Image in a docx file being discarded by the docx reader

[andrew-danieli-fj]

I've come across an odd issue during the conversion of my docx file to html. One of the images in the document is being removed during the conversion, but still being extracted to the media folder. The image is a .png format internally and opens fine from the copy in the media folder, so its not corrupted. It's also defined as a "caption" style in the document with the associated "text" caption also styled the same way, like the rest of the images. Instead of the expected <figure><figcaption></figcaption></figure> I just get <div data-custom-style="caption"><p>Figure caption text</p></div> in the html output. Converting to the "native" AST shows the same Div, [ Para ] internal structure.

Is it possible to turn on logging for the docx reader to see what it's doing internally during the conversion?

[jgm]

You can use --trace but it will probably not give you the info you need.

[andrew-danieli-fj]

Hi, --trace didn't return anything.

[jgm]

Is a custom style involved? See #10705

In any case, try this with a nightly to see if that fix handles this as well.

[TuongNM]

Looks like the indentation of the paragraph style causes the BlockQuotes:

pandoc/src/Text/Pandoc/Readers/Docx.hs

Line 652 in 70393b4

let transform = if relativeIndent pPr > 0 && not (numbered pPr) &&

[jgm]

Thanks for the diagnosis! Yes, the indentation is our crude way of detecting block quotes.

So, we could either add a pattern match here (so that the block quote around what would otherwise be a figure is simply ignored) -- would that have any bad effects?

[andrew-danieli-fj]

Are the block quotes an artifact of the conversion process? In the original document I have ~260 images scattered across 497 pages and all look the same when viewed through the MS365 Word GUI, with both the image and caption text using the caption style, and centred on the page. It's odd that only two images are affected like this.

I'd be happy with any solution that gets the images into the AST from the reader, as I've become adept at writing filters to connect images and their captions together when the reader can't combine them.

Thanks for writing such a useful tool.

[jgm]

The BlockQuotes are getting inserted because the paragraphs containing your images (the ones that don't come through) are indented. (The image that works isn't in an indented paragraph.) You can see this in Word if you click on the paragraph surrounding the image and look at the ruler line.

[jgm]

But it will be easy to fix now that we know what is happening.

[jgm]

Fixed in commit f64183f

[andrew-danieli-fj]

Thanks. I'll keep an eye out for the next release, but in the meantime, I adjusted the indentation of images, and that cured the problem. The reason they were indented was because they were large images and when centered looked wrong (e.g. to far to the left).

Thankfully I didn't need to tweak the output with a filter as they just worked. For now I'm going to add the workaround to my notes for prepping the document for the conversion.