feat(profiling): Set and StringSet

Author: morrisonlevi

Cargo.lock

@@ -21,6 +21,7 @@ name = "main"
 harness = false
 
 [dependencies]
+allocator-api2 = { version = "0.2", default-features = false, features = ["alloc"] }
 anyhow = "1.0"
 bitmaps = "3.2.0"
 byteorder = { version = "1.5", features = ["std"] }
@@ -30,6 +31,7 @@ libdd-alloc = { version = "1.0.0", path = "../libdd-alloc" }
 libdd-profiling-protobuf = { version = "1.0.0", path = "../libdd-profiling-protobuf", features = ["prost_impls"] }
 libdd-common = { version = "1.0.0", path = "../libdd-common" }
 futures = { version = "0.3", default-features = false }
+hashbrown = { version = "0.16", default-features = false }
 http = "1.0"
 hyper = { workspace = true}
 http-body-util = "0.1"
@@ -42,6 +44,7 @@ rustc-hash = { version = "1.1", default-features = false }
 serde = {version = "1.0", features = ["derive"]}
 serde_json = {version = "1.0"}
 target-triple = "0.1.4"
+thiserror = "2"
 tokio = {version = "1.23", features = ["rt", "macros"]}
 tokio-util = "0.7.1"
 zstd = { version = "0.13", default-features = false }
@@ -50,3 +53,4 @@ zstd = { version = "0.13", default-features = false }
 bolero = "0.13"
 criterion = "0.5.1"
 lz4_flex = { version = "0.9", default-features = false, features = ["std", "frame"] }
+proptest = "1"

LICENSE-3rdparty.yml

@@ -12,4 +12,4 @@ pub mod exporter;
 pub mod internal;
 pub mod iter;
 pub mod pprof;
-mod profiles;
+pub mod profiles;

libdd-profiling/Cargo.toml

@@ -1,20 +1,19 @@
 // Copyright 2023-Present Datadog, Inc. https://www.datadoghq.com/
 // SPDX-License-Identifier: Apache-2.0
 
-use anyhow::Context;
 use libdd_profiling_protobuf::prost_impls::{Profile, Sample};
-use std::io::Cursor;
 
 fn deserialize_compressed_pprof(encoded: &[u8]) -> anyhow::Result<Profile> {
     use prost::Message;
-    use std::io::Read;
 
     // The zstd bindings use FFI so they don't work under miri. This means the
     // buffer isn't compressed, so simply convert to a vec.
     #[cfg(miri)]
     let buf = encoded.to_vec();
     #[cfg(not(miri))]
     let buf = {
+        use anyhow::Context;
+        use std::io::{Cursor, Read};
         let mut decoder =
             zstd::Decoder::new(Cursor::new(encoded)).context("failed to create zstd decoder")?;
         let mut out = Vec::new();

libdd-profiling/src/lib.rs

@@ -0,0 +1,29 @@
+// Copyright 2025-Present Datadog, Inc. https://www.datadoghq.com/
+// SPDX-License-Identifier: Apache-2.0
+
+#[repr(C)]
+#[derive(Debug, thiserror::Error)]
+pub enum SetError {
+    #[error("set error: invalid argument")]
+    InvalidArgument,
+    #[error("set error: out of memory")]
+    OutOfMemory,
+}
+
+impl From<libdd_alloc::AllocError> for SetError {
+    fn from(_: libdd_alloc::AllocError) -> Self {
+        SetError::OutOfMemory
+    }
+}
+
+impl From<std::collections::TryReserveError> for SetError {
+    fn from(_: std::collections::TryReserveError) -> Self {
+        SetError::OutOfMemory
+    }
+}
+
+impl From<hashbrown::TryReserveError> for SetError {
+    fn from(_: hashbrown::TryReserveError) -> Self {
+        SetError::OutOfMemory
+    }
+}

libdd-profiling/src/pprof/test_utils.rs

@@ -0,0 +1,67 @@
+// Copyright 2025-Present Datadog, Inc. https://www.datadoghq.com/
+// SPDX-License-Identifier: Apache-2.0
+
+mod error;
+mod set;
+mod slice_set;
+mod string_set;
+mod thin_str;
+
+pub type SetHasher = core::hash::BuildHasherDefault<rustc_hash::FxHasher>;
+
+pub use error::*;
+pub use set::*;
+pub use slice_set::*;
+pub use string_set::*;
+pub use thin_str::*;
+
+use std::any::TypeId;
+
+/// Operations that somewhat abstract around Sets of single items and sets of
+/// slices for sharded sets.
+///
+/// # Safety
+///
+/// Implementors must ensure that all methods which take `&self` are safe to
+/// call under a read-lock, and all `&mut self` methods are safe to call under
+/// a write-lock, and are safe for `Send` and `Sync`.
+pub unsafe trait ShardedSetOps {
+    type Lookup<'a>: Copy
+    where
+        Self: 'a;
+
+    /// Owned payload used for insertion. For some containers (e.g. slice-backed
+    /// sets) this can be a borrowed view like `&'a [T]` because the container
+    /// copies data into its own arena during insertion.
+    type Owned<'a>
+    where
+        Self: 'a;
+
+    type Id: Copy;
+
+    /// Returns the `TypeId` of the logical element type stored by this set.
+    fn type_id(&self) -> TypeId;
+
+    fn try_with_capacity(capacity: usize) -> Result<Self, SetError>
+    where
+        Self: Sized;
+
+    fn len(&self) -> usize;
+
+    #[inline]
+    fn is_empty(&self) -> bool {
+        self.len() == 0
+    }
+
+    /// # Safety
+    /// Same safety contract as the underlying container's find_with_hash.
+    unsafe fn find_with_hash(&self, hash: u64, key: Self::Lookup<'_>) -> Option<Self::Id>;
+
+    /// # Safety
+    /// Same safety contract as the underlying container's insert_unique_uncontended_with_hash.
+    unsafe fn insert_unique_uncontended_with_hash(
+        &mut self,
+        hash: u64,
+        key: Self::Owned<'_>,
+    ) -> Result<Self::Id, SetError>;
+}