Best-effort build

Cargo.toml

@@ -61,10 +61,9 @@ unstable-simd = [
     "simdutf8/aarch64_neon",
 ]
 
-# Build yyjson as a backend.
-yyjson = [
-    "cc",
-]
+# Build yyjson as a backend and panic if it fails. The default is to attempt
+# to build and on failure fall back to another backend.
+yyjson = []
 
 [dependencies]
 ahash = { version = "0.7", default_features = false }
@@ -85,8 +84,9 @@ simdutf8 = { version = "0.1", default_features = false, features = ["std"] }
 smallvec = { version = "^1.8", default_features = false, features = ["union", "write"] }
 
 [build-dependencies]
-cc = { version = "1", optional = true  }
+cc = { version = "1" }
 pyo3-build-config = "^0.16.5"
+version_check = { version = "0.9" }
 
 [profile.release]
 codegen-units = 1

README.md

@@ -1146,19 +1146,16 @@ Probably not.
 ## Packaging
 
 To package orjson requires at least [Rust](https://www.rust-lang.org/) 1.54
-and the [maturin](https://github.com/PyO3/maturin) build tool. It benefits
-from also having `clang`. The recommended build command is:
+and the [maturin](https://github.com/PyO3/maturin) build tool. The recommended
+build command is:
 
 ```sh
-maturin build --release --strip --features=yyjson
+maturin build --release --strip
 ```
 
-To build without use of `clang`, do not specify `--features=yyjson`.
-Deserialization is much faster if built with this feature.
-
-There is a minor performance benefit on at least amd64 to building on `nightly`
-with `--features=unstable-simd`. It may be more significant on other
-architectures.
+It benefits from also having a C build environment to compile a faster
+deserialization backend. See this project's `manylinux_2_28` builds for an
+example using clang and LTO.
 
 The project's own CI tests against `nightly-2022-06-22` and stable 1.54. It
 is prudent to pin the nightly version because that channel can introduce

build.rs

@@ -1,19 +1,44 @@
 // SPDX-License-Identifier: (Apache-2.0 OR MIT)
 
-#[allow(dead_code)]
-#[cfg(feature = "yyjson")]
-fn build_yyjson() {
-    cc::Build::new()
-        .file("include/yyjson/yyjson.c")
-        .include("include/yyjson")
-        .define("YYJSON_DISABLE_WRITER", "1")
-        .define("YYJSON_DISABLE_NON_STANDARD", "1")
-        .compile("yyjson");
-}
-
 fn main() {
+    println!("cargo:rerun-if-changed=build.rs");
+    println!("cargo:rerun-if-changed=include/yyjson/*");
+    println!("cargo:rerun-if-env-changed=CC");
+    println!("cargo:rerun-if-env-changed=CFLAGS");
+    println!("cargo:rerun-if-env-changed=LDFLAGS");
+    println!("cargo:rerun-if-env-changed=RUSTFLAGS");
+    println!("cargo:rerun-if-env-changed=ORJSON_DISABLE_YYJSON");
+
     pyo3_build_config::use_pyo3_cfgs();
 
-    #[cfg(feature = "yyjson")]
-    build_yyjson();
+    if let Some(true) = version_check::supports_feature("core_intrinsics") {
+        println!("cargo:rustc-cfg=feature=\"intrinsics\"");
+    }
+
+    if let Some(true) = version_check::supports_feature("optimize_attribute") {
+        println!("cargo:rustc-cfg=feature=\"optimize\"");
+    }
+
+    if std::env::var("ORJSON_DISABLE_YYJSON").is_ok() {
+        if std::env::var("CARGO_FEATURE_YYJSON").is_ok() {
+            panic!("ORJSON_DISABLE_YYJSON and --features=yyjson both enabled.")
+        }
+    } else {
+        match cc::Build::new()
+            .file("include/yyjson/yyjson.c")
+            .include("include/yyjson")
+            .define("YYJSON_DISABLE_WRITER", "1")
+            .define("YYJSON_DISABLE_NON_STANDARD", "1")
+            .try_compile("yyjson")
+        {
+            Ok(_) => {
+                println!("cargo:rustc-cfg=feature=\"yyjson\"");
+            }
+            Err(_) => {
+                if std::env::var("CARGO_FEATURE_YYJSON").is_ok() {
+                    panic!("yyjson was enabled but the build failed. To build with a different backend do not specify the feature.")
+                }
+            }
+        }
+    }
 }

src/deserialize/utf8.rs

@@ -16,16 +16,11 @@ fn is_valid_utf8(buf: &[u8]) -> bool {
     }
 }
 
-#[cfg(all(target_arch = "aarch64", feature = "unstable-simd"))]
+#[cfg(target_arch = "aarch64")]
 fn is_valid_utf8(buf: &[u8]) -> bool {
     simdutf8::basic::from_utf8(buf).is_ok()
 }
 
