Build the engine once for every elixir+otp combination

[doorgan]

Currently if you have a workspace with multiple workspace folders and all of them use the same elixir+otp versions, we will try to build the engine for each one of them.
In other words, if your workspace has 3 folders, 3 engines will be built

To reduce work and speed up statup times, we should only try to build the engine once for each elixir+otp pair during the same session, and cache the engine path for any subsequent engine restart.
If we try to build the same engine twice at the same time, only a single build should run and both engines should use the results of that single run.

[mkaput]

correct me if I miss something; but this sounds to me like some kind of garbage collection should be considered to do as a followup, right?

[doorgan]

@mkaput can you elaborate? I'm not sure what you're referring to

[mkaput]

Ok, that's how I understand things:

Now

Expert builds the engine, identified with tuple (workspace folder, elixir version, otp version) and keep it in .expert directory

Goal

Identify engine builds with only (elixir version, otp version) and keep in a global directory, shared among workspaces

---

My concern is about the accumulation of old engine builds that are no longer used. The (elixir version, otp version) pair is an infinitely growing property => as users use Expert over time, this might happen to be a problem for some (nee node_modules or Rust's target/ stories).

Theoretically, reducing tuple arity from 3 to 2 should lower the rate of disk usage growth. But scoping engine builds in project directories had a self-cleaning property, in that users may delete long-untouched projects on their machines. By keeping all engine builds in a single place, Expert loses this property.

Perhaps I'm overreacting; that's plausible.