Otp application and libraries discussion

The difference is similar as in OS libraries vs applications (in context of system supervisor). Library is just set of functions grouped in the modules where application is something that can run on its own. Good example there can be OpenTelemetry API vs OpenTelemetry SDK where the first one is “library of functions” and second one is “independent application with own supervision tree”.

I understand the difference between them. My point on the article is that creating an “application” is an anti-pattern from the point of view of OTP that should be avoided because it reduces the control we have over the execution and how other parts of the system interact with it.

Yes, because 95% of the time when someone defines application, they wanted library instead. You rarely need to write full application for your purposes, in most cases exposing functions that spawn processes you need is much better than spawning that function for the user, but still there are use cases when it is desirable. For user convenience (like in my systemd library) or that we just need a mutable state for keeping info about rest of the system (like in telemetry, however I think that in future persistent_term may be used there) or that there is no reason for having multiple instances of the processes as it is “sidecar-like” application (like OpenTelemetry).

FYI, the article is not available anymore. In fact cevado.me is “available on GoDaddy Auctions”.

Interesting article about it at Reusable Elixir Libraries.

i’ve moved it to https://cevado.wtf/
i’ll make that post available again and link it here.

edit:
so a link to the original blog post of this topic
https://cevado.wtf/posts/otp-applications/

keep in mind that the tone that i’m using this blogpost is not a good one, very mandatory and shitting on peoples solutions. i was planning to rewrite it with a better tone and better language, but never had the time.

Thank you. I think your article should be developed to be one of the main sections of the Elixir Docs.

My sense has always been that the docs are missing treatment of these key structuring issues.

For example, the docs have a section “Library guidelines”, but I’m not aware of any discussion in the docs of the differences between a Library and an Application, or more importantly the guidance on how to structure as you provide.

I don’t feel sufficiently qualified to propose a docs PR, but I do think a great many people would appreciate having your observations and suggestions developed into a key component of the docs.

Even after many years of producing production systems your article succinctly calls out the issues I’m sure many of us have had to confront while thinking we must be naive or incompetent because there is no treatment of the issues in the docs.

Many thanks again, Regards …