Build Output Size Test Unfair?

[AlexanderKaran]

Had some feedback on the Build Output Size Test:

build output size is also not a good comparison. .output/ for Nuxt/Solid/TanStack is self contained, but SvelteKit/Astro/RR all need an additional unmeasured install from package.json - so those node_modules/ are untracked

Results here: https://frameworks.e18e.dev/dev-time/#build-and-install-times

Some deets on our methods: https://frameworks.e18e.dev/dev-time/#build-and-install-times

Do we keep it, edit the tests, edit the way we show it or remove it.

[Fuzzyma]

Show it, show a banner and the estimated extra size. Alternatively show it but only ever for one framework and don't let them compare against each other

[43081j]

So is the gap that we need to run install of the build output?

but SvelteKit/Astro/RR all need an additional unmeasured install from package.json

What exactly is this? The build output has its own package.json? What installs that usually?

[timneutkens]

For Next.js you could use output: 'standalone' which makes it isolated (no extra node_modules needed) but then you have to measure .next/standalone only, not all of .next.