Render html and pdf with Julia/PlotlyLight plots

[dpo]

Description

Hi. Thank you for making Quarto, which I use extensively in research and teaching.

I'm trying to render Quarto notebooks to PDF (think many reports from students, so it should be done from the command line with, e.g., quarto render report.qmd --execute --to pdf).
The notebooks are in Julia and produce plots with PlotlyLight.jl; a lightweight interface to Plotly that is important to Julia users as it produces plots much faster than any other plotting package, and so I find it rather essential in the classroom.

There's another package called PlotlyKaleido.jl that can be used to download a plot in svg, png, pdf, etc.

The following sample notebook illustrates what I'm trying to accomplish.
The trouble is that the downloaded figure is nowhere to be found, and therefore, conversion to PDF fails.
Rending the html is broken as well in the sense that the image is not there.

---
title: PlotlyLight and Quarto
engine: julia
---

```{julia}
#| output: false
using Pkg
Pkg.activate("plotlylight-quarto")
Pkg.add("PlotlyLight")
Pkg.add("PlotlyKaleido")

using PlotlyLight
using PlotlyKaleido
PlotlyKaleido.start()
```

# A plot with PlotlyLight

```{julia}
p = plot(x = 1:200, y = cumsum(randn(200)), type="scatter", mode="lines" )
p.layout.title.text = "Your stocks"
display(p)  # this plot appears in the html rendering
```

<!-- Now I would like this plot to appear in the pdf rendering -->

```julia
PlotlyKaleido.savefig(p, "myplot.svg")
```
![](myplot.svg)

A similar ipynb notebook works (the svg image is indeed downloaded and included in the html).

I opened an issue over at PlotlyKaleido.jl, but the fact that the image is downloaded with Jupyter makes me think that someone here may have an idea of what's going on.

Thank you in advance.

[mcanouil]

You are not executing PlotlyKaleido.savefig(p, "myplot.svg").
It's a regular code block and not a code cell that is evaluated. You need the brackets.

[dpo]

DUH. Thank you. julia and {julia} are highlighted the same and I didn't spot the error.

[dpo]

Months later, I'm back to the question above. It appears that the html rendering no longer includes the plot. If I run the simplified sample notebook

---
title: PlotlyLight and Quarto
engine: julia
---

```{julia}
#| output: false
using Pkg
Pkg.activate("plotlylight-quarto")
Pkg.add("PlotlyLight")
Pkg.add("PlotlyKaleido")

using PlotlyLight
using PlotlyKaleido
PlotlyKaleido.start()
```

# A plot with PlotlyLight

```{julia}
p = plot(x = 1:200, y = cumsum(randn(200)), type="scatter", mode="lines" )
p.layout.title.text = "Your stocks"
display(p)  # this plot no longer appears in the html rendering
```

the plot pops up in a new tab of my default browser while the html file generated does not contain the plot.
I observe this behavior from VSCode and from the command line.

I am running Quarto 1.7.30 installed from Homebrew on a Mac M1.

[dpo]

NB: The 3 lines involving PlotlyKaleido can be removed in the script above, to further simplify it. I meant to do that but forgot, sorry.

[dpo]

What happens in display(p) is just replaced with p? Same thing?

Same thing.

[MichaelHatherly]

(Maybe the recent changes in JuliaComputing/PlotlyLight.jl@a9a1870 have something to do with this change in behaviour, if it used to work alright.)

[dpo]

You are absolutely right. The plot appears in the html output with PlotlyLight 0.9, 0.10 and 0.11, but not with 0.12. I'll open an issue there to begin with. Thank you for your assistance!

[dpo]

Actually, I'm wrong. It seems to be the interaction between PlotlyLight and PlotlyKaleido that causes the issue. If I remove the 3 lines about PlotlyKaleido in my notebook, I get the expected html output.

