fix(hmr): only invalidate lastHMRTimestamp of importers if the invalidated module is not a HMR boundary

[hyf0]

Description

Try to fix #3033 while pair programming with @underfin.

Additional context

We are fixing some react HMR issues in Rspack and found that Vite has a similar issue. After web-infra-dev/rspack#2714, I start to investigate the issue of Vite, which turns out the issue has a different cause than Rspack. After a hard coding with @underfin, we locate the issue, which is caused by cycle dependency.

We came out of this fix, not sure it breaks other things.

---

What is the purpose of this pull request?

• Bug fix
• New Feature
• Documentation update
• Other

Before submitting the PR, please make sure you do the following

• Read the Contributing Guidelines.
• Read the Pull Request Guidelines and follow the PR Title Convention.
• Check that there isn't already a PR that solves the problem the same way to avoid creating a duplicate.
• Provide a description in this PR that addresses what the PR is solving, or reference the issue that it solves (e.g. fixes #123).
• Ideally, include relevant tests that fail without this PR but pass with it.

[stackblitz]

Run & review this pull request in StackBlitz Codeflow.

[patak-dev]

Thanks a lot for porting the fix to Vite @hyf0 and @underfin! It is highly appreciated ❤️

[patak-dev]

/ecosystem-ci run

[patak-dev]

/ecosystem-ci run

[vite-ecosystem-ci]

📝 Ran ecosystem CI: Open

suite	result
astro	✅ success
histoire	✅ success
iles	✅ success
ladle	✅ success
laravel	✅ success
marko	✅ success
nuxt	✅ success
previewjs	✅ success
qwik	✅ success
rakkas	✅ success
sveltekit	✅ success
vite-plugin-ssr	✅ success
vite-plugin-react	✅ success
vite-plugin-react-pages	✅ success
vite-plugin-react-swc	✅ success
vite-plugin-svelte	✅ success
vite-plugin-vue	✅ success
vite-setup-catalogue	✅ success
vitepress	✅ success
vitest	✅ success
windicss	✅ success

[ArnaudBarre]

Thanks for looking into this! There is effectively something we can do here to be smarter on invalidation. But the current changes creates two issues I think:

• If a module is a dead end, it's not invalidated
• When a dead end is found, we continue to search for boundaries which can slow down full page reload in case of big module graph

If you want I can try to look deeper on this later this weekend

Edit: I'm less sure for the second point. Because actually that can be beneficial to avoid more invalidation if other branches contains HMR boundaries. I think we need to rethink this code further to not be invalidate then search for boundaries or the opposite, but explore the graph to do both at the same time. Easier said than done

[ArnaudBarre]

So I took a deeper dive and I discovered (thanks to @patak-dev) that handling circular imports & tailwind has already created a lot of complexity and invalidate way too many things.
For example when a file change, we first walk through the module graph to invalidate, but with isHMR to false which doesn't update lastHMRTimestamp and so doesn't trigger this bug.

I will see how to improve that once I've finished my work to integrate Lightning CSS.

In the meantime I'm ok with this PR if we move the invalidate call between the propagateUpdate call and hasDeadEnd check.

Thanks again for helping fix that!

Updated in last commit