-#[cfg(all(target_arch = "aarch64", not(feature = "unstable-simd")))]
-fn is_valid_utf8(buf: &[u8]) -> bool {
-    std::str::from_utf8(buf).is_ok()
-}
-
 #[cfg(not(any(target_arch = "x86_64", target_arch = "aarch64")))]
 fn is_valid_utf8(buf: &[u8]) -> bool {
     std::str::from_utf8(buf).is_ok()

src/lib.rs

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: (Apache-2.0 OR MIT)
 
-#![cfg_attr(feature = "unstable-simd", feature(core_intrinsics))]
-#![cfg_attr(feature = "unstable-simd", feature(optimize_attribute))]
+#![cfg_attr(feature = "intrinsics", feature(core_intrinsics))]
+#![cfg_attr(feature = "optimize", feature(optimize_attribute))]
 #![allow(unused_unsafe)]
 #![allow(clippy::missing_safety_doc)]
 #![allow(clippy::redundant_field_names)]
@@ -60,7 +60,7 @@ macro_rules! opt {
 #[allow(non_snake_case)]
 #[no_mangle]
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 pub unsafe extern "C" fn orjson_init_exec(mptr: *mut PyObject) -> c_int {
     typeref::init_typerefs();
     {
@@ -193,7 +193,7 @@ pub unsafe extern "C" fn orjson_init_exec(mptr: *mut PyObject) -> c_int {
 #[allow(non_snake_case)]
 #[no_mangle]
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 pub unsafe extern "C" fn PyInit_orjson() -> *mut PyModuleDef {
     let mod_slots: Box<[PyModuleDef_Slot; 2]> = Box::new([
         PyModuleDef_Slot {
@@ -224,7 +224,7 @@ pub unsafe extern "C" fn PyInit_orjson() -> *mut PyModuleDef {
 
 #[cold]
 #[inline(never)]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 fn raise_loads_exception(err: deserialize::DeserializeError) -> *mut PyObject {
     let pos = err.pos();
     let msg = err.message;
@@ -246,7 +246,7 @@ fn raise_loads_exception(err: deserialize::DeserializeError) -> *mut PyObject {
 
 #[cold]
 #[inline(never)]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 fn raise_dumps_exception(msg: Cow<str>) -> *mut PyObject {
     unsafe {
         let err_msg =

src/serialize/error.rs

@@ -23,7 +23,7 @@ pub enum SerializeError {
 
 impl std::fmt::Display for SerializeError {
     #[cold]
-    #[cfg_attr(feature = "unstable-simd", optimize(size))]
+    #[cfg_attr(feature = "optimize", optimize(size))]
     fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
         match *self {
             SerializeError::DatetimeLibraryUnsupported => write!(f, "datetime's timezone library is not supported: use datetime.timezone.utc, pendulum, pytz, or dateutil"),

src/serialize/numpy.rs

@@ -662,7 +662,7 @@ pub enum NumpyDatetimeUnit {
 
 impl fmt::Display for NumpyDatetimeUnit {
     #[cold]
-    #[cfg_attr(feature = "unstable-simd", optimize(size))]
+    #[cfg_attr(feature = "optimize", optimize(size))]
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         let unit = match self {
             Self::NaT => "NaT",
@@ -693,7 +693,7 @@ enum NumpyDateTimeError {
 
 impl NumpyDateTimeError {
     #[cold]
-    #[cfg_attr(feature = "unstable-simd", optimize(size))]
+    #[cfg_attr(feature = "optimize", optimize(size))]
     fn into_serde_err<T: ser::Error>(self) -> T {
         let err = match self {
             Self::UnsupportedUnit(unit) => format!("unsupported numpy.datetime64 unit: {}", unit),

src/serialize/serializer.rs

@@ -101,7 +101,7 @@ macro_rules! is_subclass {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 #[inline(never)]
 pub fn pyobject_to_obtype_unlikely(obj: *mut pyo3_ffi::PyObject, opts: Opt) -> ObType {
     unsafe {

src/typeref.rs

@@ -104,7 +104,7 @@ pub static mut JsonDecodeError: *mut PyObject = 0 as *mut PyObject;
 static INIT: Once = Once::new();
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 pub fn init_typerefs() {
     INIT.call_once(|| unsafe {
         assert!(crate::deserialize::KEY_MAP
@@ -171,7 +171,7 @@ pub fn init_typerefs() {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_json_exc() -> *mut PyObject {
     let module = PyImport_ImportModule("json\0".as_ptr() as *const c_char);
     let module_dict = PyObject_GenericGetDict(module, std::ptr::null_mut());
@@ -190,7 +190,7 @@ unsafe fn look_up_json_exc() -> *mut PyObject {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_numpy_type(numpy_module: *mut PyObject, np_type: &str) -> *mut PyTypeObject {
     let mod_dict = PyObject_GenericGetDict(numpy_module, std::ptr::null_mut());
     let ptr = PyMapping_GetItemString(mod_dict, np_type.as_ptr() as *const c_char);
@@ -200,7 +200,7 @@ unsafe fn look_up_numpy_type(numpy_module: *mut PyObject, np_type: &str) -> *mut
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn load_numpy_types() -> Option<NumpyTypes> {
     let numpy = PyImport_ImportModule("numpy\0".as_ptr() as *const c_char);
     if numpy.is_null() {
@@ -226,7 +226,7 @@ unsafe fn load_numpy_types() -> Option<NumpyTypes> {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_field_type() -> NonNull<PyObject> {
     let module = PyImport_ImportModule("dataclasses\0".as_ptr() as *const c_char);
     let module_dict = PyObject_GenericGetDict(module, std::ptr::null_mut());
@@ -238,7 +238,7 @@ unsafe fn look_up_field_type() -> NonNull<PyObject> {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_enum_type() -> *mut PyTypeObject {
     let module = PyImport_ImportModule("enum\0".as_ptr() as *const c_char);
     let module_dict = PyObject_GenericGetDict(module, std::ptr::null_mut());
@@ -250,7 +250,7 @@ unsafe fn look_up_enum_type() -> *mut PyTypeObject {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_uuid_type() -> *mut PyTypeObject {
     let uuid_mod = PyImport_ImportModule("uuid\0".as_ptr() as *const c_char);
     let uuid_mod_dict = PyObject_GenericGetDict(uuid_mod, std::ptr::null_mut());
@@ -263,7 +263,7 @@ unsafe fn look_up_uuid_type() -> *mut PyTypeObject {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_datetime_type() -> *mut PyTypeObject {
     let datetime = ((*PyDateTimeAPI()).DateTime_FromDateAndTime)(
         1970,
@@ -282,7 +282,7 @@ unsafe fn look_up_datetime_type() -> *mut PyTypeObject {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_date_type() -> *mut PyTypeObject {
     let date = ((*PyDateTimeAPI()).Date_FromDate)(1, 1, 1, (*(PyDateTimeAPI())).DateType);
     let ptr = (*date).ob_type;
@@ -291,7 +291,7 @@ unsafe fn look_up_date_type() -> *mut PyTypeObject {
 }
 
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_time_type() -> *mut PyTypeObject {
     let time = ((*PyDateTimeAPI()).Time_FromTime)(0, 0, 0, 0, NONE, (*(PyDateTimeAPI())).TimeType);
     let ptr = (*time).ob_type;
@@ -301,7 +301,7 @@ unsafe fn look_up_time_type() -> *mut PyTypeObject {
 
 #[cfg(Py_3_9)]
 #[cold]
-#[cfg_attr(feature = "unstable-simd", optimize(size))]
+#[cfg_attr(feature = "optimize", optimize(size))]
 unsafe fn look_up_zoneinfo_type() -> *mut PyTypeObject {
     let module = PyImport_ImportModule("zoneinfo\0".as_ptr() as *const c_char);
     let module_dict = PyObject_GenericGetDict(module, std::ptr::null_mut());

src/util.rs

@@ -18,14 +18,14 @@ macro_rules! err {
     };
 }
 
-#[cfg(feature = "unstable-simd")]
+#[cfg(feature = "intrinsics")]
 macro_rules! unlikely {
     ($exp:expr) => {
         core::intrinsics::unlikely($exp)
     };
 }
 
-#[cfg(not(feature = "unstable-simd"))]
+#[cfg(not(feature = "intrinsics"))]
 macro_rules! unlikely {
     ($exp:expr) => {
         $exp