BufWriter's inner does not need to be an Option

[kestein]

In the implementation of BufWriter the option type for the inner writer does not seem to be necessary. There is no way to initialize the inner writer to be None and the value is unwrap()'d anywhere the value of it is needed.

[TimNN]

The Option is necessary to allow for an into_inner implementation in the presence of a Drop implementation.

(The Option is set to None during the take in into_inner).

[apasel422]

@TimNN into_inner could be rewritten to use ptr::read and mem::forget.

[TimNN]

@apasel422: Yeah, you're right, so let me rephrase: An Option is necessary for a safe into_inner implementation (in the presence of a Drop implementation).

And personally I don't think the unsafe is worth it here, since whatever cost is associated with the Option is probably negligible when compared to the cost of the actual I/O (and, IMO, the less unsafe the better).

[hanna-kruppe]

A benchmark would be good, but if there is any measurable gain to be had, then this is exactly (one of the reasons) why unsafe exists. It might be sad that this pattern can't be expressed safely, but that's a grievance with the language, not a mark against the pattern, modulo the practical concerns of how sure one is about the correctness of the implementation. But I'm not super worried about the safety since the unsafe code would be rather small, simple, and contained.

[frewsxcv]

Anyone have ideas for a benchmark for this?

[Mark-Simulacrum]

I'm going to go ahead and close as this is a pretty ugly change based on the couple PRs that tried to fix this and the additional complexity isn't really worth it (no benchmarks are able to show significant wins).