EPub Support. Adapted #123 to not use epublib. (#1131)

* Adapted #123 to not use epublib.
* Updated README.md

README.md

@@ -14,7 +14,7 @@ MarkItDown is a lightweight Python utility for converting various files to Markd
 At present, MarkItDown supports:
 
 - PDF
-- PowerPoint (reading in top-to-bottom, left-to-right order)
+- PowerPoint
 - Word
 - Excel
 - Images (EXIF metadata and OCR)
@@ -23,6 +23,7 @@ At present, MarkItDown supports:
 - Text-based formats (CSV, JSON, XML)
 - ZIP files (iterates over contents)
 - Youtube URLs
+- EPubs
 - ... and more!
 
 ## Why Markdown?

packages/markitdown/pyproject.toml

@@ -27,7 +27,7 @@ dependencies = [
   "beautifulsoup4",
   "requests",
   "markdownify",
-  "magika>=0.6.0rc1",
+  "magika>=0.6.1rc3",
   "charset-normalizer",
 ]
 

packages/markitdown/src/markitdown/__about__.py

@@ -1,4 +1,4 @@
 # SPDX-FileCopyrightText: 2024-present Adam Fourney <adamfo@microsoft.com>
 #
 # SPDX-License-Identifier: MIT
-__version__ = "0.1.0a1"
+__version__ = "0.1.0a4"

packages/markitdown/src/markitdown/__main__.py

@@ -139,7 +139,7 @@ def main():
         else:
             charset_hint = None
 
-    stream_info: str | None = None
+    stream_info = None
     if (
         extension_hint is not None
         or mime_type_hint is not None

packages/markitdown/src/markitdown/_markitdown.py

@@ -38,6 +38,7 @@ from .converters import (
     AudioConverter,
     OutlookMsgConverter,
     ZipConverter,
+    EpubConverter,
     DocumentIntelligenceConverter,
 )
 
@@ -191,6 +192,7 @@ class MarkItDown:
             self.register_converter(IpynbConverter())
             self.register_converter(PdfConverter())
             self.register_converter(OutlookMsgConverter())
+            self.register_converter(EpubConverter())
 
             # Register Document Intelligence converter at the top of the stack if endpoint is provided
             docintel_endpoint = kwargs.get("docintel_endpoint")
@@ -610,14 +612,16 @@ class MarkItDown:
         # Call magika to guess from the stream
         cur_pos = file_stream.tell()
         try:
-            stream_bytes = file_stream.read()
-
-            result = self._magika.identify_bytes(stream_bytes)
+            result = self._magika.identify_stream(file_stream)
             if result.status == "ok" and result.prediction.output.label != "unknown":
                 # If it's text, also guess the charset
                 charset = None
                 if result.prediction.output.is_text:
-                    charset_result = charset_normalizer.from_bytes(stream_bytes).best()
+                    # Read the first 4k to guess the charset
+                    file_stream.seek(cur_pos)
+                    stream_page = file_stream.read(4096)
+                    charset_result = charset_normalizer.from_bytes(stream_page).best()
+
                     if charset_result is not None:
                         charset = self._normalize_charset(charset_result.encoding)
 

packages/markitdown/src/markitdown/converters/__init__.py

@@ -18,6 +18,7 @@ from ._audio_converter import AudioConverter
 from ._outlook_msg_converter import OutlookMsgConverter
 from ._zip_converter import ZipConverter
 from ._doc_intel_converter import DocumentIntelligenceConverter
+from ._epub_converter import EpubConverter
 
 __all__ = [
     "PlainTextConverter",
@@ -37,4 +38,5 @@ __all__ = [
     "OutlookMsgConverter",
     "ZipConverter",
     "DocumentIntelligenceConverter",
+    "EpubConverter",
 ]

