Inaccurate results - maybe because of babel-register?

[jnields]

Hey - great idea but can't help but notice how inaccurate the results are

React is not as fast at SSR as marko, but it is not THAT much slower, and this has been popping up around the internet when searching for "framework performance compared" so it's a bit misleading

I didn't have time to fix everything, but just changing the SSR code for react / inferno / preact to be transpiled properly shows the following benchmarks for server-rendering the color picker:

npm run build-benchmark
sudo nice -n -20 npm run benchmark

inferno x 7,060 ops/sec ±0.50% (278 runs sampled)
marko x 8,382 ops/sec ±0.57% (277 runs sampled)
preact x 4,247 ops/sec ±0.40% (278 runs sampled)
react x 4,000 ops/sec ±1.07% (278 runs sampled)
vue x 920 ops/sec ±1.58% (206 runs sampled)

fork here: https://github.com/jnields/isomorphic-ui-benchmarks

[DylanPiercey]

@jnields sorry for taking such a long time to get back to you on this one. I've done my best to bring the repo back into a usable state with updated dependencies (including react).

I'm seeing React do a fair amount better (2-3x) but still lagging behind more significantly than what you are able to see.

I'm not sure how this could be an issue with babel/register, but I will do my best once I get a spare slot of time to pre compile all of the SSR stuff as well. One thing that is different is that now all of the packages are outputting modern JS and so maybe that was the culprit and it is resolved.

Either way I'll leave this one open until I can get the benchmarks to be precompiled for the server.

[Sebring]

A Marko 4 vs Marko 5 would also be of interest ;)