search

gcalderone / Gnuplot.jl

Julia interface to gnuplot
Other
119 stars 16 forks source link

SVG tool tips don't work in JupyterLab #21

Closed lyonsquark closed 4 years ago

lyonsquark commented 4 years ago

Hi! Thanks for making gnuplots show up in Jupyter notebooks! If I turn on enhanced mouse for the svg terminal, 

Gnuplot.options.term_svg = "svg enhanced mouse size 700,500 lw 1";

the expected tool tips don't work in the plot output in JupyterLab. This apparently is a known issue (or maybe feature) - see https://github.com/jupyterlab/jupyterlab/issues/7497 . The workaround is to give Jupyter a text/html MIME type instead of image/svg+xml. I tried this in my dev version of the package. On line https://github.com/gcalderone/Gnuplot.jl/blob/bad5e12681960a9cf4eac526c36685421638bcb3/src/Gnuplot.jl#L1520 I replaced ::MIME"image/svg+xml" with ::MIME"text/html". The plot appears correctly and tool tips work!

I have no idea if this will affect Juno or other editors. But it works for JupyterLab. Since this is such a small change, I don't think it deserves a full pull request, but I can submit one if you'd like. I'm worried about side-effects though for other systems. -- Adam

gcalderone commented 4 years ago

Thank you for reporting. The show(io::IO, ::MIME"image/svg+xml" method actually writes a svg file, not an html, so I can't change that.

The workaround you proposed works since, apparently, JupyterLab properly detects and display the svg image even if an html was required.

A possible solution is to implement a separate show(io::IO, ::MIME"text/html" method using the canvas terminal provided by gnuplot. I already prepared a branch. You can test it with:

add Gnuplot#testCanvasTerminal

and create plots as you normally do.

In this branch there is a new option, Gnuplot.options.term_html, wth default value canvas mousing. If you still prefer using a svg file in JupyterLab you can change it as follows:

Gnuplot.options.term_html = "svg enhanced mouse size 700,500 lw 1";
lyonsquark commented 4 years ago

Thanks! Trying it...

lyonsquark commented 4 years ago

You likely know this already, but I was wondering how IJulia figured out what MIME type to display. I found the priority list... https://github.com/JuliaLang/IJulia.jl/blob/8c0f770309c88db402f5ed737ef7ae4312772fa7/src/inline.jl#L5-L20

gcalderone commented 4 years ago

No I didn't know it, thank you! This means the changes in testCanvasTerminal would also affect IJulia users....

Let's fix the problem with JupyterLab, then I'll think how to make everything flexible so that a user can choose the preferred terminal for images to be exported in the IDE.

Please confirm if the new branch works properly in JupyterLab. Thanks!

lyonsquark commented 4 years ago

Ok - I have good news and bad news. Setting Gnuplot.options.term_html = "svg enhanced mouse size 700,500 lw 1"; works great.

The default Gnuplot.options.term_html = "canvas mousing" does not work at all. I just get a blank square. I think is because the html that Gnuplot writes is not embeddable without putting it within an <iframe>. The canvas terminal has a default option standalone that makes a full html file (including <!DOCTYPE HTML>). The alternate option, name, writes straight javascript to the output file. You would need to write an html wrapper to make that work.

I think a good solution is to just use svg. That works well.

lyonsquark commented 4 years ago

Also, you should change the extension of the temp file for text/html to .html. Not that it really matters, but it makes more sense than .png.

gcalderone commented 4 years ago

I added the changes in the master branch, and deleted the testCanvasTerminal branch:

]add Gnuplot#master

I also changed the way terminal options are specified. In your case:

Gnuplot.options.mime[MIME"text/html"] = "svg enhanced mouse size 700,500 lw 1"

Default value is svg enhanced dynamic, which is supposed to work without problems for the majority of users.

You may explicitly preview what is being passed to JupyterLab with, e.g.

@gp 1:9
save(MIME"text/html", output="tmp.svg")
gcalderone commented 4 years ago

The feature is now available in the version 1.3.0. Further informations in the docs and in this notebook.

Please consider closing the issue if you think the issue is solved.

lyonsquark commented 4 years ago

Thanks! Trying it now.

lyonsquark commented 4 years ago

I can confirm this now works! Thanks very much!!

lyon-fnal commented 4 years ago

I wrote too soon - multiplot doesn't work. I'll open a new issue and look at it.