when Markdown is actually parsed. # See https://github.com/docling-project/docling/issues/3613. _MARKO_AVAILABLE: bool = False _MARKO_IMPORT_ERROR: ImportError | None = None try: # pragma: no cover - import-time guard import marko import marko.element import marko.inline from marko import Markdown _MARKO_AVAILABLE = True except ImportError as e: # pragma: no cover - import-time guard _MARKO_IMPORT_ERROR = e _log = logging.getLogger(__name__) _INSTALL_HINT = ( "The 'marko' package is required to process Markdown files. " "Install it with `pip install 'docling[format-markdown]'`." ) _MARKER_BODY = "DOCLING_DOC_MD_HTML_EXPORT" _START_MARKER = f"#_#_{_MARKER_BODY}_START_#_#" _STOP_MARKER = f"#_#_{_MARKER_BODY}_STOP_#_#" class _PendingCreationType(str, Enum): """CoordOrigin.""" HEADING = "heading" LIST_ITEM = "list_item" class _HeadingCreationPayload(BaseModel): kind: Literal["heading"] = "heading" level: int class _ListItemCreationPayload(BaseModel): kind: Literal["list_item"] = "list_item" enumerated: bool _CreationPayload = Annotated[ Union[ _HeadingCreationPayload, _ListItemCreationPayload, ], Field(discriminator="kind"), ] class MarkdownDocumentBackend(DeclarativeDocumentBackend): _ENTITY_RE = re.compile(r"&(#\d+|#x[0-9a-fA-F]+|\w+);") @staticmethod def _unescape_except_pipe(text: str) -> str: def replace(match): entity = match.group(0) # entities that represent | if entity in ("|", "|", "|"): return entity return unescape(entity) return MarkdownDocumentBackend._ENTITY_RE.sub(replace, text) def _shorten_underscore_sequences(self, markdown_text: str, max_length: int = 10): # This regex will match any sequence of underscores pattern = r"_+" def replace_match(match): underscore_sequence = match.group( 0 ) # Get the full match (sequence of underscores) # Shorten the sequence if it exceeds max_length if len(underscore_sequence) > max_length: return "_" * max_length else: return underscore_sequence # Leave it unchanged if it is shorter or equal to max_length # Use re.sub to replace long underscore sequences shortened_text = re.sub(pattern, replace_match, markdown_text) if len(shortened_text) != len(markdown_text): warnings.warn("Detected potentially incorrect Markdown, correcting...") return shortened_text def _shorten_leading_dash_sequences( self, markdown_text: str, max_length: int = 10 ) -> str: pattern = re.compile( rf"^([ \t]*)(?:-\s+){{{max_length + 1},}}-?(?=\S)", re.MULTILINE ) shortened_text, count = pattern.subn(r"\1- ", markdown_text) if count > 0: warnings.warn("Detected potentially incorrect Markdown, correcting...") return shortened_text @override def __init__( self, in_doc: InputDocument, path_or_stream: Union[BytesIO, Path], options: Optional[MarkdownBackendOptions] = None, ): # Raised first so a missing optional dependency gives an actionable # message rather than a NameError when marko is dereferenced below. if not _MARKO_AVAILABLE: raise ImportError(_INSTALL_HINT) from _MARKO_IMPORT_ERROR if options is None: options = MarkdownBackendOptions() super().__init__(in_doc, path_or_stream, options) _log.debug("Starting MarkdownDocumentBackend...") # Markdown file: self.path_or_stream = path_or_stream self.valid = True self.markdown = "" # To store original Markdown string self.in_table = False self.md_table_buffer: list[str] = [] self._html_blocks: int = 0 self._image_loader: Optional[ImageResourceLoader] = None try: if isinstance(self.path_or_stream, BytesIO): text_stream = self.path_or_stream.getvalue().decode("utf-8") # remove invalid sequences # very long sequences of underscores will lead to unnecessary long processing times. # In any proper Markdown files, underscores have to be escaped, # otherwise they represent emphasis (bold or italic) self.markdown = self._shorten_underscore_sequences(text_stream) self.markdown = self._shorten_leading_dash_sequences(self.markdown) if isinstance(self.path_or_stream, Path): with open(self.path_or_stream, encoding="utf-8") as f: md_content = f.read() # remove invalid sequences # very long sequences of underscores will lead to unnecessary long processing times. # In any proper Markdown files, underscores have to be escaped, # otherwise they represent emphasis (bold or italic) self.markdown = self._shorten_underscore_sequences(md_content) self.markdown = self._shorten_leading_dash_sequences(self.markdown) self.valid = True _log.debug(self.markdown) except Exception as e: raise DocumentLoadError( f"Could not initialize MD backend for file with hash {self.document_hash}." ) from e return def _close_table(self, doc: DoclingDocument): if self.in_table: _log.debug("=== TABLE START ===") for md_table_row in self.md_table_buffer: _log.debug(md_table_row) _log.debug("=== TABLE END ===") tcells: list[TableCell] = [] result_table = [] for n, md_table_row in enumerate(self.md_table_buffer): data = [] if n == 0: header = [t.strip() for t in md_table_row.split("|")[1:-1]] for value in header: data.append(value) result_table.append(data) if n > 1: values = [t.strip() for t in md_table_row.split("|")[1:-1]] for value in values: data.append(value) result_table.append(data) for trow_ind, trow in enumerate(result_table): for tcol_ind, cellval in enumerate(trow): row_span = ( 1 # currently supporting just simple tables (without spans) ) col_span = ( 1 # currently supporting just simple tables (without spans) ) icell = TableCell( text=unescape(cellval.strip()), row_span=row_span, col_span=col_span, start_row_offset_idx=trow_ind, end_row_offset_idx=trow_ind + row_span, start_col_offset_idx=tcol_ind, end_col_offset_idx=tcol_ind + col_span, column_header=trow_ind == 0, row_header=False, ) tcells.append(icell) num_rows = len(result_table) num_cols = len(result_table[0]) self.in_table = False self.md_table_buffer = [] # clean table markdown buffer # Initialize Docling TableData table_data = TableData( num_rows=num_rows, num_cols=num_cols, table_cells=tcells ) # Populate for tcell in tcells: table_data.table_cells.append(tcell) if len(tcells) > 0: doc.add_table(data=table_data) return def _create_list_item( self, doc: DoclingDocument, parent_item: Optional[NodeItem], text: str, enumerated: bool, formatting: Optional[Formatting] = None, hyperlink: Optional[Union[AnyUrl, Path]] = None, ): item = doc.add_list_item( text=text, enumerated=enumerated, parent=parent_item, formatting=formatting, hyperlink=hyperlink, ) return item def _create_heading_item( self, doc: DoclingDocument, parent_item: Optional[NodeItem], text: str, level: int, formatting: Optional[Formatting] = None, hyperlink: Optional[Union[AnyUrl, Path]] = None, ): if level == 1: item = doc.add_title( text=text, parent=parent_item, formatting=formatting, hyperlink=hyperlink, ) else: item = doc.add_heading( text=text, level=level - 1, parent=parent_item, formatting=formatting, hyperlink=hyperlink, ) return item def _flush_creation_stack( self, *, doc: DoclingDocument, creation_stack: list[_CreationPayload], snippet_text: str, parent_item: Optional[NodeItem], list_ordered_flag_by_ref: dict[str, bool], list_last_item_by_ref: dict[str, ListItem], formatting: Optional[Formatting], hyperlink: Optional[Union[AnyUrl, Path]], ) -> Optional[NodeItem]: """ Lazily create list items / headings when we first see their inline content. Important: Marko list items/headings can contain inline nodes that are NOT RawText (e.g. CodeSpan, Link). If we only flush on RawText, pending payloads can leak to later nodes and attach to a wrong parent, producing a very deep tree. """ while len(creation_stack) > 0: to_create = creation_stack.pop() if isinstance(to_create, _ListItemCreationPayload): enumerated = ( list_ordered_flag_by_ref.get(parent_item.self_ref, False) if parent_item else False ) parent_ref = parent_item.self_ref if parent_item else None parent_item = self._create_list_item( doc=doc, parent_item=parent_item, text=snippet_text, enumerated=enumerated, formatting=formatting, hyperlink=hyperlink, ) if parent_ref: