emma is a full-fledged allocator written in pure rust that is intended to not rely on anything in the target process. This means:

Fully no_std compatible.
No direct or indirect usage of libc (or any other non-rust library): emma uses raw syscalls instead.
No usage of shared resources: Instead of brk/sbrk to modify the shared data segment, emma only ever maps its own segment(s) using mmap.

This means that multiple emma instances can be used in the same process without ever noticing one another.

Performance

emma seems not far behind (other) state-of-the-art allocators.

Target Architecture

At the moment, emma exclusively targets linux on x86_64.

Known Issues

It is not clear how emma behaves when a process is forked. (Note: child processes of multi-threaded processes may only access async-signal-safe functions until they call execve anyway.)
If a thread is terminated during memory de-/allocation (terminating a thread at any other time in any way does not bother emma):
- tls disabled: The heap may end up permanently locked, deadlocking other threads.
- tls enabled: The heap may get corrupted as soon as another thread acquires the per-thread-heap from the dead thread.
Embedding multiple emmas into one process should work, but unless their symbols are renamed they may share data structures behind the scenes.
An emma instance does not return all its resources, even if all allocations are returned. This is reasonable for a global allocator, but makes it not as useful as a temporary allocator.
emma is not async-signal-safe, i.e., you may not de-/allocate memory in a signal handler. (The same probably holds true for your default memory allocator; POSIX does not list malloc or free as async-signal safe either.)

Features

tls enabling thread-local-storage requires a nightly compiler. Enabling tls massively increases performance.
boundary-checks enables assertions at the library boundary. These assertions cost a small amount of performance.

License

Licensed under either of:

Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)

at your option.

Contributions