fix: replace runner-side path normalization with fetchModule-side resolve

[hi-ogawa]

Description

• closes virtual file system is broken in v6 #18223

Not so confident that this is a proper fix nor the issue needs fix, but I'm thinking about what would require to support #18223 and hopefully it helps deciding the fate of the issue (bug or wontfix).

One thing I noticed that it looks possible to dedupe/normalize /abs-path-to/src/etc -> /src/etc on server side (fetchModule) and module identity is still kept intact on runner side since they are keyed by resolved id. For FetchModuleResult.cache detection, I replaced getModuleByUrl with ensureEntryFromUrl and I'm wondering what's the downside of this.

todo

• decide on file:// support (resolved in feat!: support file:// resolution #18422)

[stackblitz]

Run & review this pull request in StackBlitz Codeflow.

[hi-ogawa]

normalizeAbsoluteUrl also noramlizes file:///... to support this test case, but I'm not sure if we needs to keep this if we go with externalizing file:// entirely #17369. Also v5's ssrLoadModule('file://') doesn't seem to work either.

vite/packages/vite/src/node/ssr/runtime/__tests__/server-runtime.spec.ts

Lines 17 to 20 in 5e8e7ec

	// loads the same module if id is a file url
	const fileUrl = new _URL('./fixtures/simple.js', import.meta.url)
	const mod2 = await runner.import(fileUrl.toString())
	expect(mod).toBe(mod2)

[hi-ogawa]

/ecosystem-ci run

[vite-ecosystem-ci]

📝 Ran ecosystem CI on 92f53eb: Open

suite	result	latest scheduled
histoire	❌ failure	❌ failure
nuxt	❌ failure	❌ failure
redwoodjs	❌ failure	❌ failure
remix	❌ failure	❌ failure
vike	❌ failure	❌ failure
waku	❌ failure	❌ failure

✅ analogjs, astro, ladle, laravel, marko, previewjs, quasar, qwik, rakkas, storybook, sveltekit, unocss, vite-environment-examples, vite-plugin-pwa, vite-plugin-react, vite-plugin-react-swc, vite-plugin-svelte, vite-plugin-vue, vite-setup-catalogue, vitepress, vitest, vuepress

[sapphi-red]

Since root is deprecated and does not need to be set anymore, I think we can remove root from all the examples and types in the docs.

[hi-ogawa]

Makes sense. Updated the docs and also removed the internal usages of root.
I kept the default server runner one since users can technically access it via runner.options.root.

[sapphi-red]

@hi-ogawa Do you think this can be released in a patch? or do you prefer it to be released in a minor to be safe?

[hi-ogawa]

I think we can go as patch. runner.options.root becoming optional is actually a breaking change, but people who potentially consume that is only a few people, so I guess it's alright. Other than that, this should be a reasonable fix and I need to follow up on that if this breaks anyways.

[dai-shi]

Hi there, can I get an estimate about the release time?

[sapphi-red]

@dai-shi Probably I'll cut a release in few days 🫡