Would the API name be confusing?

I think that it may be confusing that server.invalidateModule and moduleGraph.invalidateModule has the same name but they have different side effects.

Is the first parameter ok? e.g. a string vs a ModuleNode

We should probably follow the current convention in the module graph and pass a module instead.

What about server.reloadModule(module, ...)? IMO it more clearly states that it will trigger a Hot Module Reload for the module, and not only invalidate it.

reloadModule sounds nice! I updated to use that.

We should probably follow the current convention in the module graph and pass a module instead.

I also updated to a ModuleNode. I initially didn't went with it since the API would tie to the moduleGraph. And to keep it consistent with ssrLoadModule. But I feel like ModuleNode would be more futureproof too so it might be fine.

This PR would make my i18next resources HMR desperately easy (if I have the correct understanding). Invalidating my virtual module should cause all the relative i18next resource to refresh. Aside from this, it looks like a detailed mess (outside of the vite ecosystem) to invalidate old resources and load new.

Thanks for your feedback @rosskevin, it's very helpful to know real use cases for PRs 🙏🏼
This feature will be included in Vite 3.2

The server.reloadModule is perfect, just how HMR should be. I replaced ~100 LOC that were relying on internal WebSocket and spread across parts of the code, with just one. It deserves some love in the docs!