borrow sonic-rs impl

Author: Brooooooklyn

.github/workflows/CI.yml

@@ -45,6 +45,30 @@ jobs:
       - name: Run tests
         run: cargo test
 
+  miri:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@stable
+        with:
+          targets: x86_64-unknown-linux-gnu
+          components: miri
+          toolchain: nightly
+      - uses: actions/setup-node@v5
+        with:
+          node-version: 22
+          cache: 'yarn'
+      - name: Install dependencies
+        run: yarn install
+      - name: Download fixtures
+        run: node download-fixtures.js
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Run miri
+        run: cargo miri test
+        env:
+          MIRIFLAGS: "-Zmiri-disable-isolation"
   bench:
     strategy:
       matrix:

Cargo.lock

@@ -17,7 +17,6 @@ name = "escape"
 path = "examples/escape.rs"
 
 [features]
-force_aarch64_neon = [] # Force use of neon implementation on aarch64
 codspeed = ["criterion2/codspeed"]
 
 [[bench]]
@@ -26,10 +25,13 @@ harness = false
 
 [dependencies]
 anyhow = "1"
+sonic-simd = "0.1"
+thiserror = "2"
 
 [dev-dependencies]
 criterion2 = "3"
 glob = "0.3"
+rand = "0.9"
 serde = "1"
 serde_json = "1"
 v_jsonescape = "0.7"

Cargo.toml

@@ -4,17 +4,7 @@
 ![docs.rs](https://img.shields.io/docsrs/json-escape-simd)
 [![CodSpeed Badge](https://img.shields.io/endpoint?url=https://codspeed.io/badge.json)](https://codspeed.io/napi-rs/json-escape-simd)
 
-Optimized SIMD routines for escaping JSON strings. This repository contains the `json-escape-simd` crate, comparison fixtures, and Criterion benches against commonly used alternatives.
-
-> [!IMPORTANT]
->
-> On aarch64 NEON hosts the available register width is **128** bits, which is narrower than the lookup table this implementation prefers. As a result the SIMD path may not outperform the generic fallback, which is reflected in the benchmark numbers below.
->
-> On some modern macOS devices with larger register numbers, the SIMD path may outperform the generic fallback, see the [M3 max benchmark](#apple-m3-max) below.
-
-> [!NOTE]
->
-> The `force_aarch64_neon` feature flag can be used to force use of the neon implementation on aarch64. This is useful for the benchmark.
+Optimized SIMD routines for escaping JSON strings. The implementation is from [sonic-rs](https://github.com/cloudwego/sonic-rs), we only take the string escaping part to avoid the abstraction overhead.
 
 ## Benchmarks
 

README.md

@@ -2,7 +2,10 @@ use std::{fs, hint::black_box};
 
 use criterion::{Criterion, criterion_group, criterion_main};
 
-use json_escape_simd::{escape, escape_generic};
+use generic::escape_generic;
+use json_escape_simd::escape;
+
+mod generic;
 
 fn get_rxjs_sources() -> Vec<String> {
     let rxjs_paths = glob::glob("node_modules/rxjs/src/**/*.ts").unwrap();
@@ -86,6 +89,16 @@ fn run_benchmarks(c: &mut Criterion, sources: &[String], prefix: &str) {
     });
 }
 
+fn short_string_benchmark(c: &mut Criterion) {
+    let sources = vec![
+        "Hello, world!".to_string(),
+        r#"abcdefghijklmnopqrstuvwxyz .*? hello world escape json string"#.to_string(),
+        "normal string 🥹".to_string(),
+        "中文 English 🚀 \n❓ 𝄞".to_string(),
+    ];
+    run_benchmarks(c, &sources, "short string");
+}
+
 fn rxjs_benchmark(c: &mut Criterion) {
     let sources = get_rxjs_sources();
     if !sources.is_empty() {
@@ -100,5 +113,10 @@ fn fixtures_benchmark(c: &mut Criterion) {
     }
 }
 
-criterion_group!(benches, rxjs_benchmark, fixtures_benchmark);
+criterion_group!(
+    benches,
+    short_string_benchmark,
+    rxjs_benchmark,
+    fixtures_benchmark
+);
 criterion_main!(benches);

benches/escape.rs

@@ -13,15 +13,6 @@ pub fn escape_generic<S: AsRef<str>>(s: S) -> String {
     unsafe { String::from_utf8_unchecked(result) }
 }
 
