plugin priority when multiple plugins accept a given filetype

[semoi69]

Hi,
I've made a plugin to convert docx, and extract images in a folder.
But, when calling markitdown, the plugin isn't called (it's well recognized by markitdown), because markitdown calls first the builtin docx converter.

How can we do to force the execution of a particular plugin, or define a priority, (...or other idea) ?
Or is the only solution to desactivate the builtin docxconverter, but how ?

Thanks a lot for your feedback.
Btw, great work @afourney

[afourney]

Awesome. Happy to hear that people are experimenting with plugins.

In the plugin's register_converters method, you should be able to do:

markitdown.register_converter(MyNewConverter(), priority=-1)

or

markitdown.register_converter(MyNewConverter(), priority=PRIORITY_GENERIC_FILE_FORMAT-1)

BUT, I plugins are always added after built-ins, so I'm guessing something else is going wrong.

Can you add code to make sure the register_converters method is being called?

Remember you need to use --use-plugins at the CLI or MarkItDown(enable_plugins=True) at in Python to make sure plugins get loaded. I didn't want to enable them all by automatically yet, as I thought that might have some security implications.

You can also try the sample:

pip install markitdown-sample-plugin

And then do:

markitdown --use-plugins SOME_RTF_FILE.rtf

And see if that works for you -- it should.

Finally, the output of markitdown --version would be helpful. I released a version last night where plugins were broken, but then pushed a fix about 30 minutes later. Maybe you have a bad version.

[Sillocan]

Have an example that is working

__init__.py:

"""Markitdown plugin for converting a docx while keeping image content in local files."""

# The version of the plugin interface that this plugin uses.
# The only supported version is 1 for now.
__plugin_interface_version__ = 1


from typing import TYPE_CHECKING

import mammoth
from mammoth.cli import ImageWriter
from mammoth.documents import Image
from mammoth.html import Element
from markitdown import PRIORITY_SPECIFIC_FILE_FORMAT
from markitdown.converters import DocxConverter

if TYPE_CHECKING:
    from typing import Any, BinaryIO, Callable

    from markitdown import DocumentConverterResult, MarkItDown, StreamInfo


class DocxConverterWithImages(DocxConverter):
    """Subclass of DocxConverter which outputs images rather than strip them.

    TODO: Provide method to instead output alttext
    """

    def get_image_converter(self, **kwargs: Any) -> "Callable[Image, list[Element]]" | None:
        """Return argument for `convert_image` argument in `mammoth.convert_to_html`.

        NOTE: This defaults to current directory so the CLI still works.
        """
        output_dir = kwargs.get("output_dir", ".")
        return mammoth.images.img_element(ImageWriter(output_dir))

    def convert(
        self,
        file_stream: "BinaryIO",
        stream_info: "StreamInfo",
        **kwargs: Any,  # Options to pass to the converter
    ) -> "DocumentConverterResult":
        """Convert docx while handling images in user defined way."""
        style_map = kwargs.get("style_map", None)

        return self._html_converter.convert_string(
            mammoth.convert_to_html(
                file_stream,
                style_map=style_map,
                convert_image=self.get_image_converter(**kwargs),
            ).value
        )


def register_converters(markitdown: "MarkItDown", **kwargs):
    """Register MarkItDown plugin."""
    markitdown.register_converter(DocxConverterWithImages(), priority=PRIORITY_SPECIFIC_FILE_FORMAT)

pyproject.toml:

# rest of file
[project.entry-points."markitdown.plugin"]
markitdown_docx_img_plugin = "markitdown_docx_img_plugin"
# rest of file

Then install that package and run markitdown --use-plugins <file>

[afourney]

I'll likely also enable an os Environment variable ... something like MARKITDOWN_USE_PLUGINS=1 that you can set to save yourself the trouble of typing that every time.