[dpo]

So sorry to come back again with the same problematic notebook. I'm teaching a class where the students use Quarto and, to my surprise, things that worked on macOS and Linux didn't on Windows. I personally don't use Windows, but I was able to reproduce the glitch on one of the Windows 10 workstations with a fresh install of VSCode, Julia and Quarto.
The following works fine:

---
title: PlotlyLight and Quarto
engine: julia
---

```{julia}
#| output: false
using Pkg
Pkg.activate("plotlylight-quarto")
Pkg.add("PlotlyLight")
#Pkg.add("PlotlyKaleido")

using PlotlyLight
#using PlotlyKaleido
#PlotlyKaleido.start()
```

# A plot with PlotlyLight

```{julia}
p = plot(x = 1:200, y = cumsum(randn(200)), type="scatter", mode="lines" )
p.layout.title.text = "Your stocks"
display(p)  # this plot appears in the html rendering
```

However, uncomment the 3 lines about PlotlyKaleido, and quarto render hangs forever, apparently stuck on the line using Pkg. I invoke quarto render from the VSCode command palette.

This time I feel that the issue is with Quarto because running the same Julia commands at a Julia prompt works fine on all platforms.

I ran quarto check, which came out clean. (Jupyter was missing, I installed it, and the error persists; it's not clear if installing Jupyter is required or not.)

Any ideas what may be happening here? Thanks in advance!!!

[dpo]

In fact I don't think the issue is specific to PlotlyLight, PlotlyKaleido, etc. Quarto also hangs with certain other packages. For instance, this makes Quarto hang on Windows:

using Pkg
Pkg.activate("deblur")
Pkg.add("Images")
Pkg.add("ToeplitzMatrices")
using LinearAlgebra
using Images, ToeplitzMatrices

[cderv]

I don't know how QuartoNoteebookRunnner.jl does behave when activating from within a cell.

What I tried is the following

• Create a _setup.qmd document to setup the Julia env for the default project (current Quarto project)

   ---
   title: Julia env setup
   engine: julia
   ---
   
   ```{julia}
   using Pkg
   Pkg.add("Images")
   Pkg.add("ToeplitzMatrices")
   ```

• Call quarto render _setup.qmd so that the cell is run for installation

• Then in main document index.qmd don't install package

   ---
   title: "test"
   engine: julia
   ---
   
   ## Using Julia
   
   ```{julia}
   using LinearAlgebra
   using Images, ToeplitzMatrices
   ```

  This will use the default env and so the packages will be available.

I get no hang on Windows with this process.

I am using default project for Julia here (--project=.) but it can be a custom one by just adding this in the YAML for all docs

engine: julia
julia:
  exeflags: 
    - '--project=deblur'

This way Julia will be started using a specific project.

I am sharing this because it is possible that these are the installation steps using Pkg.add that create the handling. I'll let @MichaelHatherly, who knows way more, shares insights on this too.

Hope it helps.

[MichaelHatherly]

I don't know how QuartoNoteebookRunnner.jl does behave when activating from within a cell.

It's not really the way that we envisioned it being used. Our preference has been for performing all Pkg operations using a Julia REPL and then just pointing to already instantiated environment using the --project flag.

That being said, @dpo it would be best if you open an issue related to this in QNR with a reproduction of this problem and we can see what, if anything, that can be done about this.

[dpo]

Our preference has been for performing all Pkg operations using a Julia REPL and then just pointing to already instantiated environment using the --project flag.

Really? It’s much less self contained that way. In addition, the notebooks above all work correctly on macOS and linux, so it really seems to boil down to a bug on Windows.

[MichaelHatherly]

so it really seems to boil down to a bug on Windows.

Open an issue in https://github.com/PumasAI/QuartoNotebookRunner.jl, please.

Really? It’s much less self contained that way.

We just have different core use cases. I'm happy to discuss ways that we can improve the "self-contained" workflow, but these cannot be done in a discussion thread in this repo. Open issues in QNR and we can discuss there instead.