use a bump allocator again

BENCHMARKING.md

@@ -170,3 +170,24 @@ Summary
 ```
 
 Yep!
+
+## Bump Allocation (again)
+
+Let's try the bump allocator again... it seemed like a fluke that it produced such a severe performance degradation.
+
+```
+$ hyperfine './result/bin/similar-sort define < /usr/share/dict/words' './target/release/similar-sort benchmark < /usr/share/dict/words'
+Benchmark #1: ./result/bin/similar-sort define < /usr/share/dict/words
+  Time (mean ± σ):     286.2 ms ±   5.0 ms    [User: 251.4 ms, System: 72.5 ms]
+  Range (min … max):   280.8 ms … 298.4 ms    10 runs
+
+Benchmark #2: ./target/release/similar-sort benchmark < /usr/share/dict/words
+  Time (mean ± σ):     111.3 ms ±   3.4 ms    [User: 94.7 ms, System: 16.0 ms]
+  Range (min … max):   104.1 ms … 118.5 ms    24 runs
+
+Summary
+  './target/release/similar-sort benchmark < /usr/share/dict/words' ran
+    2.57 ± 0.09 times faster than './result/bin/similar-sort define < /usr/share/dict/words'
+```
+
+That seems more like what I'd expect!

Cargo.toml

@@ -7,7 +7,8 @@ edition = "2018"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-structopt = "0.3.22"
+bump_alloc = "0.1.0"
 color-eyre = "0.5.7"
 rayon = "1.5.1"
 strsim = "0.10.0"
+structopt = "0.3.22"

src/main.rs

@@ -4,6 +4,9 @@ use std::io::{self, stdin, stdout, BufRead, BufWriter, Write};
 use strsim::levenshtein;
 use structopt::StructOpt;
 
+#[global_allocator]
+static A: bump_alloc::BumpAlloc = bump_alloc::BumpAlloc::new();
+
 // works like `sort`, but sorts according to Levenshtein distance instead of
 // alphanumerically.
 #[derive(StructOpt)]
@@ -42,6 +45,7 @@ fn try_main() -> Result<()> {
     for (_, candidate) in distances {
         writeln!(out, "{}", candidate).context("could not write to stdout")?;
     }
+    out.flush().context("could not finish writing to stdout")?;
 
     Ok(())
 }