build(profiling): remove once_cell as a dependency (#3607)

Since Rust 1.70 there are types in core we can use, or in some cases
we can avoid the type and rely on get_module as a place to run the
bits that are non-const.

Author: morrisonlevi

Cargo.lock

@@ -31,7 +31,6 @@ lazy_static = { version = "1.4" }
 libc = "0.2"
 # TRACE set to max to support runtime configuration.
 log = { version = "0.4", features = ["max_level_trace", "release_max_level_trace"]}
-once_cell = { version = "1.12" }
 serde_json = {version = "1.0"}
 rand = { version = "0.8.5" }
 rand_distr = { version = "0.4.3" }

profiling/Cargo.toml

@@ -246,13 +246,16 @@ pub struct ZendExtension {
 
 pub use ZendExtension as zend_extension;
 
-impl Default for ModuleEntry {
-    fn default() -> Self {
+impl ModuleEntry {
+    /// Creates a new ModuleEntry with default values for const-compatible fields.
+    /// Non-const fields (functions, build_id) are set to null and should be initialized separately.
+    #[allow(clippy::new_without_default)]
+    pub const fn new() -> Self {
         Self {
-            size: std::mem::size_of::<Self>() as c_ushort,
+            size: core::mem::size_of::<Self>() as c_ushort,
             zend_api: ZEND_MODULE_API_NO,
-            zend_debug: ZEND_DEBUG as u8,
-            zts: USING_ZTS as u8,
+            zend_debug: ZEND_DEBUG as c_uchar,
+            zts: USING_ZTS as c_uchar,
             ini_entry: ptr::null(),
             deps: ptr::null(),
             name: c"".as_ptr(),
@@ -264,20 +267,18 @@ impl Default for ModuleEntry {
             info_func: None,
             version: ptr::null(),
             globals_size: 0,
-
             #[cfg(php_zts)]
             globals_id_ptr: ptr::null_mut(),
             #[cfg(not(php_zts))]
             globals_ptr: ptr::null_mut(),
-
             globals_ctor: None,
             globals_dtor: None,
             post_deactivate_func: None,
             module_started: 0,
             type_: MODULE_PERSISTENT as c_uchar,
             handle: ptr::null_mut(),
             module_number: -1,
-            build_id: unsafe { datadog_module_build_id() },
+            build_id: ptr::null(),
         }
     }
 }

profiling/src/bindings/mod.rs

@@ -34,14 +34,13 @@ use core::ptr;
 use lazy_static::lazy_static;
 use libdd_common::{cstr, tag, tag::Tag};
 use log::{debug, error, info, trace, warn};
-use once_cell::sync::{Lazy, OnceCell};
 use profiling::{LocalRootSpanResourceMessage, Profiler, VmInterrupt};
 use sapi::Sapi;
 use std::borrow::Cow;
 use std::cell::{BorrowError, BorrowMutError, RefCell};
 use std::ops::{Deref, DerefMut};
 use std::sync::atomic::{AtomicBool, AtomicU32, Ordering};
-use std::sync::{Arc, Once};
+use std::sync::{Arc, Once, OnceLock};
 use std::thread::{AccessError, LocalKey};
 use std::time::{Duration, Instant};
 use uuid::Uuid;
@@ -146,7 +145,7 @@ lazy_static! {
 /// Additionally, the tracer is going to ask for this in its ACTIVATE handler,
 /// so whatever it is replaced with needs to also follow the
 /// initialize-on-first-use pattern.
-static RUNTIME_ID: OnceCell<Uuid> = OnceCell::new();
+static RUNTIME_ID: OnceLock<Uuid> = OnceLock::new();
 // If ddtrace is loaded, we fetch the uuid from there instead
 extern "C" {
     pub static ddtrace_runtime_id: *const Uuid;
@@ -159,55 +158,62 @@ extern "C" {
 #[cfg(php_zts)]
 static mut GLOBALS_ID_PTR: i32 = 0;
 
