Parse DocBook "role" attributes as "class" whenever possible, including block elements

[de103252]

Docbook has a "role" attribute that is common to all DocBook elements and is intended to "provides additional, user-specified classification for an element".
For a few selected elements (e.g., phrase, indexterm, emphasis – all of them inline if I'm not mistaken), the Pandoc DocBook reader does examine the "role" attribute and handles it in a special way.
Would it be possible to handle "role" in a generic way for other DocBook elements as well so that a custom filter could examine it? (I hope I am phrasing this at least semi-correctly.)

What I want to achieve:

My toolchain converts from AsciiDoc to DocBook and from DocBook via Pandoc to docx. In AsciiDoc, I have block level elements that I give a custom "Finding" role (Custom role):

[.Finding]
Lorem ipsum dolor sit amet

which results in a <simpara role="Finding">Lorem ipsum dolor sit amet</simpara>. In the final Word output, I would then like to have a paragraph with the docx custom paragraph style "finding" (that I would define in the reference docx) applied.

As suggested in #7195, this could be done with a filter that acts on the respective elements, if only the information was carried over into the filter input. This would also address #3657.

[jgm]

Pandoc Para elements don't take attributes. The only way to handle this case would be to generate a Div with the attributes, wrapping the Para.

[de103252]

Okay, thanks—seems I'm at a dead end then, if I understand you correctly. To summarize, my goal is to carry an AsciiDoc custom role label over to a custom paragraph style name in Word, and that seems not to be possible in an elegant way with the current toolchain. What I could and will try to do is create a custom filter that recognizes text fragments (such as "Finding" and "Recommendation"). Or swallow the pill and change the Word output manually.

[jgm]

My comment wasn't meant to close this discussion. It would be possible to add a wrapper Div with the class. That would be enough to allow you to do what you need in filters, no?

[de103252]

Ah, so I did misunderstand you after all. I thougt that your suggestion was to add a "div" element in the DocBook source, but what you actually meant is to add a Div element in the internal Pandoc representation, right? Yes, I think that would work for me, as it would carry over the information to the filter and so would allow me to do my stuff there. If you think it will not break existing code, this looks like the right thing to do.
Would you wrap the Div around each and any block-level element, or just around those that have a role attribute?

[lifeunleaded]

Since we are anyways considering wrapping a Div, maybe even put it in attributes with key role to not lose the attribute name? role in DocBook is really generic and has many different uses. It would be no more difficult to access in filters than finding it in the classes list, I think. @jgm I was considering starting on it unless you already have?

[yanntrividic]

I am currently facing the same issue. Would love to help if needed!

[de103252]

Wow, I almost forgot about that. (I resorted to manually apply the Word style; an alternative would have been to write a simple VBA macro.)
But hey, I like elegant solutions—and what I suggested, I believe, would be an elegant solution to that problem. Namely, to treat DocBooks "role" attribute which is intended to serve as a general catch-all, in a general way in Pandoc.
If someone wants to address that, that would be much appreciated! I can't help though—while I do admire Haskell, I feel I'd better not even try to wrap my head around it...

[yanntrividic]

Ok, so I started working on something. I am not sure I am doing this the right way. If anybody has feedback for me, I am all ears. I still need to wrap the Para elements in Div elements with the corresponding attributes when needed, I think it will have to be done in the parseMixed function. Again, any help with this, or feedback, would be highly appreciated.