perf: Refactor underlying buffers for branchless access (#229)

* branch start, reimplement Repr for branchless string accesses, smaller code, faster in general

* update fuzzing framework

* get rid of our own asserts module, fix clippy

* more clippy and fmt fixes

* fix miri, ignore quickcheck when running miri

* re-enable tests in capacity.rs

* fix issue in test for 32-bit

* amortize growth for HeapBuffer

* cargo fmt

* respond to GitHub feedback, benchmark against compact_str v0.6

* delete the old repr

* cargo fmt and stop using saturating_div

* make the repr module private

* workflow cleanup

* cleanup comments and workflows

.github/workflows/cross_platform.yml

@@ -86,25 +86,6 @@ jobs:
           command: miri
           args: test --all-features --manifest-path=compact_str/Cargo.toml
 
-  # TODO(parkmycar) re-enable ARMv7 builds
-  # linux_arm7:
-  #   name: Linux ARMv7
-  #   runs-on: [self-hosted, linux, ARM]
-  #   steps:
-  #     - uses: actions/checkout@v2
-  #       name: Checkout Repo
-  #     - uses: actions-rs/toolchain@v1
-  #       name: Install Rust
-  #       with:
-  #         profile: minimal
-  #         toolchain: nightly
-  #         override: true
-  #     - uses: actions-rs/cargo@v1
-  #       name: cargo test
-  #       with:
-  #         command: test
-  #         args: --release --all-features --manifest-path=compact_str/Cargo.toml
-
   linux_32bit:
     name: Linux 32-bit
     runs-on: ubuntu-latest

.github/workflows/fuzz.yml

@@ -70,43 +70,6 @@ jobs:
         with:
           update_existing: true
 
-  # TODO(parkmycar): Fails in CI with "error: failed to run LLVM passes: unknown pass name 'sancov'"
-  # honggfuzz_x86_64:
-  #   name: honggfuzz [x86_64]
-  #   runs-on: ubuntu-latest
-  #   env:
-  #     HFUZZ_BUILD_ARGS: "--features=honggfuzz"
-  #     HFUZZ_RUN_ARGS: "--run_time 120 --exit_upon_crash"
-  #     RUSTFLAGS: "-Znew-llvm-pass-manager=no"
-  #   steps:
-  #     - uses: actions/checkout@v2
-  #       name: Checkout compact_str
-  #     - uses: actions-rs/toolchain@v1
-  #       name: Install Rust
-  #       with:
-  #         toolchain: nightly
-  #         override: true
-  #     - name: Install Deps
-  #       run: sudo apt-get install --no-install-recommends binutils-dev libunwind8-dev
-  #     - uses: actions-rs/cargo@v1
-  #       name: Install honggfuzz
-  #       with:
-  #         command: install
-  #         toolchain: nightly
-  #         args: honggfuzz
-  #     - uses: actions-rs/cargo@v1
-  #       name: Fuzz!
-  #       with:
-  #         command: hfuzz
-  #         args: run honggfuzz
-  #     - name: File Issue (if failure found)
-  #       if: failure()
-  #       uses: JasonEtco/create-an-issue@v2.6
-  #       env:
-  #         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-  #       with:
-  #         update_existing: true
-
   afl_x86_64:
     name: AFL++ [x86_64]
     runs-on: ubuntu-latest
@@ -177,58 +140,3 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
           update_existing: true
-
-  # TODO(parkmycar): ARMv7 runner is no longer online
-  # afl_armv7:
-  #   name: AFL++ [ARMv7]
-  #   runs-on: [self-hosted, linux, ARM]
-  #   steps:
-  #     - uses: actions/checkout@v2
-  #       name: Checkout compact_str
-  #     - uses: actions-rs/toolchain@v1
-  #       name: Install Rust
-  #       with:
-  #         profile: minimal
-  #         toolchain: nightly-2022-07-01
-  #         override: true
-  #     - name: Install cargo-afl
-  #       run: |
-  #         cargo +nightly install afl --force
-  #     - name: AFL++ Build
-  #       run: |
-  #         cd fuzz
-  #         cargo +nightly afl build --bin afl --release --features=afl
-  #     - name: Set Fuzz Time
-  #       run: |
-  #         if [[ "${{github.event_name}}" == "push" || "${{github.event_name}}" == "pull_request" ]]; then
-  #           echo "fuzz_time=120" >> $GITHUB_ENV
-  #         else
-  #           echo "fuzz_time=1800" >> $GITHUB_ENV
-  #         fi
-  #         echo "${{ env.fuzz_time }}"
-  #     - name: Fuzz!
-  #       run: |
-  #         cd fuzz
-  #         mkdir afl/out
-  #         cargo +nightly afl fuzz -i afl/in -o afl/out -D -V ${{ env.fuzz_time }} ../target/release/afl
-  #     - name: Check for Crashes
-  #       run: |
-  #         if [ "$(ls -1q ./fuzz/afl/out/default/crashes/ | wc -l)" -ne 0 ]; then exit 1; fi
-  #     # AFL generates filenames with ":", which upload-artifact fails on, so we need to "detox" them
-  #     - name: Sanitize Crash Filenames (if present)
-  #       if: failure()
-  #       run: |
-  #         detox -r -v ./fuzz/afl/out/default/crashes/
-  #     - name: Upload Crashes (if present)
-  #       if: failure()
-  #       uses: actions/upload-artifact@v2
-  #       with:
-  #         name: Upload Crashes
-  #         path: ./fuzz/afl/out/default/crashes/
-  #     - name: File Issue (if failure found)
-  #       if: failure()
-  #       uses: JasonEtco/create-an-issue@v2.6
-  #       env:
-  #         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-  #       with:
-  #         update_existing: true

bench/Cargo.toml

@@ -7,6 +7,7 @@ edition = "2021"
 
