Distributing this thing

[etpinard]

On top of the @plotly/image-exporter npm package , we'll add standalone Mac, Linux and Window binaries to the release section after every release similar to what's done currently in plotly-database-connector.

Standalone binaries ship with node.js and electron bundled up, but without the fonts we use on prod (if users want the fonts, they'll have to install them locally or use docker). We'll probably use electron-packager same as plotly-database-connector for this task, although there are other options. This part is easy.

Now, we need to figure how to distribute image-exporter to python and R API users. We'll also need to write (thin) python and R wrapper around the plotly-graph-exporter CLI command e.g in python:

from subprocess import call
import json

fig = {"data": [{"y": [1,2,1]}]}
call(['plotly-graph-exporter', json.dumps(fig)])

For distributing, I can think of three different options:

1. Include the image-exporter standalone binaries in plotly.py and plotly.R packages and write the wrapper there, similar to what we currently do with the plotly.js bundle for offline. Potential problems, we'll have to make platform-dependent installs - is that possible with pip and CRAN? Download times might increase significantly too.
2. Make new plotly-image-exporter py/pip and R/CRAN packages from this repo - which would include the py/R wrappers. This option would require platform-specific install too.
3. Make users download and install the standalone binaries from this repo's release section, add wrapper code to plotly.py and plotly.R and make them spit out errors if called when plotly-graph-exporter isn't installed.

cc @chriddyp @cpsievert @jackparmer

[cpsievert]

(3) would certainly be doable/easiest for R (assuming bundles are relatively small)

[jonmmease]

Quick thought regarding packing this for python users.

poppler and nodejs are both available in the main conda repository (at least for linux and os x, poppler actually isn't available in the window repo). I wonder if it would be feasible (sensible?) to create a conda package that has dependecies on nodejs and poppler and basically just contains the bin/, src/, and node_modules/ directories from this repo, plus a thin python subprocess wrapper around plotly-graph-exporter.js.

cc @jackparmer @chriddyp

[jackparmer]

I wonder if it would be feasible (sensible?) to create a conda package that has dependecies on nodejs and poppler and basically just contains the bin/, src/, and node_modules/ directories from this repo, plus a thin python subprocess wrapper around plotly-graph-exporter.js.

I'm all for this as long as we can get to a point where we feel it works in a bulletproof way. Otherwise be prepared for hundreds of plotly.py user complaints/issues on GitHub and the forum 😬

[etpinard]

To note, we only need poppler for eps exports. Not sure what percentage of python users export in eps, but I suspect it is small.

[jonmmease]

I was assuming poppler was needed for PDF export as well. Nice to know that this isn't the case. So poppler could be an optional dependency.

I may try out what I outlined above for some internal needs. If I do I'll report back. But, to Jack's point, this would certainly require a bit more thought, and a lot more testing, before promoting it to the plotly.py community.

[JamesCropcho]

I tend to favor one way of doing something, over multiple (yes, it's very pythonic of me, ha). So, though my voice is small in these matters, in this case I therefore advise a single vector of local installation.

[jonmmease]

@JamesCropcho That's certainly reasonable :-) What would be your preferred approach for local installation? Is it currently possible to create standalone binaries? I don't see anything in https://github.com/plotly/image-exporter/releases and I don't know enough about electron to build them myself from the repo.

[JamesCropcho]

@jmmease Local installation is actually being figured out and implemented by Plotly and me very actively right now. Hopefully this will be soon possible for Plotly users.

[etpinard]

from @jackparmer

[JamesCropcho]

Orca sounds nice.

[jackparmer]

@CPetzold and NPM support very kindly helped us acquire the NPM package name orca. 🙇 https://www.npmjs.com/package/orca is owned by NPM user plotly now.

[etpinard]

More things to think about from #78:

@n-riesco wrote:

Since the electron app only has one entry point (package.json:main), how about having a single app named plotly-exporter that takes as first argument server, graph, dash-report, ...?

I replied:

Augmenting on @n-riesco 's #78 (comment), maybe we could expose that single entry point as plotly-export (instead of plotly-exporter), so that we get very clean (and cute) sounding CLI commands like:

$ plotly-export graph first.json second.json https://plot.ly/~empet/10 --format pdf

$ plotly-export dash-report https://my-dash-report.com

$ plotly-export serve --port 9090

---

I still think a single entry point would be the option that scales the best for this project. So now we have:

$ orca graph first.json second.json https://plot.ly/~empet/10 --format pdf

$ orca dash-report https://my-dash-report.com

where the first argument indicates which orca component to use. @JamesCropcho had some reserves about orca serve. We can leave that out of the first release.