add support for new (2025) actualtext, artifact options of graphicx

[xworld21]

graphicx introduced two options at the very end of 2024:

• actualtext is additional text for accessibility (this is meant as 'replacement' text, rather than a descriptive text that goes into alt)
• artifact is a boolean specifying that the image is decorative

Implementation would take only a few lines, but I get the feeling that they will require a schema change and some thinking about the correct HTML equivalent (e.g. if I understand the intention correctly, actualtext should probably hide the fact that <img> is an image by overriding @role and setting @aria-label instead of @alt).

[dginev]

Could you link to the official documentation for these new keys?

[xworld21]

The new keys are documented at https://mirrors.ctan.org/macros/latex/required/graphics/grfguide.pdf section 4.4. The documentation does not explain a lot, but the behaviour is meant to match a tagged PDF feature. For ActualText, that seems to be 'replacement', meaning for instance that extracting text from a PDF should literally replace each figure with its ActualText, if present. I keep finding the example of a glyph that is provided as an image but is effectively part of the text.

Some old but relevant sources: ActualText https://www.w3.org/WAI/GL/WCAG-PDF-TECHS-20010913/Overview.html#checkpoint1-4, Artifact https://www.w3.org/TR/WCAG20-TECHS/PDF4.html

[dginev]

That's just a spec collision then. I think the right way for us to read actualtext is as an alias for alt. They are both meant to replace the image when it can not be accessed in its rendered form.

Similarly for @alt and @aria-label, I think having both is a historical and spec-related accident, but they achieve essentially the same purpose.

[xworld21]

That's just a spec collision then. I think the right way for us to read actualtext is as an alias for alt. They are both meant to replace the image when it can not be accessed in its rendered form.

Not so much collision, but HTML lacking the distinction. I just checked that alt is already used as text replacement if you e.g. copy text with an image and paste it as plain text. I think actualtext wants to also inhibit screen readers from spelling "graphic" after an image, but that does not seem possible with HTML without replacing <img> with something else entirely (a <span> with fixed size and background-image?). That must make decorated drop caps quite annoying.

A simple change would be to map both actualtext and alt to alt, with preference for alt which should contain more information. artifact should simply enforce alt="", that is, prevent LaTeXML from extracting the description from somewhere else, like a caption.

A more sophisticated solution might be to map actualtext to alt and alt to aria-description when both are present.

[dginev]

A simple change would be to map both actualtext and alt to alt, with preference for alt which should contain more information.

Yes, agreed. Making that more sophisticated may just end up confusing/redundant.

artifact should simply enforce alt="", that is, prevent LaTeXML from extracting the description from somewhere else, like a caption.

I think the best practice there is aria-hidden="true" instead.