diff --git a/api/sixtyfps-cpp/include/sixtyfps_sharedarray.h b/api/sixtyfps-cpp/include/sixtyfps_sharedarray.h
index af0d4f356..dfa06dd37 100644
--- a/api/sixtyfps-cpp/include/sixtyfps_sharedarray.h
+++ b/api/sixtyfps-cpp/include/sixtyfps_sharedarray.h
@@ -9,6 +9,7 @@
 LICENSE END */
 #pragma once
 #include "sixtyfps_sharedarray_internal.h"
+#include <atomic>
 
 namespace sixtyfps {
 
@@ -16,26 +17,29 @@ template<typename T>
 struct SharedArray
 {
     SharedArray()
-    {
-        cbindgen_private::sixtyfps_shared_array_new_null(reinterpret_cast<SharedArray<uint8_t> *>(this));
-    }
+        : inner(const_cast<SharedArrayHeader*>(reinterpret_cast<const SharedArrayHeader*>(
+            cbindgen_private::sixtyfps_shared_array_empty())))
+    { }
 
     SharedArray(const SharedArray &other)
+        : inner(other.inner)
     {
-        cbindgen_private::sixtyfps_shared_array_clone(
-                reinterpret_cast<SharedArray<uint8_t> *>(this),
-                reinterpret_cast<const SharedArray<uint8_t> *>(&other));
+        if (inner->refcount > 0) {
+            ++inner->refcount;
+        }
     }
     ~SharedArray()
     {
-        cbindgen_private::sixtyfps_shared_array_drop(reinterpret_cast<SharedArray<uint8_t> *>(this));
+        drop();
     }
     SharedArray &operator=(const SharedArray &other)
     {
-        cbindgen_private::sixtyfps_shared_array_drop(reinterpret_cast<SharedArray<uint8_t> *>(this));
-        cbindgen_private::sixtyfps_shared_array_clone(
-                reinterpret_cast<SharedArray<uint8_t> *>(this),
-                reinterpret_cast<const SharedArray<uint8_t> *>(&other));
+        if (other.inner == inner) { return *this; }
+        drop();
+        inner = other.inner;
+        if (inner->refcount > 0) {
+            ++inner->refcount;
+        }
         return *this;
     }
     SharedArray &operator=(SharedArray &&other)
@@ -44,7 +48,36 @@ struct SharedArray
         return *this;
     }
 
+    const T *begin() const {
+        return reinterpret_cast<const T *>(inner + 1);
+    }
+
+    const T *end() const {
+        return begin() + inner->size;
+    }
+
 private:
-    void *inner; // opaque
+
+    void drop() {
+        if (inner->refcount > 0 && (--inner->refcount) == 0) {
+            auto b = begin(), e = end();
+            for (auto it = b; it < e; ++it) {
+                it->~T();
+            }
+        }
+        cbindgen_private::sixtyfps_shared_array_free(
+            reinterpret_cast<uint8_t *>(inner),
+            sizeof(SharedArrayHeader) + inner->capacity * sizeof(T),
+            alignof(SharedArrayHeader));
+    }
+
+    // Unfortunately, this cannot be generated by cbindgen because std::atomic is not understood
+    struct SharedArrayHeader {
+        std::atomic<std::intptr_t> refcount;
+        std::size_t size;
+        std::size_t capacity;
+    };
+    static_assert(alignof(T) <= alignof(SharedArrayHeader), "Not yet supported because we would need to add padding");
+    SharedArrayHeader *inner;
 };
 }
diff --git a/sixtyfps_runtime/corelib/lib.rs b/sixtyfps_runtime/corelib/lib.rs
index 164788931..282552eb6 100644
--- a/sixtyfps_runtime/corelib/lib.rs
+++ b/sixtyfps_runtime/corelib/lib.rs
@@ -73,7 +73,7 @@ pub mod tests;
 pub fn use_modules() -> usize {
     tests::sixtyfps_mock_elapsed_time as usize
         + signals::ffi::sixtyfps_signal_init as usize
-        + sharedarray::ffi::sixtyfps_shared_array_drop as usize
+        + sharedarray::ffi::sixtyfps_shared_array_empty as usize
         + layout::solve_grid_layout as usize
         + item_tree::ffi::sixtyfps_visit_item_tree as usize
         + graphics::ffi::sixtyfps_new_path_elements as usize
diff --git a/sixtyfps_runtime/corelib/sharedarray.rs b/sixtyfps_runtime/corelib/sharedarray.rs
index 9689e394d..509af4288 100644
--- a/sixtyfps_runtime/corelib/sharedarray.rs
+++ b/sixtyfps_runtime/corelib/sharedarray.rs
@@ -10,74 +10,95 @@ LICENSE END */
 //! module for the SharedArray and related things
 #![allow(unsafe_code)]
 #![warn(missing_docs)]
+use core::fmt::Debug;
 use core::mem::MaybeUninit;
-use std::{fmt::Debug, fmt::Display, ops::Deref};
-use triomphe::{Arc, HeaderWithLength, ThinArc};
+use core::ops::Deref;
+use core::ptr::NonNull;
+use core::sync::atomic;
+use std::alloc;
+
+#[repr(C)]
+struct SharedArrayHeader {
+    refcount: atomic::AtomicIsize,
+    size: usize,
+    capacity: usize,
+}
+
+#[repr(C)]
+struct SharedArrayInner<T> {
+    header: SharedArrayHeader,
+    data: MaybeUninit<T>,
+}
+
+fn compute_inner_layout<T>(capacity: usize) -> alloc::Layout {
+    alloc::Layout::new::<SharedArrayHeader>()
+        .extend(alloc::Layout::array::<T>(capacity).unwrap())
+        .unwrap()
+        .0
+}
+
+unsafe fn drop_inner<T>(inner: NonNull<SharedArrayInner<T>>) {
+    debug_assert_eq!(inner.as_ref().header.refcount.load(core::sync::atomic::Ordering::Relaxed), 0);
+    let data_ptr = inner.as_ref().data.as_ptr();
+    for x in 0..inner.as_ref().header.size {
+        drop(core::ptr::read(data_ptr.add(x)));
+    }
+    alloc::dealloc(
+        inner.as_ptr() as *mut u8,
+        compute_inner_layout::<T>(inner.as_ref().header.capacity),
+    )
+}
+
+/// Allocate the memory for the SharedArray with the given capacity. Return the inner with size and refcount set to 1
+fn alloc_with_capacity<T>(capacity: usize) -> NonNull<SharedArrayInner<T>> {
+    let ptr = unsafe { alloc::alloc(compute_inner_layout::<T>(capacity)) };
+    unsafe {
+        core::ptr::write(
+            ptr as *mut SharedArrayHeader,
+            SharedArrayHeader { refcount: 1.into(), size: 0, capacity },
+        );
+    }
+    NonNull::new(ptr).unwrap().cast()
+}
 