list_last_item_by_ref[parent_ref] = cast(ListItem, parent_item) elif isinstance(to_create, _HeadingCreationPayload): # Not keeping as parent_item as logic for correctly tracking # that not implemented yet (section components not captured # as heading children in marko) self._create_heading_item( doc=doc, parent_item=parent_item, text=snippet_text, level=to_create.level, formatting=formatting, hyperlink=hyperlink, ) return parent_item def _iterate_elements( # noqa: C901 self, *, element: marko.element.Element, depth: int, doc: DoclingDocument, visited: set[marko.element.Element], creation_stack: list[ _CreationPayload ], # stack for lazy item creation triggered deep in marko's AST (on RawText) list_ordered_flag_by_ref: dict[str, bool], list_last_item_by_ref: dict[str, ListItem], parent_item: Optional[NodeItem] = None, formatting: Optional[Formatting] = None, hyperlink: Optional[Union[AnyUrl, Path]] = None, ): if element in visited: return # Iterates over all elements in the AST # Check for different element types and process relevant details if ( isinstance(element, marko.block.Heading) or isinstance(element, marko.block.SetextHeading) ) and len(element.children) > 0: self._close_table(doc) _log.debug( " - Heading level %s, content: %s", element.level, element.children[0].children, # type: ignore ) if len(element.children) > 1: # inline group will be created further down parent_item = self._create_heading_item( doc=doc, parent_item=parent_item, text="", level=element.level, formatting=formatting, hyperlink=hyperlink, ) else: creation_stack.append(_HeadingCreationPayload(level=element.level)) elif isinstance(element, marko.block.List): has_non_empty_list_items = False for child in element.children: if isinstance(child, marko.block.ListItem) and len(child.children) > 0: has_non_empty_list_items = True break self._close_table(doc) _log.debug(" - List %s", "ordered" if element.ordered else "unordered") if has_non_empty_list_items: parent_item = doc.add_list_group(name="list", parent=parent_item) list_ordered_flag_by_ref[parent_item.self_ref] = element.ordered elif ( isinstance(element, marko.block.ListItem) and len(element.children) > 0 and isinstance((child := element.children[0]), marko.block.Paragraph) and len(child.children) > 0 ): self._close_table(doc) _log.debug(" - List item") enumerated = ( list_ordered_flag_by_ref.get(parent_item.self_ref, False) if parent_item else False ) non_list_children: list[marko.element.Element] = [ item for item in child.children if not isinstance(item, marko.block.ListItem) ] if len(non_list_children) > 1: # inline group will be created further down parent_ref: Optional[str] = ( parent_item.self_ref if parent_item else None ) parent_item = self._create_list_item( doc=doc, parent_item=parent_item, text="", enumerated=enumerated, formatting=formatting, hyperlink=hyperlink, ) if parent_ref: list_last_item_by_ref[parent_ref] = cast(ListItem, parent_item) else: creation_stack.append(_ListItemCreationPayload(enumerated=enumerated)) elif isinstance(element, marko.inline.Image): self._close_table(doc) _log.debug(" - Image with alt: %s, url: %s", element.title, element.dest) fig_caption: Optional[TextItem] = None if element.title is not None and element.title != "": title = unescape(element.title) fig_caption = doc.add_text( label=DocItemLabel.CAPTION, text=title, formatting=formatting, hyperlink=hyperlink, ) image_ref = self._load_image_ref(element.dest) doc.add_picture(parent=parent_item, image=image_ref, caption=fig_caption) elif isinstance(element, marko.inline.Emphasis): _log.debug(" - Emphasis: %s", element.children) formatting = deepcopy(formatting) if formatting else Formatting() formatting.italic = True elif isinstance(element, marko.inline.StrongEmphasis): _log.debug(" - StrongEmphasis: %s", element.children) formatting = deepcopy(formatting) if formatting else Formatting() formatting.bold = True elif isinstance(element, marko.inline.Link): _log.debug(" - Link: %s", element.children) hyperlink = TypeAdapter(Optional[Union[AnyUrl, Path]]).validate_python( element.dest ) elif isinstance(element, marko.inline.RawText | marko.inline.Literal): _log.debug(" - RawText/Literal: %s", element.children) original_text = ( element.children if isinstance(element.children, str) else "" ) snippet_text = unescape(original_text.strip()) is_table_row = "|" in snippet_text and ( self.in_table or original_text.lstrip().startswith("|") ) if is_table_row: self.in_table = True if self.in_table and snippet_text: snippet_text = self._unescape_except_pipe(original_text.strip()) # If we're in a table, keep adding text (for formatted content in cells) if self.md_table_buffer: self.md_table_buffer[len(self.md_table_buffer) - 1] += snippet_text else: self.md_table_buffer.append(snippet_text) elif snippet_text: # Not in table - close any pending table and process as regular text self._close_table(doc) if creation_stack: parent_item = self._flush_creation_stack( doc=doc, creation_stack=creation_stack, snippet_text=snippet_text, parent_item=parent_item, list_ordered_flag_by_ref=list_ordered_flag_by_ref, list_last_item_by_ref=list_last_item_by_ref, formatting=formatting, hyperlink=hyperlink, ) else: doc.add_text( label=DocItemLabel.TEXT, parent=parent_item, text=snippet_text, formatting=formatting, hyperlink=hyperlink, ) elif isinstance(element, marko.inline.CodeSpan): self._close_table(doc) _log.debug(" - Code Span: %s", element.children) snippet_text = str(element.children).strip() # If this CodeSpan is the only content of a list item / heading, Marko won't # emit RawText. Flush pending creations here to avoid leaking payloads. if creation_stack and snippet_text: parent_item = self._flush_creation_stack( doc=doc, creation_stack=creation_stack, snippet_text=snippet_text, parent_item=parent_item, list_ordered_flag_by_ref=list_ordered_flag_by_ref, list_last_item_by_ref=list_last_item_by_ref, formatting=formatting, hyperlink=hyperlink, ) # Represent CodeSpan as the container's text; avoid adding a duplicate CodeItem. return doc.add_code( parent=parent_item, text=snippet_text, formatting=formatting, hyperlink=hyperlink, ) elif ( isinstance(element, marko.block.CodeBlock | marko.block.FencedCode) and len(element.children) > 0 and isinstance((child := element.children[0]), marko.inline.RawText) and len(snippet_text := (child.children.strip())) > 0 ): self._close_table(doc) _log.debug(" - Code Block: %s", element.children) doc.add_code( parent=parent_item, text=snippet_text, formatting=formatting, hyperlink=hyperlink, ) elif isinstance(element, marko.inline.LineBreak): if self.in_table: _log.debug("Line break in a table") self.md_table_buffer.append("") elif isinstance(element, marko.block.HTMLBlock): self._html_blocks += 1 self._close_table(doc) _log.debug("HTML Block: %s", element) if ( len(element.body) > 0 ): # If Marko doesn't return any content for HTML block, skip it html_block = element.body.strip() # wrap in markers to enable post-processing in convert() text_to_add = f"{_START_MARKER}{html_block}{_STOP_MARKER}" doc.add_code( parent=parent_item, text=text_to_add, formatting=formatting, hyperlink=hyperlink, ) else: if not isinstance(element, str): self._close_table(doc) _log.debug("Some other element: %s", type(element).__name__) if ( isinstance(element, marko.block.Paragraph | marko.block.Heading) and len(element.children) > 1 ): parent_item = doc.add_inline_group(parent=parent_item) processed_block_types = ( marko.block.CodeBlock, marko.block.FencedCode, marko.inline.RawText, ) # Iterate through the element's children (if any) if hasattr(element, "children") and not isinstance( element, processed_block_types ): for child in element.children: if ( isinstance(element, marko.block.ListItem) and isinstance(child, marko.block.List) and parent_item and list_last_item_by_ref.get(parent_item.self_ref, None) ): _log.debug( "walking into new List hanging from item of parent