Unfortunately, this can't really be done efficiently with how allocation works in Rust. A String is always allocated with alignment 1, but an Arc's control block requires usize alignment. An allocated pointer can't be deallocated with a different alignment than it was allocated with, nor can its alignment be changed when it is reallocated. Thus, the from_raw_parts() idea would be unusable for any types with less than usize alignment.

(And the control block really does need to be aligned, since otherwise we couldn't perform atomic operations on it.)

Something that is optimal under some set of assumptions and under some specific conditions is likely not optimal under all conditions. Does that mean it's not optimal under any conditions? If not, is something that is optimal under more conditions not "more optimal"?

I love Cow, except when it goes mut.

I presume they're trying to avoid having to worry about the lifetime, so they want something that owns its contents.

From 1983:

Although "optimum" is an absolute term, like "unique", it became common verbal practice to make it relative: "not quite optimum" or "less optimum" or "not very optimum". Mel called the maximum time-delay locations the "most pessimum".

https://users.cs.utah.edu/~elb/folklore/mel-annotated/node1....

I prefer "more betterer".

The typical strategy for doing that in a context like this is string interning. That's literally what most of the types described in the post are doing behind the scenes, in one form or another. The post is literally about doing your suggestion -- just using a library to do so, and leveraging the type system to reduce some of the boilerplate.