Replace ed25519 with libsodium

ed25519 hasn’t been updated in years and is lagging behind what’s otherwise available: https://github.com/future-tense/ed25519-dalek#benchmarks

ed25519 might have been a good choice in terms of a smaller binary size, but this only used for node anyway, so the extra bloat doesn’t matter as much.

I’d suggest libsodium/sodium-native, since that’s an actual npm package, and since signature verification isn’t that big of a deal; I’d be surprised if most people even know it’s in the SDK.

I’d also suggest using the native bindings for key generation, instead of only using it for signing and verification.

Can do the work, just want a go/no-go

[morleyzhi]

I like those ideas to reduce the bundle size. I don’t think reduce -> keyBy is a big enough upgrade to warrant the file size increase.

As far as tree-shaking is concerned: I worked on this a few weeks ago but came to the conclusion that it requires a lot of complicated changes, and I’m not positive the output will be worth it. Especially if we can just write Lodash out of the packages. Here’s the rough todo list for that, which has to be done to all 3 libs:

• Upgrade webpack to 3 or 4
• Upgrade webpack’s peer dependencies to the latest versions
• Rewrite all module.exports use to use imports / exports instead. All instances have to be rewritten for tree-shaking to work
• Figure out which files have side effects and which are side-effect-free. Mark the former.
• Modify all the unit and integration test setups to import the rewritten library the correct way such that tests continue to pass

You have to very precisely manage your codebase to get tree-shaking working, and I wasn’t confident at the time that I could do that in a reasonable amount of time.

[morleyzhi]

I just updated stellar-base and stellar-sdk to remove node builtins from the bundle. Unless my methodology was wrong, I was used your webpack config with Webpack Bundle Analyzer, and it no longer includes the crypto polyfill or elliptic. It does still include tweetnacl.