-#[derive(Clone)]
 #[repr(C)]
 /// SharedArray holds a reference-counted read-only copy of `[T]`.
-pub struct SharedArray<T: 'static> {
-    /// Invariant: The usize header is the `len` of the vector, the contained buffer is `[T]`
-    inner: ThinArc<usize, MaybeUninit<T>>,
+pub struct SharedArray<T> {
+    inner: NonNull<SharedArrayInner<T>>,
 }
 
-struct PaddingFillingIter<'a, U> {
-    iter: &'a mut dyn Iterator<Item = MaybeUninit<U>>,
-    pos: usize,
-    len: usize,
-    padding_elements: usize,
-}
-
-impl<'a, U> PaddingFillingIter<'a, U> {
-    fn new(len: usize, iter: &'a mut dyn Iterator<Item = MaybeUninit<U>>) -> Self {
-        let alignment = core::mem::align_of::<usize>();
-        let mut padding_elements = if len == 0 { 1 } else { 0 }; // ThinArc can't deal with empty arrays, so add padding for empty arrays.
-
-        // Add padding to ensure that the size in bytes is a multiple of the pointer alignment. This can mean different
-        // increments depending on whether sizeof(U) is less or greater than align_of(usize).
-        loop {
-            let size_in_bytes = (len + padding_elements) * core::mem::size_of::<U>();
-            let byte_aligned_size = (size_in_bytes + alignment - 1) & !(alignment - 1);
-            let padding_bytes = byte_aligned_size - size_in_bytes;
-            if padding_bytes == 0 {
-                break;
+impl<T> Drop for SharedArray<T> {
+    fn drop(&mut self) {
+        unsafe {
+            if self.inner.as_ref().header.refcount.load(atomic::Ordering::Relaxed) < 0 {
+                return;
+            }
+            if self.inner.as_ref().header.refcount.fetch_sub(1, atomic::Ordering::SeqCst) == 1 {
+                drop_inner(self.inner)
             }
-            padding_elements += 1;
         }
-
-        Self { iter, pos: 0, len, padding_elements }
     }
 }
 
-impl<'a, U: Clone> Iterator for PaddingFillingIter<'a, U> {
-    type Item = MaybeUninit<U>;
-    fn next(&mut self) -> Option<MaybeUninit<U>> {
-        let pos = self.pos;
-        self.pos += 1;
-        if pos < self.len {
-            self.iter.next()
-        } else if pos < self.len + self.padding_elements {
-            Some(MaybeUninit::uninit())
-        } else {
-            None
+impl<T> Clone for SharedArray<T> {
+    fn clone(&self) -> Self {
+        unsafe {
+            if self.inner.as_ref().header.refcount.load(atomic::Ordering::Relaxed) > 0 {
+                self.inner.as_ref().header.refcount.fetch_add(1, atomic::Ordering::SeqCst);
+            }
+            return SharedArray { inner: self.inner };
         }
     }
-    fn size_hint(&self) -> (usize, Option<usize>) {
-        let l = self.len + self.padding_elements;
-        (l, Some(l))
-    }
 }
-impl<'a, U: Clone> core::iter::ExactSizeIterator for PaddingFillingIter<'a, U> {}
 