packages/markitdown/src/markitdown/converters/_bing_serp_converter.py

@@ -1,6 +1,7 @@
 import io
 import re
 import base64
+import binascii
 from urllib.parse import parse_qs, urlparse
 from typing import Any, BinaryIO, Optional
 from bs4 import BeautifulSoup
@@ -60,6 +61,8 @@ class BingSerpConverter(DocumentConverter):
         stream_info: StreamInfo,
         **kwargs: Any,  # Options to pass to the converter
     ) -> DocumentConverterResult:
+        assert stream_info.url is not None
+
         # Parse the query parameters
         parsed_params = parse_qs(urlparse(stream_info.url).query)
         query = parsed_params.get("q", [""])[0]
@@ -79,6 +82,9 @@ class BingSerpConverter(DocumentConverter):
         _markdownify = _CustomMarkdownify()
         results = list()
         for result in soup.find_all(class_="b_algo"):
+            if not hasattr(result, "find_all"):
+                continue
+
             # Rewrite redirect urls
             for a in result.find_all("a", href=True):
                 parsed_href = urlparse(a["href"])

packages/markitdown/src/markitdown/converters/_epub_converter.py

@@ -0,0 +1,147 @@
+import os
+import zipfile
+import xml.dom.minidom as minidom
+
+from typing import BinaryIO, Any, Dict, List
+
+from ._html_converter import HtmlConverter
+from .._base_converter import DocumentConverter, DocumentConverterResult
+from .._stream_info import StreamInfo
+
+ACCEPTED_MIME_TYPE_PREFIXES = [
+    "application/epub",
+    "application/epub+zip",
+    "application/x-epub+zip",
+]
+
+ACCEPTED_FILE_EXTENSIONS = [".epub"]
+
+MIME_TYPE_MAPPING = {
+    ".html": "text/html",
+    ".xhtml": "application/xhtml+xml",
+}
+
+
+class EpubConverter(HtmlConverter):
+    """
+    Converts EPUB files to Markdown. Style information (e.g.m headings) and tables are preserved where possible.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self._html_converter = HtmlConverter()
+
+    def accepts(
+        self,
+        file_stream: BinaryIO,
+        stream_info: StreamInfo,
+        **kwargs: Any,  # Options to pass to the converter
+    ) -> bool:
+        mimetype = (stream_info.mimetype or "").lower()
+        extension = (stream_info.extension or "").lower()
+
+        if extension in ACCEPTED_FILE_EXTENSIONS:
+            return True
+
+        for prefix in ACCEPTED_MIME_TYPE_PREFIXES:
+            if mimetype.startswith(prefix):
+                return True
+
+        return False
+
+    def convert(
+        self,
+        file_stream: BinaryIO,
+        stream_info: StreamInfo,
+        **kwargs: Any,  # Options to pass to the converter
+    ) -> DocumentConverterResult:
+        with zipfile.ZipFile(file_stream, "r") as z:
+            # Extracts metadata (title, authors, language, publisher, date, description, cover) from an EPUB file."""
+
+            # Locate content.opf
+            container_dom = minidom.parse(z.open("META-INF/container.xml"))
+            opf_path = container_dom.getElementsByTagName("rootfile")[0].getAttribute(
+                "full-path"
+            )
+
+            # Parse content.opf
+            opf_dom = minidom.parse(z.open(opf_path))
+            metadata: Dict[str, Any] = {
+                "title": self._get_text_from_node(opf_dom, "dc:title"),
+                "authors": self._get_all_texts_from_nodes(opf_dom, "dc:creator"),
+                "language": self._get_text_from_node(opf_dom, "dc:language"),
+                "publisher": self._get_text_from_node(opf_dom, "dc:publisher"),
+                "date": self._get_text_from_node(opf_dom, "dc:date"),
+                "description": self._get_text_from_node(opf_dom, "dc:description"),
+                "identifier": self._get_text_from_node(opf_dom, "dc:identifier"),
+            }
+
+            # Extract manifest items (ID → href mapping)
+            manifest = {
+                item.getAttribute("id"): item.getAttribute("href")
+                for item in opf_dom.getElementsByTagName("item")
+            }
+
+            # Extract spine order (ID refs)
+            spine_items = opf_dom.getElementsByTagName("itemref")
+            spine_order = [item.getAttribute("idref") for item in spine_items]
+
+            # Convert spine order to actual file paths
+            base_path = "/".join(
+                opf_path.split("/")[:-1]
+            )  # Get base directory of content.opf
+            spine = [
+                f"{base_path}/{manifest[item_id]}" if base_path else manifest[item_id]
+                for item_id in spine_order
+                if item_id in manifest
+            ]
+
+            # Extract and convert the content
+            markdown_content: List[str] = []
+            for file in spine:
+                if file in z.namelist():
+                    with z.open(file) as f:
+                        filename = os.path.basename(file)
+                        extension = os.path.splitext(filename)[1].lower()
+                        mimetype = MIME_TYPE_MAPPING.get(extension)
+                        converted_content = self._html_converter.convert(
+                            f,
+                            StreamInfo(
+                                mimetype=mimetype,
+                                extension=extension,
+                                filename=filename,
+                            ),
+                        )
+                        markdown_content.append(converted_content.markdown.strip())
+
+            # Format and add the metadata
+            metadata_markdown = []
+            for key, value in metadata.items():
+                if isinstance(value, list):
+                    value = ", ".join(value)
+                if value:
+                    metadata_markdown.append(f"**{key.capitalize()}:** {value}")
+
+            markdown_content.insert(0, "\n".join(metadata_markdown))
+
+            return DocumentConverterResult(
+                markdown="\n\n".join(markdown_content), title=metadata["title"]
+            )
+
+    def _get_text_from_node(self, dom: minidom.Document, tag_name: str) -> str | None:
+        """Convenience function to extract a single occurrence of a tag (e.g., title)."""
+        texts = self._get_all_texts_from_nodes(dom, tag_name)
+        if len(texts) > 0:
+            return texts[0]
+        else:
+            return None
+
+    def _get_all_texts_from_nodes(
+        self, dom: minidom.Document, tag_name: str
+    ) -> List[str]:
+        """Helper function to extract all occurrences of a tag (e.g., multiple authors)."""
+        texts: List[str] = []
+        for node in dom.getElementsByTagName(tag_name):
+            if node.firstChild and hasattr(node.firstChild, "nodeValue"):
+                texts.append(node.firstChild.nodeValue.strip())
+        return texts

