Consider switching to undici library for http requests

[ppedziwiatr]

https://www.npmjs.com/package/undici

Simple benchmark:

That's 2x the difference...

[ppedziwiatr]

obviously just for the node.js env.

A bit more direct comparison with Axios:

....I believe this could have a huge impact on - for example - loading pages from the GQL endpoint.

[ppedziwiatr]

Yup, just like I've expected - huge performance boost while loading SmartWeave interacitons for our "loot" contract - from ~61s on default arweave.js (with Axios underneath) to ~40s on undici 🚀 20s difference...

arweave.js/axios

undici:

[rosmcmahon]

it does not support the browser. axios is a good all-rounder.

if you replaced axios with undici for nodejs, would you also replace it with something in the browser? are native browser fetch calls faster?

the code base is already split with a custom solution supporting nodejs crypto and webcrypto, so it's not totally unreasonable to split it further - although the cypto-drivers could probably be replaced by a single external library ? - but that's off topic

I will let someone else way in with opinion here, but if we decide to go that way would you be willing to do a PR for it?

[ppedziwiatr]

I believe the best solution would be to prepare an isomorphic-fetch, with undici as a library used by node (sth. similar to https://www.npmjs.com/package/isomorphic-fetch).
We will prepare sth. similar for our contracts sdk and will let you know if it works as expected. If it will, you could consider using it in arweave-js.

[ppedziwiatr]

maybe it could also speed up a bit #128

[ppedziwiatr]

we've created https://github.com/redstone-finance/redstone-isomorphic (which has isomorphic version of fetch - uses undici for node and native fetch for web).
But I assume you're currently not interested in using this (and we've already created a wrapper for arweave-js in our SDK), so I'm closing this issue.