#lol #alloc #allocator

no-std lol_alloc

lol_alloc: A laughably simple wasm global_allocator

7 unstable releases (3 breaking)

0.4.0 Apr 28, 2023
0.3.0 Nov 23, 2022
0.2.0 Oct 15, 2022
0.1.3 Aug 11, 2022

#58 in Memory management

Download history 1032/week @ 2023-08-16 1114/week @ 2023-08-23 1111/week @ 2023-08-30 994/week @ 2023-09-06 848/week @ 2023-09-13 1038/week @ 2023-09-20 1421/week @ 2023-09-27 965/week @ 2023-10-04 1273/week @ 2023-10-11 933/week @ 2023-10-18 1116/week @ 2023-10-25 1039/week @ 2023-11-01 1063/week @ 2023-11-08 1056/week @ 2023-11-15 1018/week @ 2023-11-22 910/week @ 2023-11-29

4,205 downloads per month
Used in 2 crates (via purewasm-bindgen)

MIT license

612 lines


A laughably simple wasm global_allocator.

Like wee_alloc, but smaller since I used skinnier letters in the name.

lol_alloc is a collection of simple wasm global_allocators.

I wrote lol_alloc to learn about allocators (I hadn't written one before) and because wee_alloc seems unmaintained and has a leak. After looking at wee_alloc's implementation (which I failed to understand or fix), I wanted to find out how hard it really is to make a wasm global_allocator, and it seemed like providing one could be useful to the rust wasm community.


You can replace the global_allocator with LockedAllocator<FreeListAllocator> for wasm32 builds using:

extern crate alloc;

#[cfg(target_arch = "wasm32")]
use lol_alloc::{FreeListAllocator, LockedAllocator};

#[cfg(target_arch = "wasm32")]
static ALLOCATOR: LockedAllocator<FreeListAllocator> = LockedAllocator::new(FreeListAllocator::new());

For slightly smaller file size and slightly better performance, single threaded WASM applications can use AssumeSingleThreaded instead of LockedAllocator:

extern crate alloc;

use lol_alloc::{AssumeSingleThreaded, FreeListAllocator};

// SAFETY: This application is single threaded, so using AssumeSingleThreaded is allowed.
static ALLOCATOR: AssumeSingleThreaded<FreeListAllocator> =
    unsafe { AssumeSingleThreaded::new(FreeListAllocator::new()) };

Applications which do not need any allocator can use FailAllocator:

extern crate alloc;

#[cfg(target_arch = "wasm32")]
use lol_alloc::FailAllocator;

#[cfg(target_arch = "wasm32")]
static ALLOCATOR: FailAllocator = FailAllocator;

Applications which only do a bounded small number of allocations and thus don't require freeing can use one of the leaking allocators. LeakingPageAllocator (shown below), AssumeSingleThreaded<LeakingAllocator> and LockedAllocator<LeakingAllocator> are the best options for this case:

extern crate alloc;

#[cfg(target_arch = "wasm32")]
use lol_alloc::LeakingPageAllocator;

#[cfg(target_arch = "wasm32")]
static ALLOCATOR: LeakingPageAllocator = LeakingPageAllocator;

Thread Safety

LeakingAllocator and FreeListAllocator are NOT Sync and must be wrapped in either LockedAllocator or the unsafe AssumeSingleThreaded to assign to a static (this is enforced by the Rust type system). Multithreading is possible in wasm these days: do not use AssumeSingleThreaded unless you are confident that all allocations and freeing will happen from a single thread.

FailAllocator, LeakingPageAllocator are thread-safe and do not need any wrapping.


A few projects have apparently used this library, and there have been no reported issues (none reported success either, so use at your own risk).

FreeListAllocator has pretty good test suite, and the rest of the allocators are trivial, and had at least minimal testing.

If you use it, please report any bugs. If it actually works for you, also let me know (you can post an issue with your report).

Sizes of allocators include overhead from example (compiled with rustc 1.65.0 and wasm-pack 0.10.3):

  • FailAllocator: 195 bytes: errors on allocations. Operations are O(1).
  • LeakingPageAllocator: 230 bytes: Allocates pages for each allocation. Operations are O(1).
  • LeakingAllocator: Bump pointer allocator, growing the heap as needed and does not reuse/free memory. Operations are O(1). No allocation space overhead other than for alignment.
    • AssumeSingleThreaded<LeakingAllocator>: 356 bytes.
    • LockedAllocator<LeakingAllocator>: 484 bytes.
  • FreeListAllocator: Free list based allocator. Operations (both allocation and freeing) are O(length of free list), but it does coalesce adjacent free list nodes. Rounds allocations up to at least 2 words in size, but otherwise should use all the space. Even gaps from high alignment allocations end up in its free list for use by smaller allocations.
    • AssumeSingleThreaded<FreeListAllocator>: 654 bytes.
    • LockedAllocator<FreeListAllocator>: 775 bytes.
  • Builtin Rust allocator: 5034 bytes.

If you can afford the extra code size, use the builtin rust allocator: it is a much better allocator.

Supports only wasm32: other targets may build, but the allocators will not work on them (except: FailAllocator, it errors on all platforms just fine).


Asymptotic behavior is documented in the above status section along with code size.

These allocators are all optimized for simplicity (and thus code size and maintainability), and not runtime performance. Runtime performance data for them them (and comparing to the built in allocator) has not been collected but would be interesting: feel free to send patches with benchmarks and/or performance data.

If you care about runtime performance, you should probably use the built in allocator, or at least carefully measure the impact of using any of these allocators on your code.


Soundness of the pointer manipulation in this library is currently unclear. Since wasm32::memory_grow does not return a pointer there is no "original pointer" so the Strict Provenance rules can not be followed. Attempting to determine if this library's use of pointes at least meets the requirements for being dereferenceable when it dereferences them is similarly challenging as that is defined as:

dereferenceable: the memory range of the given size starting at the pointer must all be within the bounds of a single allocated object.

The definition of "allocated object" is not clear here. If the growable wasm heap counts as a single allocated object, then all these allocators are likely ok (in this aspect at least). However if each call to wasm32::memory_grow is considered to create a new allocated object, then the free list coalescing in FreeListAllocator in unsound and could result in undefined behavior.


There are some normal rust unit tests (run with cargo test), which use a test implementation of MemoryGrower.

There are also some wasm-pack tests (run with wasm-pack test --node lol_alloc)

Size testing:

wasm-pack build --release example && wc -c example/pkg/lol_alloc_example_bg.wasm

Change log


  • Make implementations which are wasm only not available on other targets, instead of failing at runtime.
  • Hint docs.rs to only build for wasm.


  • Add AssumeSingleThreaded.
  • Remove unsound Sync implementations for FreeListAllocator and LeakingAllocator: use AssumeSingleThreaded and its unsafe AssumeSingleThreaded::new function instead: this puts all known safety issues in this library behind an unsafe function.
  • Remove default FreeListAllocator type parameter from LockedAllocator.


  • Add LockedAllocator.