packages/markitdown/src/markitdown/converters/_outlook_msg_converter.py

@@ -9,7 +9,7 @@ from .._exceptions import MissingDependencyException, MISSING_DEPENDENCY_MESSAGE
 _dependency_exc_info = None
 olefile = None
 try:
-    import olefile
+    import olefile  # type: ignore[no-redef]
 except ImportError:
     # Preserve the error and stack trace for later
     _dependency_exc_info = sys.exc_info()
@@ -56,12 +56,13 @@ class OutlookMsgConverter(DocumentConverter):
 
         # Brue force, check if it's an Outlook file
         try:
-            msg = olefile.OleFileIO(file_stream)
-            toc = "\n".join([str(stream) for stream in msg.listdir()])
-            return (
-                "__properties_version1.0" in toc
-                and "__recip_version1.0_#00000000" in toc
-            )
+            if olefile is not None:
+                msg = olefile.OleFileIO(file_stream)
+                toc = "\n".join([str(stream) for stream in msg.listdir()])
+                return (
+                    "__properties_version1.0" in toc
+                    and "__recip_version1.0_#00000000" in toc
+                )
         except Exception as e:
             pass
         finally:
@@ -89,7 +90,11 @@ class OutlookMsgConverter(DocumentConverter):
                 _dependency_exc_info[2]
             )
 
+        assert (
+            olefile is not None
+        )  # If we made it this far, olefile should be available
         msg = olefile.OleFileIO(file_stream)
+
         # Extract email metadata
         md_content = "# Email Message\n\n"
 
@@ -121,6 +126,7 @@ class OutlookMsgConverter(DocumentConverter):
 
     def _get_stream_data(self, msg: Any, stream_path: str) -> Union[str, None]:
         """Helper to safely extract and decode stream data from the MSG file."""
+        assert olefile is not None
         assert isinstance(
             msg, olefile.OleFileIO
         )  # Ensure msg is of the correct type (type hinting is not possible with the optional olefile package)

packages/markitdown/src/markitdown/converters/_rss_converter.py

@@ -66,7 +66,7 @@ class RssConverter(DocumentConverter):
             file_stream.seek(cur_pos)
         return False
 
