SVG images in Markdown are not rendered in latex export

[amueller]

This is a bit of a tricky one I guess.
By default latex doesn't support svg images, but they are a nice format if you want to show vector graphics in a browser. I could convert to png but that seems not great.
there is a svg package that could be used to convert svg packages to pdf on the fly.

[jakobgager]

Actually svgs should be converted to pdf using inkscape during the nbconversion. How do you use/embed the svg?

[amueller]

![some_svg](some.svg) in a markup cell. I'll try a minimal reproduction later.

[amueller]

Ok, so I did ![svg_thing](overfitting_underfitting_cartoon.svg) which gave me

! LaTeX Error: Unknown graphics extension: .svg.

See the LaTeX manual or LaTeX Companion for explanation.
Type  H <return>  for immediate help.
 ...                                              

l.255 ...ics{overfitting_underfitting_cartoon.svg}

? 
! Emergency stop.
 ...                                              

l.255 ...ics{overfitting_underfitting_cartoon.svg}

!  ==> Fatal error occurred, no output PDF file produced!
Transcript written on notebook.log.

when trying to convert to pdf.

the generated latex is

    \begin{figure}[htbp]
\centering
\includegraphics{overfitting_underfitting_cartoon.svg}
\caption{svg\_thing}
\end{figure}

[jakobgager]

You are right. In your particular case only the syntax is converted from markdown to latex but the linked content not.
If the svg data is embedded in the notebook, e.g. created by a matplotlib plot this svg data is converted during the extraction to pdf. Hence, it would be a good idea to write a preprocessor which captures all linked content and converts is as required. Here, it would be possible a well to download images which are linked via a web address.

[mpacer]

Is there a reason this is better to have as a jupyter extension rather than as part of core nbconvert preprocessors?

@Carreau, @takluyver, @minrk how do we determine what goes in the extensions repo and what is incorporated into core projects? Is there a system for migrating things? This seems like important svg support and may address other issues that I'm having difficulty finding right now.

[amueller]

I agree with @michaelpacer that this would be good to have in the main repo.

[takluyver]

@michaelpacer sometimes we ask people to make an extension if something is too specific for the main repos, or we're not sure about how to design it. But in this case, it's an extension largely because @juhasch, who maintains the repo with a collection of extensions, got round to doing it.

There's no system to migrate things to or from extensions, it would just involve making a pull request like any other feature.

Note that we already have svg2pdf conversion, but as @jakobgager said, it doesn't currently handle svgs linked into the notebook in markdown cells.

[romor001]

Ran into this issue when trying to convert my notebooks contaning SVG images in Markdown into Pdf.
Any workaround recommended other than converting the SVGs to PNG?

[MichaelAdolph]

@romor001 I also ran into trouble with converting SVGs. I hunted down the problem, and indeed: there are two problems:

Some specs: I'm using jupyter-nbconvert 6.0.7 on python 3.8.6, windows PC.
Inkscape is installed into the default folder C:\Program Files\Inkscape

And here is the first problem, the whitespace in "Program Files" makes the call fail, because it is not enclosed in double quotes or in another way escaped.

The second problem is that the initial --export-filename will raise an error because auf an overwriting risk, it should be changed to --export-type. But than you have to change the value (from a filename to just "pdf").

All in all I change the site-packages\nbconvert\preprocessors\svg2pdf.py the following way:

    @default('command')
    def _command_default(self):
        self.inkscape_path_string = f'\"{self.inkscape}\"' # added this to escape the whitespace
        major_verison = self.inkscape_version.split('.')[0]
        export_option = ' --export-type' if int(major_verison) > 0 else ' --export-pdf'
        gui_option = '' if int(major_verison) > 0 else ' --without-gui'
        to_filename = 'pdf' if int(major_verison) > 0 else to_filename # Need this for backward-capability

        return '{inkscape}{gui_option}{export_option}='.format(
            inkscape=self.inkscape_path_string, export_option=export_option, gui_option=gui_option
        ) + f'{to_filename}' + " {from_filename}" # needed to update the string, otherwise it did not work for my (but that is probably my fault)

Now converting Jupyter Notebooks works.

I'll look at a way to update the package and create a pull request.

[MichaelAdolph]

I added a pull request for this issue #1469 , this should also work for issue #1325

[aravindh-krishnamoorthy]

In PR #2190 a workaround is provided that hooks onto \includegraphics and replaces it with \includesvg for SVG files. But the resulting LaTeX files need -shell-escape to compile as \includesvg uses Inkscape for conversion, see also issue #2192.

I understand that -shell-escape may not be secure and might allow malicious users to include arbitrary commands in a Python notebook, which will activate during LaTeX compilation or export.

On the other hand, I also do not think messing with ![](file.svg) in a markdown cell and converting it to PDF on the fly is a good idea.

So, not sure what's the best way forward here.