Examples, datasets, and movies

[dichotommy]

This issue aims to give an update on the current feelings of the @meilisearch/docs-team surrounding our examples and use of the movies.json dataset.

Discussion summary

• In the current state of examples on the site, it’s not clear if an example is referring to the actual movies.json dataset or just demonstrating a concept with a theoretical "Movies" index. There have been cases of users getting confused when they test a command on movies.json and find that they get a different response than the one displayed on the docs.
• One solution would be making all examples on the site consistent with movies.json or any other dataset. While this is achievable, it would have the side effect of limiting our flexibility and bloating example responses with unnecessary fields. Longer examples = less clear examples.
• A better solution would be to avoid referring to any existing dataset with our examples. That way we will have maximum freedom, nobody will be tempted to test our examples, and we won't be burdened with keeping them up-to-date with dataset changes.
• How to do this? We could rewrite all of our code samples, examples, and responses to refer to a single imaginary dataset, e.g. tv-shows, or use various hypothetical datasets as we see fit, e.g. shoes, pets, tv-shows.
• However, we already have example commands and responses written in eight (?) languages about a Movies dataset, so it seems like a waste to throw that away.
• Instead, we think it would be better to change Quick Start to use a different dataset, then remove movies.json from the docs site.
• This way we are still able to describe our concepts through Movies, something our users are familiar with, but there is no confusion with a literal dataset. Users who follow the Quick Start and subsequent tutorials will still have the opportunity to work with an actual dataset, but on the rest of the site examples will be purely demonstrative, referring to a theoretical Movies dataset.

Next steps

We have a total rewrite of the Quick Start guide planned for Q4 2021. This would be a nice opportunity to find/create a dataset to replace movies.json and make the switch.

The only limitations (to my knowledge) are that the dataset must be small enough to be a quick download, and should be able to show off all of MeiliSearch's most important features. Since it's a common and intuitive use-case, an eCommerce dataset would be ideal.