Feature Request: Support SVG output for bokeh

[syntonym]

Currently pandoc-plot with the bokeh toolkit only supports png and html output. Bokeh itself can output svgs source. This is especially relevant for bokeh as the bokeh png export unfortunately ignores DPI and always generates pngs with 92 DPI (ticket) which makes it less useful for pdf files.

[LaurentRDC]

Yes we can do this! I'll add support for this over lunch.

BTW, how do you set the dpi on a Bokeh plot?

[syntonym]

BTW, how do you set the dpi on a Bokeh plot?

As far as I know you can't currently. Bokeh itself always deals in css pixels, when exported the DPI is up to the browser, and apparently selenium always uses 92 DPI. Hm, apparently that is not true this blogpost says there is an easy way to set DPI. But that should be implemented in bokeh.

Does pandoc-plot currently support multiple bokeh plots/layouts? Bokeh SVGs unfortunately don't support this, but I don't think there is any good way around that.

[LaurentRDC]

pandoc-plot can only include one file per figure. I have not tested using layouts, but there's no reason it should not be possible in the near future.

Right now, for technical reasons, pandoc-plot determines the most recent bokeh Plot object and saves that to file.

I think the most general plotting object in Bokeh is a Document, so we would need to tweak pandoc-plot a little to create multi-plot layouts by saving the most recent Document.

[LaurentRDC]

So SVG support is in (I just tested it).

I also changed (in 8599869) the way the "current" figure is found, which implies that plots made with bokeh.layouts will be saved appropriately. This will come in the next version (probably 0.8.2.0).

Is there anything else related to bokeh that you think we should include in the next release?

[syntonym]

There is actually one more thing which bothers me with bokeh plots:
If bokeh plots include a lot of data, no plot and no text/html/other plots following are rendered.
This is because bokeh generated HTML as following:

<script type="application/json" id="$ID">
    {"id": $DATA}
</script>
<script type="text/javascript">

(function() {
     var fn = function() {
     ...
     root.Bokeh.embed.embed_items($JSON_WITH_ID_$ID, render_items);
     ...
     };
     if (document.readyState != "loading") fn();
     else document.addEventListener("DOMContentLoaded", fn);
 })()
</script>

So if $DATA is big, this just takes a long time to parse.
This is inline, so text/html/plots/whatever after these script tags are not shown until the script tag is parsed.

One possible solution is moving the first script tag with all the data to the bottom of the document.
The cons for this is 1) additional work needs to be done to identify this script tag (there should be some support in the bokeh library, see this function 2) now the filter needs to be stateful to include something at the bottom of the document and 3) pandoc-plot is now tied more closely to the bokeh library, making it more brittle.

This can actually be circumvented by using bokeh.sources.AjaxDataSource and putting the data as separate json files.

All in all I don't think it's worth solving this in pandoc-plot.

[syntonym]

One more thing came to my mind, although unrelated to bokeh: Currently pandoc-plot caches figures on the sourcecode used. When figures depend on external code or data, these dependencies are not taken into account and the figures are not rerendered when the dependencies change. Would it be possible to incorporate this?

One possible solution would be an option depends=FILE which would additionally hash FILE. Option for multiple files would be great.

A different solution would be an optional user defined function which evaluates whether the figure needs regeneration. This seems very complicated.

Another option could be to inject a function "open" into the code and automatically registering dependencies whenever this function is used to read files.

[LaurentRDC]

There is actually one more thing which bothers me with bokeh plots:
If bokeh plots include a lot of data, no plot and no text/html/other plots following are rendered.
This is because bokeh generated HTML as following:

<script type="application/json" id="$ID">
    {"id": $DATA}
</script>
<script type="text/javascript">

(function() {
     var fn = function() {
     ...
     root.Bokeh.embed.embed_items($JSON_WITH_ID_$ID, render_items);
     ...
     };
     if (document.readyState != "loading") fn();
     else document.addEventListener("DOMContentLoaded", fn);
 })()
</script>

So if $DATA is big, this just takes a long time to parse.
This is inline, so text/html/plots/whatever after these script tags are not shown until the script tag is parsed.

One possible solution is moving the first script tag with all the data to the bottom of the document.
The cons for this is 1) additional work needs to be done to identify this script tag (there should be some support in the bokeh library, see this function 2) now the filter needs to be stateful to include something at the bottom of the document and 3) pandoc-plot is now tied more closely to the bokeh library, making it more brittle.

This can actually be circumvented by using bokeh.sources.AjaxDataSource and putting the data as separate json files.

All in all I don't think it's worth solving this in pandoc-plot.

Actually, pandoc-plot parses <script> tags to be to embed them correctly. Is this something that can be fixed by replacing <script> with <script async> ? That's an easy change. Please open another issue for this.

Edit: it appears that replacing <script> with <script defer> would fix it. That's easy.

[LaurentRDC]

One more thing came to my mind, although unrelated to bokeh: Currently pandoc-plot caches figures on the sourcecode used. When figures depend on external code or data, these dependencies are not taken into account and the figures are not rerendered when the dependencies change. Would it be possible to incorporate this?

One possible solution would be an option depends=FILE which would additionally hash FILE. Option for multiple files would be great.

A different solution would be an optional user defined function which evaluates whether the figure needs regeneration. This seems very complicated.

Another option could be to inject a function "open" into the code and automatically registering dependencies whenever this function is used to read files.

I can see adding a parameter cache=True|False on a particular plot, to always re-render a figure if it depends on external (non-deterministic) data. Please open another issue for this :)

[LaurentRDC]

The change is live in the latest version (0.9.0.0). Thank you for your suggestion.