I’ve played some Minecraft modpacks and found some insights.
First, why does it matter? Because Julia and Minecraft modpacks share many structures, independent modules working together, and the same module in different contexts can feel and behave completely differently.
While it’s possible to add individual mods to your playthrough, you would usually take a pre-packaged modpack, usually with a few tweaks to glue them together, and with a guidebook.
Back to Julia, there are many packages, and composed together, they can do great things, this is opposed to a few large, tightly coupled packages other ecosystems use. However, this comes with many issues, for example, if package A documents itself with package B as some examples, the documentation might be incomplete when using it with package C. The code might not even work as intended. This is a possible solution to the correctness and documentation issue while maintaining the generic nature of the code. The Minecraft modpack community offers an insight. People use packs, not individual mods.
Make package bundles. Julia is composable, but it doesn’t mean the user always want to be the one who does the composition. Maybe even write some codes to glue everything together.
Test package bundles, The package behavior might not be tested on all possible contexts, but you can test them on cases, for this reason, even when the individual packages are tested, the complete bundle must be tested again.
Document package bundles. It’s not enough to document individual packages, but how they’re used together.
For example, a data science bundle might include a Dataframe package, Flux, with certain selected backends, and so on.
Since packages can be composed together in a variety of non-trivial ways, maybe documentation and correctness issue is solved by test and documentation of particular combination instead.
This solution, however, does have its negatives. It’s more work to be done, for example, even if they aren’t necessarily works from package authors.