-/// The function `get_module` is what makes this a PHP module. Please do not
-/// call this directly; only let it be called by the engine. Generally it is
-/// only called once, but if someone accidentally loads the module twice then
-/// it might get called more than once, though it will warn and not use the
-/// consecutive return value.
+/// Module dependencies for the profiler extension.
+static MODULE_DEPS: [zend::ModuleDep; 8] = [
+    zend::ModuleDep::required(cstr!("standard")),
+    zend::ModuleDep::required(cstr!("json")),
+    zend::ModuleDep::optional(cstr!("ddtrace")),
+    // Optionally, be dependent on these event extensions so that the functions they provide
+    // are registered in the function table and we can hook into them.
+    zend::ModuleDep::optional(cstr!("ev")),
+    zend::ModuleDep::optional(cstr!("event")),
+    zend::ModuleDep::optional(cstr!("libevent")),
+    zend::ModuleDep::optional(cstr!("uv")),
+    zend::ModuleDep::end(),
+];
+
+/// The module entry for the profiler extension. Fields that aren't
+/// const-compatible are set in get_module().
+static mut MODULE: zend::ModuleEntry = zend::ModuleEntry {
+    deps: MODULE_DEPS.as_ptr(),
+    name: PROFILER_NAME.as_ptr(),
+    functions: ptr::null(), // Will be set in get_module()
+    module_startup_func: Some(minit),
+    module_shutdown_func: Some(mshutdown),
+    request_startup_func: Some(rinit),
+    request_shutdown_func: Some(rshutdown),
+    info_func: Some(minfo),
+    version: PROFILER_VERSION.as_ptr(),
+    globals_size: 1,
+    #[cfg(php_zts)]
+    globals_id_ptr: ptr::addr_of_mut!(GLOBALS_ID_PTR),
+    #[cfg(not(php_zts))]
+    globals_ptr: ptr::null_mut(),
+    globals_ctor: Some(ginit),
+    globals_dtor: Some(gshutdown),
+    post_deactivate_func: Some(prshutdown),
+    build_id: ptr::null(), // Will be set in get_module()
+    ..zend::ModuleEntry::new()
+};
+
+/// The function `get_module` is what makes this a PHP module.
+///
+/// # Safety
+///
+/// Do not call this function manually; it will be called by the engine.
+/// Generally it is  only called once, but if someone accidentally loads the
+/// module twice then it might get called more than once, though it will warn
+/// and not use the consecutive return value.
 #[no_mangle]
-pub extern "C" fn get_module() -> &'static mut zend::ModuleEntry {
-    static DEPS: [zend::ModuleDep; 8] = [
-        zend::ModuleDep::required(cstr!("standard")),
-        zend::ModuleDep::required(cstr!("json")),
-        zend::ModuleDep::optional(cstr!("ddtrace")),
-        // Optionally, be dependent on these event extensions so that the functions they provide
-        // are registered in the function table and we can hook into them.
-        zend::ModuleDep::optional(cstr!("ev")),
-        zend::ModuleDep::optional(cstr!("event")),
-        zend::ModuleDep::optional(cstr!("libevent")),
-        zend::ModuleDep::optional(cstr!("uv")),
-        zend::ModuleDep::end(),
-    ];
-
-    // In PHP modules written in C, this just returns the address of a global,
-    // mutable variable. In Rust, you cannot initialize such a complicated
-    // global variable because of initialization order issues that have been
-    // found through decades of C++ experience.
-    // There are a variety of ways to deal with this; this is just one way.
-    static mut MODULE: Lazy<zend::ModuleEntry> = Lazy::new(|| zend::ModuleEntry {
-        name: PROFILER_NAME.as_ptr(),
-        // SAFETY: php_ffi.c defines this correctly
-        functions: unsafe { bindings::ddog_php_prof_functions },
-        module_startup_func: Some(minit),
-        module_shutdown_func: Some(mshutdown),
-        request_startup_func: Some(rinit),
-        request_shutdown_func: Some(rshutdown),
-        info_func: Some(minfo),
-        version: PROFILER_VERSION.as_ptr(),
-        post_deactivate_func: Some(prshutdown),
-        deps: DEPS.as_ptr(),
-        globals_ctor: Some(ginit),
-        globals_dtor: Some(gshutdown),
-        globals_size: 1,
-        #[cfg(php_zts)]
-        globals_id_ptr: ptr::addr_of_mut!(GLOBALS_ID_PTR),
-        #[cfg(not(php_zts))]
-        globals_ptr: ptr::null_mut(),
-        ..Default::default()
-    });
+pub unsafe extern "C" fn get_module() -> *mut zend::ModuleEntry {
+    let module = ptr::addr_of_mut!(MODULE);
 
