SVGs can create content streams that reference resources / indirect objects not present in PDF

[reknih]

Hi Felix,
I could not help but notice that you recently added SVG support to printpdf using my svg2pdf crate (thanks!).

Converting an SVG to a PDF often requires more than a single content stream/object. svg2pdf heavily uses the /Do operator to draw other XObjects and references many objects through the Resources dictionary of the page. printpdf currently discards all of the svg2pdf output except for the main content stream.

This means that SVGs using the features listed below will produce content streams with broken references:

• transparency of any kind, be it fill or stroke opacity or whole groups
• gradients
• patterns
• masks
• external images
• some transformed groups
• filters (e.g. blend modes) -- if I'll ever support those 😬

To fix this, printpdf would have to collect all objects in the Page's Resources dictionary and the Resources dictionary of all referenced XObjects recursively and carry them over to the target file.

In order to reproduce the bug, I converted the attached SVG to a PDF file with printpdf and a standalone svg2pdf setup. The printpdf result cannot be opened in Acrobat and only shows a solid orange rectangle in Pdfium while the svg2pdf output reproduces the output of an SVG viewer.

The original SVG (on OneDrive because GitHub does not allow the file format in issues)
The printpdf result (pardon the margins)
The svg2pdf result

[fschutt]

I suspected that, but I personally just use this to create maps and to instantiate SVG symbols in my map (without bloating the file size):

For my purpose it's okay, since I only use simple symbols (basic paths, etc.). Right now I don't need to support gradients or complex symbols, it's just important that I can use the SVG format, so that I can edit symbol sets using an SVG editor.

Thanks for making svg2pdf by the way, placing SVG symbols in the map was the last problem I had with putting a proper map together.

[08-3]

Is there a way to mitigate that ? by a transformation of the source SVG by any chance ?
I'd like to use an SVG in printpdf which doesn't render well, while svg2pdf renders it correctly.

[fschutt]

Since printpdf now has a full PDF parser, it parses the output of svg2pdf again. So technically it could add all these things to the PDF resources.

[fschutt]

This is "fixed" in 0.8.

Now, obviously there are SVG things that are not translated properly, but the entire PDF is parsed output by svg2pdf is parsed, which includes the resources.

Yes, gradients / patterns are currently not supported. But the infrastructure is there to technically support it.