[Tracking issue] Improving build performance & experience

[nickschot]

Build performance can still be improved a lot. Especially rebuilds take too much time right now. The main tasks which consume time are the bundling (rollup) and the forking/reloading of the process.

Not entirely sure if the forking/reloading can be done any quicker, but for development an alternative solution could be found.

• Enable incremental rebuilds for Rollup ( see PR feat: enable rollup cache feature #700 )
• Improve performance of loading new bundle ( see issue Disconnect event not listened to on worker shutdown #702, PR fix: gracefully shutdown workers instead of waiting for the timeout to pass (fixes #702) #703 )
• Cache to file system in between builds (i.e. when exiting lux-cli)
• Make sure the "building" indicator always displays and disappears
• Prevent duplicate processes (cancel build on new change?)

[nickschot]

We are losing a lot of time because the worker.once('exit') is never called in time (so we always get the 5 second timeout).

It seems this is done intentionally. I guess to allow the worker to gracefully shutdown (?). It does slow down rebuild by a fixed 5 seconds. Fixed in #703

[nickschot]

For the last 2 points we still need some reliable reproduction. I encountered it often but have not yet been able to reliably reproduce. Likely timing/lock issue in the watcher.