list %s", parent_item.self_ref, ) parent_item = list_last_item_by_ref[parent_item.self_ref] self._iterate_elements( element=child, depth=depth + 1, doc=doc, visited=visited, creation_stack=creation_stack, list_ordered_flag_by_ref=list_ordered_flag_by_ref, list_last_item_by_ref=list_last_item_by_ref, parent_item=parent_item, formatting=formatting, hyperlink=hyperlink, ) def _get_image_loader(self) -> ImageResourceLoader: """Lazily build the shared image-resource loader. Resolving and decoding image sources (``data:`` URIs, local files, and remote URLs) together with the relevant safety limits is shared with the HTML backend through :class:`ImageResourceLoader`, so that logic is not duplicated here. """ if self._image_loader is None: md_options = cast(MarkdownBackendOptions, self.options) self._image_loader = ImageResourceLoader( enable_local_fetch=md_options.enable_local_fetch, enable_remote_fetch=md_options.enable_remote_fetch, max_image_data_base64_bytes=md_options.max_image_data_base64_bytes, ) return self._image_loader def _load_image_ref(self, dest: str) -> Optional[ImageRef]: """Resolve and decode a Markdown image source into an ``ImageRef``. Returns ``None`` when image loading is disabled, the source is empty, or the image cannot be loaded. """ md_options = cast(MarkdownBackendOptions, self.options) if not md_options.fetch_images or not dest: return None base_path = ( str(md_options.source_uri) if md_options.source_uri is not None else None ) return self._get_image_loader().load_image_ref(dest, base_path) def is_valid(self) -> bool: return self.valid def unload(self): if isinstance(self.path_or_stream, BytesIO): self.path_or_stream.close() self.path_or_stream = None @classmethod def supports_pagination(cls) -> bool: return False @classmethod def supported_formats(cls) -> set[InputFormat]: return {InputFormat.MD} def convert(self) -> DoclingDocument: _log.debug("converting Markdown...") origin = DocumentOrigin( filename=self.file.name or "file", mimetype="text/markdown", binary_hash=self.document_hash, ) doc = DoclingDocument(name=self.file.stem or "file", origin=origin) if self.is_valid(): # Parse the markdown into an abstract syntax tree (AST) marko_parser = Markdown() parsed_ast = marko_parser.parse(self.markdown) # Start iterating from the root of the AST self._iterate_elements( element=parsed_ast, depth=0, doc=doc, parent_item=None, visited=set(), creation_stack=[], list_ordered_flag_by_ref={}, list_last_item_by_ref={}, ) self._close_table(doc=doc) # handle any last hanging table # if HTML blocks were detected, export to HTML and delegate to HTML backend if self._html_blocks > 0: # export to HTML html_backend_cls = HTMLDocumentBackend html_str = doc.export_to_html() def _restore_original_html(txt, regex): _txt, count = re.subn(regex, "", txt) if count != self._html_blocks: raise RuntimeError( "An internal error has occurred during Markdown conversion." ) return _txt # restore original HTML by removing previously added markers for regex in [ rf"
\s*\s*{_START_MARKER}",
                    rf"{_STOP_MARKER}\s*\s*
", ]: html_str = _restore_original_html(txt=html_str, regex=regex) self._html_blocks = 0 # delegate to HTML backend stream = BytesIO(bytes(html_str, encoding="utf-8")) md_options = cast(MarkdownBackendOptions, self.options) html_options = HTMLBackendOptions( enable_local_fetch=md_options.enable_local_fetch, enable_remote_fetch=md_options.enable_remote_fetch, fetch_images=md_options.fetch_images, source_uri=md_options.source_uri, infer_furniture=False, add_title=False, ) in_doc = InputDocument( path_or_stream=stream, format=InputFormat.HTML, backend=html_backend_cls, filename=self.file.name, backend_options=html_options, ) html_backend_obj = html_backend_cls( in_doc=in_doc, path_or_stream=stream, options=html_options, ) doc = html_backend_obj.convert() else: raise RuntimeError( f"Cannot convert md with {self.document_hash} because the backend failed to init." ) return doc