Thanks @imbrem for investing into rust-multiaddr performance characteristics.

Can you please provide a benchmark that proves that this pull request improves performance?

I added some basic benchmarks in the branch https://github.com/imbrem/rust-multiaddr/tree/multiaddr-benches; these indicate a 25% performance improvement in creating and cloning ipv4 addresses, and about the same performance for ipv6 addresses. The biggest performance improvements should be in manipulating ipv4 addresses, but making a benchmark for this is more complicated; I should be able to find some time to do this in a few days.

The biggest performance improvements should be in manipulating ipv4 addresses

I can not think of a place where we manipulate an IPv4 address in a hot-path.

If you want I can try to benchmark memory usage, which should be significantly improved as we eliminate the Arc allocation holding the Vec. Another option is to modify EcoVec to have a longer buffer so it can hold an ipv6 address inline or more; this would leave memory the same as the original (since the larger EcoVec would have a larger storage footprint) but now more multiaddrs would share in the performance improvement. Issue is I’d probably have to fork EcoVec in the latter case since it adds a lot to the complexity.

If you want I can try to benchmark memory usage, which should be significantly improved as we eliminate the Arc allocation holding the Vec.

You could test with some downstream application using rust-libp2p, e.g. https://github.com/mxinden/kademlia-exporter/. Maybe the change here shows up in heaptrack of a CPU flamegraph.