-    def _feed_type(self, doc: Any) -> str:
+    def _feed_type(self, doc: Any) -> str | None:
         if doc.getElementsByTagName("rss"):
             return "rss"
         elif doc.getElementsByTagName("feed"):
@@ -130,10 +130,10 @@ class RssConverter(DocumentConverter):
         Returns None if the feed type is not recognized or something goes wrong.
         """
         root = doc.getElementsByTagName("rss")[0]
-        channel = root.getElementsByTagName("channel")
-        if not channel:
-            return None
-        channel = channel[0]
+        channel_list = root.getElementsByTagName("channel")
+        if not channel_list:
+            raise ValueError("No channel found in RSS feed")
+        channel = channel_list[0]
         channel_title = self._get_data_by_tag_name(channel, "title")
         channel_description = self._get_data_by_tag_name(channel, "description")
         items = channel.getElementsByTagName("item")
@@ -141,8 +141,6 @@ class RssConverter(DocumentConverter):
             md_text = f"# {channel_title}\n"
         if channel_description:
             md_text += f"{channel_description}\n"
-        if not items:
-            items = []
         for item in items:
             title = self._get_data_by_tag_name(item, "title")
             description = self._get_data_by_tag_name(item, "description")
@@ -183,5 +181,6 @@ class RssConverter(DocumentConverter):
             return None
         fc = nodes[0].firstChild
         if fc:
-            return fc.data
+            if hasattr(fc, "data"):
+                return fc.data
         return None

packages/markitdown/src/markitdown/converters/_transcribe_audio.py

@@ -7,7 +7,19 @@ from .._exceptions import MissingDependencyException
 # Save reporting of any exceptions for later
 _dependency_exc_info = None
 try:
+    # Suppress some deprecation warnings from the speech_recognition library
+    import warnings
+
+    warnings.filterwarnings(
+        "ignore", category=DeprecationWarning, module="speech_recognition"
+    )
+    warnings.filterwarnings(
+        "ignore",
+        category=SyntaxWarning,
+        module="pydub",  # TODO: Migrate away from pydub
+    )
     import speech_recognition as sr
+
     import pydub
 except ImportError:
     # Preserve the error and stack trace for later

packages/markitdown/src/markitdown/converters/_wikipedia_converter.py

@@ -1,7 +1,7 @@
 import io
 import re
+import bs4
 from typing import Any, BinaryIO, Optional
-from bs4 import BeautifulSoup
 
 from .._base_converter import DocumentConverter, DocumentConverterResult
 from .._stream_info import StreamInfo
@@ -57,7 +57,7 @@ class WikipediaConverter(DocumentConverter):
     ) -> DocumentConverterResult:
         # Parse the stream
         encoding = "utf-8" if stream_info.charset is None else stream_info.charset
-        soup = BeautifulSoup(file_stream, "html.parser", from_encoding=encoding)
+        soup = bs4.BeautifulSoup(file_stream, "html.parser", from_encoding=encoding)
 
         # Remove javascript and style blocks
         for script in soup(["script", "style"]):
@@ -72,9 +72,8 @@ class WikipediaConverter(DocumentConverter):
 
         if body_elm:
             # What's the title
-            if title_elm and len(title_elm) > 0:
-                main_title = title_elm.string  # type: ignore
-                assert isinstance(main_title, str)
+            if title_elm and isinstance(title_elm, bs4.Tag):
+                main_title = title_elm.string
 
             # Convert the page
             webpage_text = f"# {main_title}\n\n" + _CustomMarkdownify().convert_soup(

packages/markitdown/src/markitdown/converters/_youtube_converter.py

@@ -3,9 +3,10 @@ import json
 import time
 import io
 import re
+import bs4
+import warnings
 from typing import Any, BinaryIO, Optional, Dict, List, Union
 from urllib.parse import parse_qs, urlparse, unquote
-from bs4 import BeautifulSoup
 
 from .._base_converter import DocumentConverter, DocumentConverterResult
 from .._stream_info import StreamInfo
@@ -13,6 +14,11 @@ from ._markdownify import _CustomMarkdownify
 
 # Optional YouTube transcription support
 try:
+    warnings.filterwarnings(
+        "ignore",
+        category=SyntaxWarning,
+        module="youtube_transcript_api",  # Patch submitted to youtube-transcript-api
+    )
     from youtube_transcript_api import YouTubeTranscriptApi
 
