[bartread]

Careful now: people will start accusing you of NIH.

But I fully agree with you. Likely what you need is a tiny subset of the capabilities wrapped up by the higher level abstraction, so implement them directly.

Over time you may find you need additional capabilities (although that's far from a given) and, if and when you do, you'll need to make decisions about whether to implement them directly, or wrap everything in a higher level abstraction (or use a third party abstraction).

The point is that if you ever do need these additional capabilities there's a good chance it's because you've been successful enough to enter "good problem to have" territory because you didn't waste time getting distracted by them earlier on and instead chose to focus on work that enabled that success.

[marginalia_nu]

If you wanna NIH you can just build your own docker. It's just an abstraction around some newer syscalls for process isolation. There's really not much magic to be found if you look into how it's done.

You can probably have a working prototype up in a weekend if you've got some systems programming experience.

[bartread]

I'm sure this has been on HN more than once but here's Bocker as an example of what you're talking about: https://news.ycombinator.com/item?id=42224670