From b70f62a64a455ab0ef6f1df02ef14ae52388950b Mon Sep 17 00:00:00 2001
From: Julien Portalier <julien@portalier.com>
Date: Wed, 8 Jan 2025 15:04:59 +0100
Subject: [PATCH] Use Crystal::PointerLinkedList instead of Deque in Mutex

The flat array also doesn't have much impact since it's never traversed:
we only reach the head or the tail once (to dequeue or dequeue one).

This spares a number of GC allocations since the Deque needs to be
allocated and also a buffer that will have to be reallocated sometimes
(and will only ever grow).
---
 src/fiber/waiting.cr | 13 +++++++++++++
 src/mutex.cr         | 20 ++++++++++++--------
 src/wait_group.cr    | 16 +++-------------
 3 files changed, 28 insertions(+), 21 deletions(-)
 create mode 100644 src/fiber/waiting.cr

diff --git a/src/fiber/waiting.cr b/src/fiber/waiting.cr
new file mode 100644
index 000000000000..8c4d0138a9f1
--- /dev/null
+++ b/src/fiber/waiting.cr
@@ -0,0 +1,13 @@
+class Fiber
+  # :nodoc:
+  struct Waiting
+    include Crystal::PointerLinkedList::Node
+
+    def initialize(@fiber : Fiber)
+    end
+
+    def enqueue : Nil
+      @fiber.enqueue
+    end
+  end
+end
diff --git a/src/mutex.cr b/src/mutex.cr
index 780eac468201..ad9bc1ac2ca5 100644
--- a/src/mutex.cr
+++ b/src/mutex.cr
@@ -1,3 +1,4 @@
+require "fiber/waiting"
 require "crystal/spin_lock"
 
 # A fiber-safe mutex.
@@ -22,9 +23,9 @@ class Mutex
   @state = Atomic(Int32).new(UNLOCKED)
   @mutex_fiber : Fiber?
   @lock_count = 0
-  @queue = Deque(Fiber).new
+  @queue = Crystal::PointerLinkedList(Fiber::Waiting).new
   @queue_count = Atomic(Int32).new(0)
-  @lock = Crystal::SpinLock.new
+  @spin = Crystal::SpinLock.new
 
   enum Protection
     Checked
@@ -59,7 +60,9 @@ class Mutex
     loop do
       break if try_lock
 
-      @lock.sync do
+      waiting = Fiber::Waiting.new(Fiber.current)
+
+      @spin.sync do
         @queue_count.add(1)
 
         if @state.get(:relaxed) == UNLOCKED
@@ -71,7 +74,7 @@ class Mutex
           end
         end
 
-        @queue.push Fiber.current
+        @queue.push pointerof(waiting)
       end
 
       Fiber.suspend
@@ -116,17 +119,18 @@ class Mutex
       return
     end
 
-    fiber = nil
-    @lock.sync do
+    waiting = nil
+    @spin.sync do
       if @queue_count.get == 0
         return
       end
 
-      if fiber = @queue.shift?
+      if waiting = @queue.shift?
         @queue_count.add(-1)
       end
     end
-    fiber.enqueue if fiber
+
+    waiting.try(&.value.enqueue)
   end
 
   def synchronize(&)
diff --git a/src/wait_group.cr b/src/wait_group.cr
index cf1ca8900e8f..efa19f15e1b4 100644
--- a/src/wait_group.cr
+++ b/src/wait_group.cr
@@ -1,4 +1,5 @@
 require "fiber"
+require "fiber/waiting"
 require "crystal/spin_lock"
 require "crystal/pointer_linked_list"
 
@@ -31,17 +32,6 @@ require "crystal/pointer_linked_list"
 # wg.wait
 # ```
 class WaitGroup
-  private struct Waiting
-    include Crystal::PointerLinkedList::Node
-
-    def initialize(@fiber : Fiber)
-    end
-
-    def enqueue : Nil
-      @fiber.enqueue
-    end
-  end
-
   # Yields a `WaitGroup` instance and waits at the end of the block for all of
   # the work enqueued inside it to complete.
   #
@@ -59,7 +49,7 @@ class WaitGroup
   end
 
   def initialize(n : Int32 = 0)
-    @waiting = Crystal::PointerLinkedList(Waiting).new
+    @waiting = Crystal::PointerLinkedList(Fiber::Waiting).new
     @lock = Crystal::SpinLock.new
     @counter = Atomic(Int32).new(n)
   end
@@ -128,7 +118,7 @@ class WaitGroup
   def wait : Nil
     return if done?
 
-    waiting = Waiting.new(Fiber.current)
+    waiting = Fiber::Waiting.new(Fiber.current)
 
     @lock.sync do
       # must check again to avoid a race condition where #done may have