-#[inline]
-pub fn escape_into_generic<S: AsRef<str>>(s: S, output: &mut Vec<u8>) {
-    let s = s.as_ref();
-    let bytes = s.as_bytes();
-    output.push(b'"');
-    escape_inner(bytes, output);
-    output.push(b'"');
-}
-
 #[inline]
 // Slightly modified version of
 // <https://github.com/serde-rs/json/blob/d12e943590208da738c092db92c34b39796a2538/src/ser.rs#L2079>
@@ -140,49 +131,3 @@ pub(crate) static HEX_BYTES: [HexPair; 32] = [
     HexPair(b'1', b'e'),
     HexPair(b'1', b'f'),
 ];
-
-// Optimized escape table with 8-byte arrays for fast bulk writing
-// First element is the length of escape sequence, followed by the escape bytes
-pub(crate) static ESCAPE_TABLE: [(u8, [u8; 8]); 256] = {
-    let mut table = [(0u8, [0u8; 8]); 256];
-
-    // Control characters \u0000 - \u001f
-    table[0x00] = (6, *b"\\u0000\0\0");
-    table[0x01] = (6, *b"\\u0001\0\0");
-    table[0x02] = (6, *b"\\u0002\0\0");
-    table[0x03] = (6, *b"\\u0003\0\0");
-    table[0x04] = (6, *b"\\u0004\0\0");
-    table[0x05] = (6, *b"\\u0005\0\0");
-    table[0x06] = (6, *b"\\u0006\0\0");
-    table[0x07] = (6, *b"\\u0007\0\0");
-    table[0x08] = (2, *b"\\b\0\0\0\0\0\0");
-    table[0x09] = (2, *b"\\t\0\0\0\0\0\0");
-    table[0x0A] = (2, *b"\\n\0\0\0\0\0\0");
-    table[0x0B] = (6, *b"\\u000b\0\0");
-    table[0x0C] = (2, *b"\\f\0\0\0\0\0\0");
-    table[0x0D] = (2, *b"\\r\0\0\0\0\0\0");
-    table[0x0E] = (6, *b"\\u000e\0\0");
-    table[0x0F] = (6, *b"\\u000f\0\0");
-    table[0x10] = (6, *b"\\u0010\0\0");
-    table[0x11] = (6, *b"\\u0011\0\0");
-    table[0x12] = (6, *b"\\u0012\0\0");
-    table[0x13] = (6, *b"\\u0013\0\0");
-    table[0x14] = (6, *b"\\u0014\0\0");
-    table[0x15] = (6, *b"\\u0015\0\0");
-    table[0x16] = (6, *b"\\u0016\0\0");
-    table[0x17] = (6, *b"\\u0017\0\0");
-    table[0x18] = (6, *b"\\u0018\0\0");
-    table[0x19] = (6, *b"\\u0019\0\0");
-    table[0x1A] = (6, *b"\\u001a\0\0");
-    table[0x1B] = (6, *b"\\u001b\0\0");
-    table[0x1C] = (6, *b"\\u001c\0\0");
-    table[0x1D] = (6, *b"\\u001d\0\0");
-    table[0x1E] = (6, *b"\\u001e\0\0");
-    table[0x1F] = (6, *b"\\u001f\0\0");
-
-    // Special characters
-    table[0x22] = (2, *b"\\\"\0\0\0\0\0\0"); // "
-    table[0x5C] = (2, *b"\\\\\0\0\0\0\0\0"); // \
-
-    table
-};

src/generic.rs renamed to benches/generic.rs

@@ -1,12 +1,10 @@
 use std::fs;
 
-use json_escape_simd::{escape, escape_generic};
+use json_escape_simd::escape;
 
 fn main() {
     for fixture in get_rxjs_sources() {
         let encoded = escape(&fixture);
-        let encoded_fallback = escape_generic(&fixture);
-        assert_eq!(encoded, encoded_fallback);
         assert_eq!(encoded, sonic_rs::to_string(&fixture).unwrap());
         assert_eq!(encoded, serde_json::to_string(&fixture).unwrap());
     }