Improve speed of to_string for Classes

[hgzimmerman]

The prior code would reallocate for every item in the set that is pushed into the buffer. This should make one bigger allocation for the Vec<&str> (assuming it uses the size_hint? - worst case it makes a log_n number of allocations to accommodate the str references) and another for joining it into the final String with spaces.

[jstarry]

@hgzimmerman could you please add benchmark results to PRs with perf changes like this?

[hgzimmerman]

How do you suggest that they be integrated into the project?

This one would be easy enough to copy relevant code into a separate project, run the benchmark and throw the project away when done.
Or instead should a benchmarks directory (analogous to tests or examples) be created?
Or finally, just leave a benchmark as part of a tests or benchmarks module within the existing file?

While I await your response, I'll pursue the first option.

[jstarry]

Yeah I was suggesting running benchmarks locally and pasting the results (ideally in the PR description). I'll add a formal process for this soon

[hgzimmerman]

Benchmarks are ran using the nightly test::Bencher.
Tests are ran on Classes constructed from variations on the string: "class1 class2 class3 ...";
old indicates the present master.
new indicates the proposed change.

Tests were ran against sets of 1, 2, 5, and 10 classes as more are unlikely to be used under typical use cases.

---

Results

test tests::bench_new_1  ... bench:          32 ns/iter (+/- 2)
test tests::bench_new_2  ... bench:          37 ns/iter (+/- 4)
test tests::bench_new_5  ... bench:          50 ns/iter (+/- 5)
test tests::bench_new_10 ... bench:          67 ns/iter (+/- 8)

test tests::bench_old_1  ... bench:          30 ns/iter (+/- 3)
test tests::bench_old_2  ... bench:          46 ns/iter (+/- 4)
test tests::bench_old_5  ... bench:          84 ns/iter (+/- 3)
test tests::bench_old_10 ... bench:         159 ns/iter (+/- 13)

Analysis

The proposed change is within the margin of error in the 1 class case, while meaningfully faster for inputs larger than 1.
In the broader scope of things, this change makes little difference, likely saving only a ten-thousandth of a second per full-app redraw.

[jstarry]

Thanks for adding the benchmark results!