Trouble saving images during build

[t-makaro]

Details

Sphinx-gallery is trying to add support for images that are saved to disk when an example is run:sphinx-gallery/sphinx-gallery#417 (comment)

I am trying to get a prototype of this to work over at: t-makaro/animatplot#14

My docs build works perfectly locally or on cirleci, but for some reason when run on RTD the image files do not seem to be produced properly. (This would be much easier to debug if I could download the built source off RTD). The images are being saved to disk using matplotlib's PillowWriter, but for some reason on RTD only the first frame is saved.

• Read the Docs project URL: https://animatplot.readthedocs.io/en/latest/auto_examples/index.html
• Build URL (if applicable): https://readthedocs.org/projects/animatplot/builds/7828348/
• Read the Docs username (if applicable): t-makaro

Expected Result

Two images are produced, and there are both animated. See:

https://145-87026896-gh.circle-artifacts.com/0/home/circleci/project/docs/build/html/auto_examples/plot_sqwell.html

Actual Result

Two images are produced, but they are static.

https://animatplot.readthedocs.io/en/latest/auto_examples/plot_sqwell.html#sphx-glr-auto-examples-plot-sqwell-py

[t-makaro]

If I switch to using matplotlib's imagemagickwriter to save the gif. Then the build fails completely: https://readthedocs.org/projects/animatplot/builds/7828320/

with:

outfile must be *.htm or *.html

[stsewd]

What are the requirements to use this feature? Maybe there is something that needs to be installed?

[stsewd]

I'll take a look on Monday

[t-makaro]

The features are still under development I'm installing a dev version of sphinx-gallery by listing the git branch in the requirements file. As far as I can tell, it is installed properly it does run my example code and produces an image which it then adds to the auto-generated documents. (Ultimately, it uses matplotlib and Pillow to save the image to disk). The problem is that the gif isn't produced properly on RTD, yet local builds and circleci work fine.

[stsewd]

I'm not familiar with the libs you are using, neither the process that they follow to generate the images, so I can't tell if this is a rtd problem. What commands are you using to build the docs locally?

[stsewd]

Ok, I have investigated this a little more, I found this happens when building with the rtd-sphinx-extension. Maybe the lib is relying on the builder to always be html? If not, maybe it's something happening with the rtd-sphinx-extension.

You can test this locally with:

pip install readthedocs-sphinx-ext

Adding to your cong.py file this

extensions = [
    'readthedocs_ext.readthedocs',
]

And building your docs with

sphinx-build -b readthedocs -d _build/doctrees-readthedocs -D language=en . _build/html

[stsewd]

Nevermind about the above comment, I don't think the extension is the problem (looks like I just needed to refresh the page to trigger the gif). I'm thinkgin something related to the libs installed locally. Maybe the pandoc version?

[stsewd]

Our latest image is using pandoc 1.16.0.2

[stsewd]

Ok, I was able to build the docs correctly inside the docker container, still not sure what is happening inside the rtd code. But something must be failing silently.

[stsewd]

Just adding that the generated git is half the size from the one generated outside rtd. Maybe a permissions issue?

[stsewd]

So, I finally was able to replicate this inside the container, if I do all the steps using a virtualenv, the generated gif is static (one frame, half the original size, etc). If I do all the steps using the global python interpreter, then the gif is generated correctly. I'm not really sure why is this happening. Locally, I was able to generate the correct gif using a virtualenv.