feat: support stream decoder

.gitignore

@@ -8,3 +8,5 @@ rust-
 *.data
 opensource_git_commit.log
 Cargo.lock
+*.profraw
+*.profdata

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "sonic-rs"
-version = "0.3.10"
+version = "0.3.11"
 authors = ["Volo Team <[email protected]>"]
 edition = "2021"
 description = "Sonic-rs is a fast Rust JSON library based on SIMD"
@@ -15,7 +15,6 @@ categories = ["encoding", "parser-implementations"]
 
 [dependencies]
 cfg-if = "1.0"
-arrayref = "0.3"
 serde = { version = "1.0", features = ["rc", "derive"] }
 itoa = "1.0"
 ryu = "1.0"
@@ -26,7 +25,6 @@ bytes = "1.4"
 thiserror = "1.0"
 simdutf8 = "0.1"
 parking_lot = "0.12"
-page_size = "0.6"
 
 
 [target.'cfg(not(target_env = "msvc"))'.dev-dependencies]
@@ -78,6 +76,10 @@ harness = false
 name = "get_from"
 harness = false
 
+[[bench]]
+name = "value_operator"
+harness = false
+
 [features]
 default = []
 

benches/value_operator.rs

@@ -0,0 +1,83 @@
+#[macro_use]
+extern crate criterion;
+use std::io::Read;
+
+use criterion::{criterion_group, BatchSize, Criterion};
+use sonic_rs::JsonValueTrait;
+
+fn bench_get(c: &mut Criterion) {
+    let core_ids = core_affinity::get_core_ids().unwrap();
+    core_affinity::set_for_current(core_ids[0]);
+
+    let mut data = Vec::new();
+    let root = env!("CARGO_MANIFEST_DIR").to_owned();
+    std::fs::File::open(root + concat!("/benches/testdata/twitter.json"))
+        .unwrap()
+        .read_to_end(&mut data)
+        .unwrap();
+
+    let sonic_value: sonic_rs::Value = sonic_rs::from_slice(&data).unwrap();
+    let serde_value: serde_json::Value = serde_json::from_slice(&data).unwrap();
+
+    assert_eq!(
+        sonic_value["statuses"][4]["entities"]["media"][0]["source_status_id_str"].as_str(),
+        Some("439430848190742528")
+    );
+    assert_eq!(
+        serde_value["statuses"][4]["entities"]["media"][0]["source_status_id_str"].as_str(),
+        Some("439430848190742528")
+    );
+
+    let mut group = c.benchmark_group("value");
+    group.bench_with_input("sonic-rs::value_get", &sonic_value, |b, data| {
+        b.iter_batched(
+            || data,
+            |value| {
+                let _ =
+                    value["statuses"][4]["entities"]["media"][0]["source_status_id_str"].as_str();
+            },
+            BatchSize::SmallInput,
+        )
+    });
+
+    group.bench_with_input("serde_json::value_get", &serde_value, |b, data| {
+        b.iter_batched(
+            || data,
+            |value| {
+                let _ =
+                    value["statuses"][4]["entities"]["media"][0]["source_status_id_str"].as_str();
+            },
+            BatchSize::SmallInput,
+        )
+    });
+
+    group.bench_with_input("sonic_rs::value_new", &sonic_value, |b, data| {
+        b.iter_batched(
+            || data,
+            |_value| {
+                let mut value = sonic_rs::Array::new();
+                for i in 0..100 {
+                    value.push(i);
+                }
+            },
+            BatchSize::SmallInput,
+        )
+    });
+
+    group.bench_with_input("serde_json::value_new", &serde_value, |b, data| {
+        b.iter_batched(
+            || data,
+            |_value| {
+                let mut value = serde_json::Value::Array(Vec::new());
+                let array = &mut value.as_array_mut().unwrap();
+                for i in 0..100 {
+                    array.push(serde_json::Value::from(i as f64));
+                }
+            },
+            BatchSize::SmallInput,
+        )
+    });
+}
+
+criterion_group!(benches, bench_get);
+criterion_main!(benches);

src/input.rs

@@ -9,7 +9,7 @@ use crate::{parser::as_str, util::private::Sealed};
 #[derive(Debug, Clone, Eq, PartialEq, Ord, PartialOrd, Hash)]
 pub enum JsonSlice<'de> {
     Raw(&'de [u8]),
-    FastStr(FastStr),
+    FastStr(FastStr), // note: FastStr maybe inlined and in the stack.
 }
 
 impl<'de> JsonSlice<'de> {

src/lazyvalue/iterator.rs

@@ -34,7 +34,14 @@ use crate::{
 ///     }
 /// }
 /// ```
-pub struct ObjectJsonIter<'de>(ObjectInner<'de>);
+pub struct ObjectJsonIter<'de> {
+    json: JsonSlice<'de>,
+    parser: Option<Parser<Read<'de>>>,
+    strbuf: Vec<u8>,
+    first: bool,
+    ending: bool,
+    check: bool,
+}
 
 /// A lazied iterator for JSON array text. It will parse the JSON when iterating.
 ///
@@ -63,26 +70,15 @@ pub struct ObjectJsonIter<'de>(ObjectInner<'de>);
 ///     }
 /// }
 /// ```
-pub struct ArrayJsonIter<'de>(ArrayInner<'de>);
-
-struct ObjectInner<'de> {
-    json: JsonSlice<'de>,
-    parser: Option<Parser<Read<'de>>>,
-    strbuf: Vec<u8>,
-    first: bool,
-    ending: bool,
-    check: bool,
-}
-
-struct ArrayInner<'de> {
+pub struct ArrayJsonIter<'de> {
     json: JsonSlice<'de>,
     parser: Option<Parser<Read<'de>>>,
     first: bool,
     ending: bool,
     check: bool,
 }
 