-impl<T: Clone> SharedArray<T> {
+impl<T> SharedArray<T> {
     fn as_ptr(&self) -> *const T {
-        self.inner.slice.as_ptr() as *const T
+        unsafe { self.inner.as_ref().data.as_ptr() }
     }
 
     /// Size of the string, in bytes
     pub fn len(&self) -> usize {
-        self.inner.header.header
+        unsafe { self.inner.as_ref().header.size }
     }
 
     /// Return a slice to the array
@@ -86,67 +107,54 @@ impl<T: Clone> SharedArray<T> {
     }
 
     /// Constructs a new SharedArray from the given iterator.
-    pub fn from_iter(iter: impl Iterator<Item = T> + ExactSizeIterator) -> Self {
-        let len = iter.len();
-        let item_iter = &mut iter.map(|item| MaybeUninit::new(item));
-        let iter = PaddingFillingIter::new(len, item_iter);
-
-        SharedArray {
-            inner: Arc::into_thin(Arc::from_header_and_iter(
-                HeaderWithLength::new(len, iter.size_hint().0),
-                iter,
-            )),
+    pub fn from_iter(mut iter: impl Iterator<Item = T> + ExactSizeIterator) -> Self {
+        let capacity = iter.len();
+        let inner = alloc_with_capacity::<T>(capacity);
+        let mut result = SharedArray { inner };
+        let mut size = 0;
+        while let Some(x) = iter.next() {
+            assert_ne!(size, capacity);
+            unsafe {
+                core::ptr::write(result.inner.as_mut().data.as_mut_ptr().add(size), x);
+                size += 1;
+                result.inner.as_mut().header.size = size;
+            }
         }
+        result
     }
 
     /// Constructs a new SharedArray from the given slice.
-    pub fn from(slice: &[T]) -> Self {
+    pub fn from(slice: &[T]) -> Self
+    where
+        T: Clone,
+    {
         SharedArray::from_iter(slice.iter().cloned())
     }
 }
 
-impl<T: Clone> Deref for SharedArray<T> {
+impl<T> Deref for SharedArray<T> {
     type Target = [T];
     fn deref(&self) -> &Self::Target {
         self.as_slice()
     }
 }
-trait StaticNull: Sized + 'static {
-    const NULL: once_cell::sync::Lazy<ThinArc<usize, MaybeUninit<Self>>>;
-}
-impl<T: Clone + Default + Sized + 'static> StaticNull for T {
-    const NULL: once_cell::sync::Lazy<ThinArc<usize, MaybeUninit<T>>> =
-        once_cell::sync::Lazy::new(|| {
-            let len = 0;
-            let null_iter = &mut std::iter::empty();
-            let iter = PaddingFillingIter::new(len, null_iter);
 
-            Arc::into_thin(Arc::from_header_and_iter(
-                HeaderWithLength::new(len, iter.size_hint().0),
-                iter,
-            ))
-        });
-}
+static SHARED_NULL: SharedArrayHeader =
+    SharedArrayHeader { refcount: std::sync::atomic::AtomicIsize::new(-1), size: 0, capacity: 0 };
 
-impl<T: Clone + Default> Default for SharedArray<T> {
+impl<T> Default for SharedArray<T> {
     fn default() -> Self {
-        SharedArray { inner: StaticNull::NULL.clone() }
+        SharedArray { inner: NonNull::from(&SHARED_NULL).cast() }
     }
 }
 
-impl<T: Clone + Debug> Debug for SharedArray<T> {
+impl<T: Debug> Debug for SharedArray<T> {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         self.as_slice().fmt(f)
     }
 }
 
-impl<T: Clone + Debug> Display for SharedArray<T> {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        self.as_slice().fmt(f)
-    }
-}
-
-impl<T: Clone> AsRef<[T]> for SharedArray<T> {
+impl<T> AsRef<[T]> for SharedArray<T> {
     #[inline]
     fn as_ref(&self) -> &[T] {
         self.as_slice()
@@ -156,14 +164,85 @@ impl<T: Clone> AsRef<[T]> for SharedArray<T> {
 impl<T, U> PartialEq<U> for SharedArray<T>
 where
     U: ?Sized + AsRef<[T]>,
-    T: Clone + PartialEq,
+    T: PartialEq,
 {
     fn eq(&self, other: &U) -> bool {
         self.as_slice() == other.as_ref()
     }
 }
 
