From 9bbf9be0716116e107012f1e9a4fb3a254569611 Mon Sep 17 00:00:00 2001
From: CodeGraph Test <codegraph@example.invalid>
Date: Sat, 3 Oct 2026 18:03:54 +0800
Subject: [PATCH] perf(cli): allocate through mimalloc in the musl release
 builds

musl's allocator serializes threads, so the static Linux release binaries
got slower as threads were added: a full index of this repository took
129 s at 32 threads against 50 s at one, where a glibc build takes 5 s.

Use mimalloc (v2, MIT) as the global allocator on musl targets only. The
dependency is target-gated, so glibc, macOS and Windows builds keep their
system allocator and their dependency graph is unchanged.

Measured with the release toolchain (cargo-zigbuild 0.23.0, zig 0.15.1,
rustc 1.98.0) on the same archive: 32 threads 129 s -> 8.2-8.4 s over
three runs, 4 threads 70 s -> 13 s, 1 thread 50 s -> 36 s. Peak RSS at 32
threads rises from 249 MB to about 500 MB. Nodes, edges, unresolved refs
and file rows are identical to 0.53.1's.
---
 Cargo.lock                       | 19 +++++++++++++++++++
 README.md                        |  3 +++
 crates/codegraph-cli/Cargo.toml  |  5 +++++
 crates/codegraph-cli/src/main.rs |  7 +++++++
 docs/readme/README.zh-CN.md      |  2 ++
 5 files changed, 36 insertions(+)

diff --git a/Cargo.lock b/Cargo.lock
index d88e65f9..a251a2a4 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -615,6 +615,7 @@ dependencies = [
  "indicatif",
  "interprocess",
  "jsonc-parser",
+ "mimalloc",
  "predicates",
  "rayon",
  "regex",
@@ -1906,6 +1907,15 @@ version = "0.2.189"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3eaf3ede3fee6db1a4c2ee091bf8a8b4dccdc6d17f656fb07896ee72867612f2"
 
+[[package]]
+name = "libmimalloc-sys"
+version = "0.1.49"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a45a52f43e1c16f667ccfe4dd8c85b7f7c204fd5e3bf46c5b0db9a5c3c0b8e9"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "libsqlite3-sys"
 version = "0.38.2"
@@ -1968,6 +1978,15 @@ version = "2.8.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "88904434abc2901f197fe8cc55f0445e7ded921dba5911dad2e2b39b48e663c4"
 
+[[package]]
+name = "mimalloc"
+version = "0.1.52"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2d4139bb28d14ad1facf21d5eb8825051b326e172d216b39f6d31df53cc97862"
+dependencies = [
+ "libmimalloc-sys",
+]
+
 [[package]]
 name = "mime"
 version = "0.3.17"
diff --git a/README.md b/README.md
index 389a9dbd..134f01a0 100644
--- a/README.md
+++ b/README.md
@@ -82,6 +82,9 @@ GitHub Releases publish these archive families:
 | Windows x86_64      | `x86_64-pc-windows-msvc`     | `.zip`    |
 | Windows ARM64       | `aarch64-pc-windows-msvc`    | `.zip`    |
 
+The Linux archives are static musl builds that allocate through mimalloc; the
+macOS and Windows archives use the platform allocator.
+
 Each release also includes `SHA256SUMS`. GitHub CLI users can additionally verify
 an archive's build provenance:
 
diff --git a/crates/codegraph-cli/Cargo.toml b/crates/codegraph-cli/Cargo.toml
index fee53a8e..17876ad6 100644
--- a/crates/codegraph-cli/Cargo.toml
+++ b/crates/codegraph-cli/Cargo.toml
@@ -44,6 +44,11 @@ tracing = { workspace = true }
 tracing-subscriber = { workspace = true }
 time = { workspace = true }
 
+# musl's allocator serializes threads; the static Linux release builds use
+# mimalloc instead. Other targets keep their system allocator.
+[target.'cfg(target_env = "musl")'.dependencies]
+mimalloc = { version = "0.1.52", default-features = false, features = ["v2"] }
+
 [dev-dependencies]
 rusqlite = { workspace = true }
 codegraph-bench = { path = "../codegraph-bench" }
diff --git a/crates/codegraph-cli/src/main.rs b/crates/codegraph-cli/src/main.rs
index 310f49a7..e0b2c167 100644
--- a/crates/codegraph-cli/src/main.rs
+++ b/crates/codegraph-cli/src/main.rs
@@ -42,6 +42,13 @@ mod installer;
 mod structural_gate;
 mod viewer_gate;
 
+/// musl's allocator serializes threads, so the static Linux release builds got
+/// slower as threads were added: a full index of this repository took over two
+/// minutes at 32 threads, against 51 s at one. Through mimalloc it takes 8 s.
+#[cfg(target_env = "musl")]
+#[global_allocator]
+static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
 /// Test-only: the ONE process-wide environment lock for this binary.
 ///
 /// `cargo test` runs every unit test of this binary on threads of a SINGLE
diff --git a/docs/readme/README.zh-CN.md b/docs/readme/README.zh-CN.md
index 0f3154ad..9825c256 100644
--- a/docs/readme/README.zh-CN.md
+++ b/docs/readme/README.zh-CN.md
@@ -76,6 +76,8 @@ GitHub Releases 发布以下归档系列：
 | Windows x86_64      | `x86_64-pc-windows-msvc`     | `.zip`    |
 | Windows ARM64       | `aarch64-pc-windows-msvc`    | `.zip`    |
 
+Linux 归档是静态链接的 musl 构建，内存分配走 mimalloc；macOS 与 Windows 归档使用系统分配器。
+
 每个 Release 还包含 `SHA256SUMS`。GitHub CLI 用户可进一步验证归档构建来源：
 
 ```bash