Experiment with different loading mechanisms

[stefannibrasil]

To make users only pay for what they use, ultimately avoiding loading all generators by default, we need to experiment with different ways of loading the generators:

• Lazy load
• Zeitwerk

We want to run experiments to compare benchmarks and the pros and cons of each mechanism taking maintainability, fewer breaking changes, easier adoption factors as guiding points.

[amatsuda]

I actually do have a branch already doing this that I created 2 years ago (carefully crafted organic code!) main...amatsuda:faker:i18n_lazy_loadable

TBH I don't remember all the details, and perhaps it wasn't 100% done. Maybe what was not done yet was benchmarking and crafting the PR description (I mean I thought I have to explain why we need this, etc., but now I feel like I can skip that part now), but if you're interested in this patch, I'm happy to finish this.

[stefannibrasil]

hi @amatsuda that's awesome, thanks for sharing! We will take a look and get back to you 🚀

[stefannibrasil]

hi @amatsuda just to let you know, we haven't forgotten about your branch. Yay for carefully crafted organic code!

I haven't had time yet to review it carefully... in the meantime, do you think it would be possible to make it an opt-in configuration? We also have revamped the benchmark suites, in case that might be helpful.

cc @thdaraujo

[stefannibrasil]

Results for both lazy loading and autoloading (zeitwerk) are ready for discussion.

[stefannibrasil]

Both results have been documented in https://github.com/faker-ruby/faker/tree/main/experiments.
Next step is to decide the solution and see how to integrate external generators so users can create their own generators.

[stefannibrasil]

We are going to move forward with lazy loading since it allows adding external generators. The idea is to cut a beta release and let users report any bugs for the next weeks.