-impl<T: Clone + PartialEq> Eq for SharedArray<T> {}
+impl<T: Eq> Eq for SharedArray<T> {}
+
+impl<T: Clone> IntoIterator for SharedArray<T> {
+    type Item = T;
+    type IntoIter = IntoIter<T>;
+    fn into_iter(self) -> Self::IntoIter {
+        IntoIter(unsafe {
+            if self.inner.as_ref().header.refcount.load(atomic::Ordering::Relaxed) == 1 {
+                let inner = self.inner;
+                std::mem::forget(self);
+                inner.as_ref().header.refcount.store(0, atomic::Ordering::Relaxed);
+                IntoIterInner::UnShared(inner, 0)
+            } else {
+                IntoIterInner::Shared(self, 0)
+            }
+        })
+    }
+}
+
+enum IntoIterInner<T> {
+    Shared(SharedArray<T>, usize),
+    // Elements up to the usize member are already moved out
+    UnShared(NonNull<SharedArrayInner<T>>, usize),
+}
+
+impl<T> Drop for IntoIterInner<T> {
+    fn drop(&mut self) {
+        match self {
+            IntoIterInner::Shared(..) => { /* drop of SharedArray takes care of it */ }
+            IntoIterInner::UnShared(inner, begin) => unsafe {
+                debug_assert_eq!(inner.as_ref().header.refcount.load(atomic::Ordering::Relaxed), 0);
+                let data_ptr = inner.as_ref().data.as_ptr();
+                for x in (*begin)..inner.as_ref().header.size {
+                    drop(core::ptr::read(data_ptr.add(x)));
+                }
+                alloc::dealloc(
+                    inner.as_ptr() as *mut u8,
+                    compute_inner_layout::<T>(inner.as_ref().header.capacity),
+                )
+            },
+        }
+    }
+}
+
+/// An iterator that moves out of a SharedArray.
+///
+/// This `struct` is created by the `into_iter` method on [`SharedArray`] (provided
+/// by the [`IntoIterator`] trait).
+pub struct IntoIter<T>(IntoIterInner<T>);
+
+impl<T: Clone> Iterator for IntoIter<T> {
+    type Item = T;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match &mut self.0 {
+            IntoIterInner::Shared(array, moved) => {
+                let result = array.as_slice().get(*moved).cloned();
+                *moved += 1;
+                result
+            }
+            IntoIterInner::UnShared(inner, begin) => unsafe {
+                if *begin < inner.as_ref().header.size {
+                    let r = core::ptr::read(inner.as_ref().data.as_ptr().add(*begin));
+                    *begin += 1;
+                    Some(r)
+                } else {
+                    None
+                }
+            },
+        }
+    }
+}
 
 #[test]
 fn simple_test() {
@@ -184,24 +263,20 @@ pub(crate) mod ffi {
     use super::*;
 
     #[no_mangle]
-    /// This function is used for the low-level C++ interface to allocate the backing vector for an empty shared array.
-    pub unsafe extern "C" fn sixtyfps_shared_array_new_null(out: *mut SharedArray<u8>) {
-        core::ptr::write(out, SharedArray::<u8>::default());
+    /// This function is used for the low-level C++ interface to allocate the backing vector of a SharedArray.
+    pub unsafe extern "C" fn sixtyfps_shared_array_allocate(size: usize, align: usize) -> *mut u8 {
+        std::alloc::alloc(std::alloc::Layout::from_size_align(size, align).unwrap())
     }
 
     #[no_mangle]
-    /// This function is used for the low-level C++ interface to clone a shared array by increasing its reference count.
-    pub unsafe extern "C" fn sixtyfps_shared_array_clone(
-        out: *mut SharedArray<u8>,
-        source: &SharedArray<u8>,
-    ) {
-        core::ptr::write(out, source.clone());
+    /// This function is used for the low-level C++ interface to deallocate the backing vector of a SharedArray
+    pub unsafe extern "C" fn sixtyfps_shared_array_free(ptr: *mut u8, size: usize, align: usize) {
+        std::alloc::dealloc(ptr, std::alloc::Layout::from_size_align(size, align).unwrap())
     }
 
     #[no_mangle]
-    /// This function is used for the low-level C++ interface to decrease the reference count of a shared array.
-    pub unsafe extern "C" fn sixtyfps_shared_array_drop(out: *mut SharedArray<u8>) {
-        // ?? This won't call drop on the right type...
-        core::ptr::read(out);
+    /// This function is used for the low-level C++ interface to initialize the empty SharedArray.
+    pub unsafe extern "C" fn sixtyfps_shared_array_empty() -> *const u8 {
+        &SHARED_NULL as *const _ as *const u8
     }
 }