 [dependencies]
 compact_str = { path = "../compact_str" }
+compact_str_6 = { package = "compact_str", version = "0.6" }
 criterion = { version = "0.3", features = ["html_reports"] }
 iai = "0.1.1"
 smartstring = "1.0.1"

bench/benches/compact_str.rs

@@ -2,6 +2,7 @@ use compact_str::{
     CompactString,
     ToCompactString,
 };
+use compact_str_6::CompactString as CompactString6;
 use criterion::{
     criterion_group,
     criterion_main,
@@ -31,7 +32,7 @@ fn bench_new(c: &mut Criterion) {
     c.bench_with_input(
         BenchmarkId::new("CompactString::new", "59 chars"),
         &"I am a very long string that will get allocated on the heap",
-        |b, word| b.iter(|| CompactString::new(word)),
+        |b, &word| b.iter(|| CompactString::from(word)),
     );
 
     c.bench_with_input(
@@ -91,5 +92,67 @@ fn bench_to_compact_string(c: &mut Criterion) {
     );
 }
 
+fn bench_repr_creation(c: &mut Criterion) {
+    let mut group = c.benchmark_group("Creation");
+
+    let words: Vec<String> = vec![0, 11, 12, 22, 23, 24, 25, 50]
+        .into_iter()
+        .map(|len| (0..len).into_iter().map(|_| 'a').collect())
+        .collect();
+
+    for word in words {
+        group.bench_with_input(
+            BenchmarkId::new("CompactString6", word.len()),
+            &word,
+            |b, w| b.iter(|| CompactString6::new(w)),
+        );
+
+        group.bench_with_input(
+            BenchmarkId::new("CompactString", word.len()),
+            &word,
+            |b, w| b.iter(|| CompactString::new(w)),
+        );
+
+        group.bench_with_input(
+            BenchmarkId::new("std::String", word.len()),
+            &word,
+            |b, w| b.iter(|| String::from(w)),
+        );
+    }
+}
+
+fn bench_repr_access(c: &mut Criterion) {
+    let mut group = c.benchmark_group("Access");
+
+    let words: Vec<String> = vec![0, 11, 12, 23, 24, 50]
+        .into_iter()
+        .map(|len| (0..len).into_iter().map(|_| 'a').collect())
+        .collect();
+
+    for word in words {
+        let compact = CompactString6::new(&word);
+        group.bench_with_input(
+            BenchmarkId::new("CompactString6", compact.len()),
+            &compact,
+            |b, c| b.iter(|| c.as_str()),
+        );
+
+        let compact = CompactString::new(&word);
+        group.bench_with_input(
+            BenchmarkId::new("CompactString", compact.len()),
+            &compact,
+            |b, c| b.iter(|| c.as_str()),
+        );
+
+        let std_str = String::from(&word);
+        group.bench_with_input(
+            BenchmarkId::new("String", std_str.len()),
+            &std_str,
+            |b, s| b.iter(|| s.as_str()),
+        );
+    }
+}
+criterion_group!(repr_benches, bench_repr_creation, bench_repr_access);
+
 criterion_group!(compact_str, bench_new, bench_to_compact_string);
-criterion_main!(compact_str);
+criterion_main!(compact_str, repr_benches);

compact_str/Cargo.toml

@@ -21,16 +21,20 @@ rkyv = { version = "0.7", optional = true }
 serde = { version = "1", optional = true }
 
 castaway = "0.2.1"
+cfg-if = "1"
 itoa = "1"
 ryu = "1"
+static_assertions = "1"
 
 [dev-dependencies]
 cfg-if = "1"
 proptest = { version = "1", default-features = false, features = ["std"] }
+quickcheck = "1"
 quickcheck_macros = "1"
 rayon = "1"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
+test-case = "2"
 test-strategy = "0.2"
 
 [package.metadata.docs.rs]

compact_str/src/features/bytes.rs

@@ -37,7 +37,7 @@ impl CompactString {
     /// ```
     #[cfg_attr(docsrs, doc(cfg(feature = "bytes")))]
     pub fn from_utf8_buf<B: Buf>(buf: &mut B) -> Result<Self, Utf8Error> {
-        Repr::from_utf8_buf(buf).map(|repr| CompactString { repr })
+        Repr::from_utf8_buf(buf).map(CompactString)
     }
 
     /// Converts a buffer of bytes to a [`CompactString`], without checking that the provided buffer
@@ -63,7 +63,7 @@ impl CompactString {
     #[cfg_attr(docsrs, doc(cfg(feature = "bytes")))]
     pub unsafe fn from_utf8_buf_unchecked<B: Buf>(buf: &mut B) -> Self {
         let repr = Repr::from_utf8_buf_unchecked(buf);
-        CompactString { repr }
+        CompactString(repr)
     }
 }