-    // SAFETY: well, it's at least as safe as what every single C extension does.
-    unsafe { &mut *ptr::addr_of_mut!(MODULE) }
+    // Set fields that aren't const-compatible.
+    unsafe {
+        ptr::addr_of_mut!((*module).functions).write(bindings::ddog_php_prof_functions);
+        ptr::addr_of_mut!((*module).build_id).write(bindings::datadog_module_build_id());
+    }
+    module
 }
 
 unsafe extern "C" fn ginit(_globals_ptr: *mut c_void) {

profiling/src/lib.rs

@@ -30,13 +30,12 @@ use libdd_profiling::api::{
 use libdd_profiling::exporter::Tag;
 use libdd_profiling::internal::Profile as InternalProfile;
 use log::{debug, info, trace, warn};
-use once_cell::sync::OnceCell;
 use std::borrow::Cow;
 use std::collections::HashMap;
 use std::hash::Hash;
 use std::num::NonZeroI64;
 use std::sync::atomic::{AtomicBool, AtomicPtr, AtomicU32, Ordering};
-use std::sync::{Arc, Barrier};
+use std::sync::{Arc, Barrier, OnceLock};
 use std::thread::JoinHandle;
 use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
 
@@ -58,7 +57,7 @@ const UPLOAD_CHANNEL_CAPACITY: usize = 8;
 
 /// The global profiler. Profiler gets made during the first rinit after an
 /// minit, and is destroyed on mshutdown.
-static mut PROFILER: OnceCell<Profiler> = OnceCell::new();
+static mut PROFILER: OnceLock<Profiler> = OnceLock::new();
 
 /// Order this array this way:
 ///  1. Always enabled types.
@@ -657,7 +656,7 @@ const DDPROF_TIME: &str = "ddprof_time";
 const DDPROF_UPLOAD: &str = "ddprof_upload";
 
 impl Profiler {
-    /// Will initialize the `PROFILER` OnceCell and makes sure that only one thread will do so.
+    /// Will initialize the `PROFILER` OnceLock and makes sure that only one thread will do so.
     pub fn init(system_settings: &SystemSettings) {
         // SAFETY: the `get_or_init` access is a thread-safe API, and the
         // PROFILER is only being mutated in single-threaded phases such as

profiling/src/profiling/mod.rs

@@ -1,6 +1,6 @@
 use crate::SAPI;
 use libc::sched_yield;
-use once_cell::sync::OnceCell;
+use std::cell::OnceCell;
 use std::mem::MaybeUninit;
 use std::thread::JoinHandle;
 use std::time::{Duration, Instant};

profiling/src/profiling/thread_utils.rs

@@ -1,12 +1,12 @@
 use crate::zend::sapi_request_info;
 use log::warn;
-use once_cell::sync::OnceCell;
 use std::borrow::Cow;
 use std::collections::HashMap;
 use std::ffi::{CStr, OsStr};
 use std::fmt::{Display, Formatter};
 use std::os::unix::ffi::OsStrExt;
 use std::path::Path;
+use std::sync::OnceLock;
 
 // todo: unify with ../component/sapi
 #[derive(Copy, Clone, Eq, PartialEq)]
@@ -27,7 +27,7 @@ pub enum Sapi {
 
 impl Sapi {
     pub fn from_name(name: &str) -> Sapi {
-        static SAPIS: OnceCell<HashMap<&str, Sapi>> = OnceCell::new();
+        static SAPIS: OnceLock<HashMap<&str, Sapi>> = OnceLock::new();
         let sapis = SAPIS.get_or_init(|| {
             HashMap::from_iter([
                 ("apache2handler", Sapi::Apache2Handler),