(Sphinx caches the parsed input files in the doctrees directory)
Build the same Sphinx project using the html builder
(Sphinx will read the cached doctrees which reference the PDF images instead of the SVG images)
I think the proper way to handle this would be to subclass ImageConverter and handle conversion as a post-transform. The drawio exporting should be moved from DrawIOBase to this new class. sphinx.ext.imgconverter is an example of how ImageConverter can be subclassed.
A workaround is to use separate doctrees directories for each builder.
Steps to reproduce:
I think the proper way to handle this would be to subclass ImageConverter and handle conversion as a post-transform. The drawio exporting should be moved from DrawIOBase to this new class. sphinx.ext.imgconverter is an example of how ImageConverter can be subclassed.
A workaround is to use separate doctrees directories for each builder.