     IS_YOUTUBE_TRANSCRIPT_CAPABLE = True
@@ -72,21 +78,31 @@ class YouTubeConverter(DocumentConverter):
     ) -> DocumentConverterResult:
         # Parse the stream
         encoding = "utf-8" if stream_info.charset is None else stream_info.charset
-        soup = BeautifulSoup(file_stream, "html.parser", from_encoding=encoding)
+        soup = bs4.BeautifulSoup(file_stream, "html.parser", from_encoding=encoding)
 
         # Read the meta tags
-        metadata: Dict[str, str] = {"title": soup.title.string}
+        metadata: Dict[str, str] = {}
+
+        if soup.title and soup.title.string:
+            metadata["title"] = soup.title.string
+
         for meta in soup(["meta"]):
+            if not isinstance(meta, bs4.Tag):
+                continue
+
             for a in meta.attrs:
                 if a in ["itemprop", "property", "name"]:
-                    content = meta.get("content", "")
-                    if content:  # Only add non-empty content
-                        metadata[meta[a]] = content
+                    key = str(meta.get(a, ""))
+                    content = str(meta.get("content", ""))
+                    if key and content:  # Only add non-empty content
+                        metadata[key] = content
                     break
 
         # Try reading the description
         try:
             for script in soup(["script"]):
+                if not isinstance(script, bs4.Tag):
+                    continue
                 if not script.string:  # Skip empty scripts
                     continue
                 content = script.string
@@ -161,7 +177,7 @@ class YouTubeConverter(DocumentConverter):
             if transcript_text:
                 webpage_text += f"\n### Transcript\n{transcript_text}\n"
 
-        title = title if title else soup.title.string
+        title = title if title else (soup.title.string if soup.title else "")
         assert isinstance(title, str)
 
         return DocumentConverterResult(

packages/markitdown/tests/_test_vectors.py

@@ -211,4 +211,22 @@ GENERAL_TEST_VECTORS = [
         ],
         must_not_include=[],
     ),
+    FileTestVector(
+        filename="test.epub",
+        mimetype="application/epub+zip",
+        charset=None,
+        url=None,
+        must_include=[
+            "**Authors:** Test Author",
+            "A test EPUB document for MarkItDown testing",
+            "# Chapter 1: Test Content",
+            "This is a **test** paragraph with some formatting",
+            "* A bullet point",
+            "* Another point",
+            "# Chapter 2: More Content",
+            "*different* style",
+            "> This is a blockquote for testing",
+        ],
+        must_not_include=[],
+    ),
 ]

packages/markitdown/tests/test_cli_vectors.py

@@ -114,7 +114,9 @@ def test_input_from_stdin_without_hints(shared_tmp_dir, test_vector) -> None:
     )
 
     stdout = result.stdout.decode(locale.getpreferredencoding())
-    assert result.returncode == 0, f"CLI exited with error: {result.stderr}"
+    assert (
+        result.returncode == 0
+    ), f"CLI exited with error: {result.stderr.decode('utf-8')}"
     for test_string in test_vector.must_include:
         assert test_string in stdout
     for test_string in test_vector.must_not_include:

packages/markitdown/tests/test_module_vectors.py

@@ -47,7 +47,6 @@ def test_guess_stream_info(test_vector):
         # mimetype or extension, so we'll special-case them here.
         if test_vector.filename in [
             "test_outlook_msg.msg",
-            "test_mskanji.csv",  # See: https://github.com/google/magika/issues/983
         ]:
             return
 
@@ -96,15 +95,6 @@ def test_convert_stream_without_hints(test_vector):
     """Test the conversion of a stream with no stream info."""
     markitdown = MarkItDown()
 
-    # For some limited exceptions, we can't guarantee the exact
-    # mimetype or extension, so we'll special-case them here.
-    if test_vector.filename in [
-        # This appears to be a subtle bug in magika.
-        # See: https://github.com/google/magika/issues/983
-        "test_mskanji.csv",
-    ]:
-        return
-
     with open(os.path.join(TEST_FILES_DIR, test_vector.filename), "rb") as stream:
         result = markitdown.convert(stream, url=test_vector.url)
         for string in test_vector.must_include: