Module worker descendant script fetching regression

[domenic]

Discussion continued from #9499.

Apparently in #8253 we stopped using the fetch client settings object (i.e., the "outer" environment settings object) when fetching the descendants of a module worker. This is incorrect.

You can find the old spec in https://html.spec.whatwg.org/commit-snapshots/2edc8089d4aa2db44a324f44b3fe4e4898524217/#fetch-a-worklet/module-worker-script-graph

This impacts, e.g., what CSP or other policies applies to the module fetches. So this is definitely a bug.

@nicolo-ribaudo, can you work on fixing this to the state before your PR?

[nicolo-ribaudo]

Yes I can work on this 👍

[allstarschh]

@domenic Can you explain why fetching the descendants needs the environment settings object from the document? Because after fetching the top-level worker module script, worker will have its own environment settings object.

[domenic]

Because after fetching the top-level worker module script, worker will have its own environment settings object.

That's not true, or at least if it is, it is only because of weird specification-only things. You can't create the worker thread until you have all the code you plan to run inside of it, which means the whole dependency graph.

[domfarolino]

Dynamic imports inside of module works will use the worker script's settings object though, right?