`--embed-resources` is not respected in PDF conversion with WeasyPrint

[jamiethecat]

Explain the problem.
The intermediary HTML when converting a DOCX or ODT to PDF with WeasyPrint always contains images as paths, even when --embed-resources is used.

Command to reproduce:

test.docx

pandoc test.docx --embed-resources --pdf-engine=weasyprint --output=test.pdf

Which fails with:

ERROR: Failed to load image at 'file:///home/monster/Downloads/media/image1.png': URLError: <urlopen error [Errno 2] No such file or directory: '/home/monster/Downloads/media/image1.png'>

The error matches the error you get when you use WeasyPrint directly without using --embed-resources in the original HTML conversion:

pandoc test.docx --output=test.html
weasyprint test.html test.pdf

The following works correctly:

pandoc test.docx --embed-resources --output=test.html
weasyprint test.html test.pdf

Pandoc version?
3.7.0.2

[jgm]

See https://github.com/jgm/pandoc/blob/main/src/Text/Pandoc/App.hs#L318-L339
Because output is to a PDF, we don't follow a code path where the embed-resources transformation is applied.

Does this error message come from pandoc or weasyprint? (when you run them separately as you tried above)

Perhaps the problem is that weasyprint doesn't deal with file:/// URIs?

[jamiethecat]

Does this error message come from pandoc or weasyprint? (when you run them separately as you tried above)

WeasyPrint

Perhaps the problem is that weasyprint doesn't deal with file:/// URIs?

Pandoc outputs it as a relative path, so turning it into a URI is WeasyPrint's doing:

<p><img src="media/image1.png"
style="width:6.69306in;height:6.69306in" /></p>

[jgm]

And is there a file media/image1.png relative to the working directory?

[jamiethecat]

There isn't, I think it's getting that from the contents of the DOCX file (going off of the name and path, although in reality it's in test.docx/word/media/image1.png).

[jgm]

OK. With our other PDF engines, there is code that handles this sort of thing by extracting from the mediabag. It could be that the WeasyPrint interface code doesn't do this; that's something I could look into.

[jgm]

You can probably work around this by using a two step approach.

pandoc test.docx --output=test.html -s --extract-media=media
weasyprint test.html test.pdf

[jgm]

OK, I changed it so that if there is a nonempty mediabag, everything is embeded prior to passing the HTML to weasyprint.