Alt-Text from draw.io document text

[DennyWebb]

Feature consideration/request: Parse the draw.io document for all text, then embed that text as the alt text of the image so that searches/search engines can parse it and make image contents searchable.

[LukeCarrier]

Hi @DennyWebb, thanks for the suggestion. Whilst I appreciate this was some time ago, is there any chance you'd be able to provide some example source diagrams containing the alternate text and what you'd expect the summary to be?

My understanding is that it's generally better to limit alt attribute text to ~150 characters, as it could otherwise be a hassle to users of screen readers. I would imagine we'd be better off using HTML <figure> and <figcaption> for that purpose.

Some possibly helpful context:

• Diagrams.net supports adding tooltips to elements, but not the root.
• You can apply arbitrary data (in key-value pairs) to both the root and to elements. If you use this, below the root element of the diagram the object will have an alt="blah" attribute.
• There's also a text plugin that exposes an Extras -> Extract Text... menu item.

[DennyWebb]

Sure! Sort of... without giving away anything sensitive, imagine a draw IO that's one network diagram out of a dozen or so different ones across a large organization... so servernames, workstation names, switch names, etc all over the place in the draw file. Then, I have a report of a machine acting up, I can quickly go and search to find the draw image without having to dig to figure out which department that machine is hosted in.

Granted, this can also be solved by better naming conventions/etc to make these things more obvious... but... people are people.

For a more specific example: In my specific case it was actually a function name from an error log, and trying to find which application that function existed in. There were some issues with the logging so we weren't getting a complete picture.

The figcaption is a great suggestion. Really it wouldn't even have to be on the image itself though, just at the page level as something that would make the site search identify that page as containing the content, regardless on if it's actually displayed.

[LukeCarrier]

Since the primary concern here is making the text from diagrams searchable, I think the path of least resistance is #40, where we'll make it possible to inline the contents of an exported SVG rather than embedding it as an <object>/<img>.

[LukeCarrier]

Closed #40 in favour of #56, which should land in v0.9.0. Unfortunately it doesn't look as though the search indexer recurses into <svg> elements to find the text, so I need to troubleshoot this a little further.

I'm going to keep this out of the next release for now, but will try to get back around to it soon. I suspect this might need some changes to MkDocs itself.