fix(preload): allow ignoring dep errors

[danielroe]

Description

resolves #18042

This allows individual failures of dep preloads without a chain reaction (which Promise.all forces). By default the behaviour is the same, but if frameworks or users call e.preventDefault on the chunk errors it is still possible for the preload/import to succeed, which is I think valid in the case of CSS failures.

[stackblitz]

Run & review this pull request in StackBlitz Codeflow.

[patak-dev]

Looks good to me. Thanks for the issue and PR @danielroe!
We can merge this and later backport it so we release it in v5 too. It seems ok for a patch to me.

[sapphi-red]

I confirmed that Promise.allSettled is also supported by our default modern browser target.
https://caniuse.com/mdn-javascript_builtins_promise_allsettled

[patak-dev]

@danielroe vite@5.4.4 has been released backporting this PR

[johndunderhill]

@patak-dev @danielroe I think this broke dynamic preload error handling:

https://vitejs.dev/guide/build.html#load-error-handling

Since Vite 5.4.4, I am no longer receiving the vite:preloadError event. Something seems to be crashing before the event can be emitted:

My preload testing is done as follows:

1. Build the site for production (with chunking and hashes).
2. Preview the site.
3. Browse to the site, but don't touch anything.
4. Stop the preview server.
5. Delete the production build.
6. Make a change.
7. Rebuild the site.
8. Verify the hashes of some of the chunks are different.
9. Restart the preview server.
10. Go and hover over some of the links (triggers a preload).

Environment is Solidjs with the Solid Router, but I don't think this matters much here. This never fires:

window.addEventListener('vite:preloadError', (event) => {
	console.log('Global vite:preloadError event:', event);
});

John

[johndunderhill]

@patak-dev @danielroe I've confirmed this. This took a long time to track down.

In Vite 5.4.3, the type error still occurs, but there is no 'uncaught in Promise' and the Vite event is produced as expected:

John

[patak-dev]

That is strange. The code looks correct. @johndunderhill, would you create an issue with a minimal reproduction so we can track and fix this regression?

[danielroe]

Yes, I see the behaviour change. Basically errors called when calling baseModule() no longer trigger vite:preloadError. Will raise a PR to ensure that also causes the same error.

[patak-dev]

Thanks for the report @johndunderhill, and thanks for quickly fixing this @danielroe
It should be resolved in vite@5.4.5