Claims about preloader are not backed by data

[yoavweiss]

The explainer says: "While this indeed benefited typical websites of the time, it does not benefit modern sites that are properly marked up with async/defer scripts where appropriate"

It'd be good to show data that proves this is indeed the case

[eligrey]

True, it would be nice to see that data. Is it possible to easily disable speculative loading alone in the Chrome source code, or is it too deeply intertwined with speculative parsing? I skimmed some of the relevant source and it seems like this should be doable.

I'd like to note a distinction between the goals of this API and the benefits being mentioned. Preload definitely benefits the speed of typical websites, and 'no-preload' even with a speed decrease still infers privacy benefits to the end-user in the third party JS library use-case.

[yoavweiss]

True, it would be nice to see that data. Is it possible to easily disable speculative loading alone in the Chrome source code, or is it too deeply intertwined with speculative parsing? I skimmed some of the relevant source and it seems like this should be doable.

One could create a custom Chromium build which bails early on HTMLPreloadScanner::Scan to test your hypothesis.

'no-preload' even with a speed decrease still infers privacy benefits to the end-user in the third party JS library use-case.

I believe the behavior you're after is the one implemented in Chromium when markup based meta CSP is present. (done with the same tradeoffs in mind)

[eligrey]

I believe the behavior you're after is the one implemented in Chromium when markup based meta CSP is present. (done with the same tradeoffs in mind)

I tried triggering this code path with a lax in-markup meta CSP to no avail. This was in combination with a JS library that also (synchronously) emits a generated strict meta CSP.

[eligrey]

I'll look into getting some benchmarks later. I might not have figures ready for the F2F this thursday though.

I've also changed the proposal text to state "[…] it does not always benefit […]" instead of "[…] it does not benefit […]"

[yoavweiss]

I tried triggering this code path with a lax in-markup meta CSP to no avail. This was in combination with a JS library that also (synchronously) emits a generated strict meta CSP.

Seems like this works in some cases, but not others sigh I'll file a crbug to figure this out

[yoavweiss]

Digging further, it seems like Chromium is doing the right thing, and stalling the preloads until the HTML-based meta CSP is processed, but doesn't stop the preload scanner altogether, so it's not something you could use here.