Release official Docker image + pixel exact image tests

[antoinerg]

Closes #187

@scjody suggested we reuse our existing Dockerfile to publish our official Docker images. In order to do so, I had to adapt its ENTRYPOINT to support CLI invocation (see #187 (comment) for details). If the first argument is either --version, --help or graph we assume Orca is used as a CLI. If the first argument is serve or anything else, we launch the server as we used to do.

• Publish Docker image as plotly/orca
• Support CLI invocation
• Advertise our Docker image in the documentation
• Add image tests

image tests

The inclusion of image tests in this PR eliminates the risk of missing dependencies in the final Docker image. By using software rendering, we get reproducible images that we can compare pixel per pixel with a set of baselines in an exact manner. By visually testing all image formats (pdf, eps, svg, png, emf), we catch missing system dependencies such as Poppler, Inkscape, Mathjax or missing fonts. Hopefully, this will catch regressions whenever we perform system upgrades.

• document image tests in CONTRIBUTING.md
• runs automatically on CircleCI as a final job

@plotly/devops please have a look at the new entrypoint.

[jonmmease]

Thanks for working on this @antoinerg, this is going to be really helpful!

I'd like to see if we can work through how Python/R/Julia users could take advantage of orca as a docker image without each library needing to include docker related logic. Would it be possible to distribute a shell script, orca.sh, that presents that same interface as the native orca binary? Something like

if orca graph then call docker run -i quay.io/plotly/orca graph "$@"

if orca serve --port=8123 then call docker run -d -p 8123:8123 quay.io/plotly/orca "$@"

The R and Julia libraries use the orca graph command line API, while plotly.py uses the orca serve API with --port and --graph-only options. If we had a shell script like this for users to put on their path then I think all of the client libraries would work without modification. Can docker pull add something to the PATH or would that have to be a manual step?

What do you think?

[antoinerg]

Thanks for working on this @antoinerg, this is going to be really helpful!

I'm glad to see lots of people are excited about this one!

I'd like to see if we can work through how Python/R/Julia users could take advantage of orca as a docker image without each library needing to include docker related logic. Would it be possible to distribute a shell script, orca.sh, that presents that same interface as the native orca binary?

That is certainly one of the goal here. Making an orca.sh shell script calling Docker as you suggested would be great and I will try to add one to this PR.

Can docker pull add something to the PATH or would that have to be a manual step?

As far as I know, it cannot. It would have to be a manual step.

[scjody]

Sounds like a great idea to me!

Minor request: if you're adding a tool that users are expected to run, it should not have a .sh (or .bash, .py, etc.) extension. Rationale

Also note that we have a company-wide bash style guide (mostly incorporates Progrium's).

[antoinerg]

@jonmmease I wrote a little shell script in commit 94bb0a4 to call Orca in Docker. Please see the commit message for details. Also, can you test it out on MacOS and adapt it if it doesn't work. Anyway, let me know what you think!

[antoinerg]

Any comments for this PR? @etpinard @jonmmease @plotly/devops

Thank you all!

[etpinard]

💃 from me