-impl<'de> ObjectInner<'de> {
+impl<'de> ObjectJsonIter<'de> {
     fn new(json: JsonSlice<'de>, check: bool) -> Self {
         Self {
             json,
@@ -131,7 +127,7 @@ impl<'de> ObjectInner<'de> {
     }
 }
 
-impl<'de> ArrayInner<'de> {
+impl<'de> ArrayJsonIter<'de> {
     fn new(json: JsonSlice<'de>, check: bool) -> Self {
         Self {
             json,
@@ -219,7 +215,7 @@ impl<'de> ArrayInner<'de> {
 /// }
 /// ```
 pub fn to_object_iter<'de, I: JsonInput<'de>>(json: I) -> ObjectJsonIter<'de> {
-    ObjectJsonIter(ObjectInner::new(json.to_json_slice(), true))
+    ObjectJsonIter::new(json.to_json_slice(), true)
 }
 
 /// Traverse the JSON array text through a lazy iterator. The JSON parsing will doing when
@@ -256,7 +252,7 @@ pub fn to_object_iter<'de, I: JsonInput<'de>>(json: I) -> ObjectJsonIter<'de> {
 /// }
 /// ```
 pub fn to_array_iter<'de, I: JsonInput<'de>>(json: I) -> ArrayJsonIter<'de> {
-    ArrayJsonIter(ArrayInner::new(json.to_json_slice(), true))
+    ArrayJsonIter::new(json.to_json_slice(), true)
 }
 
 /// Traverse the JSON text through a lazy object iterator. The JSON parsing will doing when
@@ -293,7 +289,7 @@ pub fn to_array_iter<'de, I: JsonInput<'de>>(json: I) -> ArrayJsonIter<'de> {
 /// }
 /// ```
 pub unsafe fn to_object_iter_unchecked<'de, I: JsonInput<'de>>(json: I) -> ObjectJsonIter<'de> {
-    ObjectJsonIter(ObjectInner::new(json.to_json_slice(), false))
+    ObjectJsonIter::new(json.to_json_slice(), false)
 }
 
 /// Traverse the JSON text through a lazy object iterator. The JSON parsing will doing when
@@ -327,22 +323,22 @@ pub unsafe fn to_object_iter_unchecked<'de, I: JsonInput<'de>>(json: I) -> Objec
 /// }
 /// ```
 pub unsafe fn to_array_iter_unchecked<'de, I: JsonInput<'de>>(json: I) -> ArrayJsonIter<'de> {
-    ArrayJsonIter(ArrayInner::new(json.to_json_slice(), false))
+    ArrayJsonIter::new(json.to_json_slice(), false)
 }
 
 impl<'de> Iterator for ObjectJsonIter<'de> {
     type Item = Result<(FastStr, LazyValue<'de>)>;
 
     fn next(&mut self) -> Option<Self::Item> {
-        self.0.next_entry_impl(self.0.check)
+        self.next_entry_impl(self.check)
     }
 }
 
 impl<'de> Iterator for ArrayJsonIter<'de> {
     type Item = Result<LazyValue<'de>>;
 
     fn next(&mut self) -> Option<Self::Item> {
-        self.0.next_elem_impl(self.0.check)
+        self.next_elem_impl(self.check)
     }
 }
 

src/lib.rs

@@ -42,6 +42,7 @@ pub use crate::pointer::{JsonPointer, PointerNode, PointerTree};
 pub use crate::serde::{
     from_slice, from_slice_unchecked, from_str, to_string, to_string_pretty, to_vec, to_vec_pretty,
     to_writer, to_writer_pretty, Deserializer, JsonNumberTrait, Number, RawNumber, Serializer,
+    StreamDeserializer,
 };
 #[doc(inline)]
 pub use crate::value::{

src/parser.rs

@@ -5,7 +5,6 @@ use std::{
     str::from_utf8_unchecked,
 };
 
-use arrayref::array_ref;
 use faststr::FastStr;
 use serde::de::{self, Expected, Unexpected};
 use smallvec::SmallVec;
@@ -1217,7 +1216,7 @@ where
         let reader = &mut self.read;
 
         while let Some(chunk) = reader.peek_n(64) {
-            let input = array_ref![chunk, 0, 64];
+            let input = unsafe { &*(chunk.as_ptr() as *const [_; 64]) };
             if let Some(count) = skip_container_loop(
                 input,
                 &mut prev_instring,
@@ -1291,7 +1290,7 @@ where
 
         // then we use simd to accelerate skipping space
         while let Some(chunk) = reader.peek_n(64) {
-            let chunk = array_ref![chunk, 0, 64];
+            let chunk = unsafe { &*(chunk.as_ptr() as *const [_; 64]) };
             let bitmap = unsafe { get_nonspace_bits(chunk) };
             if bitmap != 0 {
                 self.nospace_bits = bitmap;
@@ -1351,7 +1350,7 @@ where
 
         // then we use simd to accelerate skipping space
         while let Some(chunk) = reader.peek_n(64) {
-            let chunk = array_ref![chunk, 0, 64];
+            let chunk = unsafe { &*(chunk.as_ptr() as *const [_; 64]) };
             let bitmap = unsafe { get_nonspace_bits(chunk) };
             if bitmap != 0 {
                 self.nospace_bits = bitmap;