Improve generalized/sourcemap fetcher

[paulirish]

#9459 landed our generalized fetcher (with an iframe).

We may have some areas where it can be improved.

CSP sensitivity.. I didn't test, but I believe a frame-src directive would introduce problems for this technique.

content-type sensitivity. In #12064 we see that if a map is served with application/octet-stream it'll trigger a user-visible download while Lighthouse is fetching it. DevTools, meanwhile, doesn't have that behavior when it fetches.

devtools generalized loading…

There's been some updates..

Sigurd added the Network.loadNetworkResource protocol method for sourcemaps. He recently added metrics to record how various load attempts succeed/fail.

AFAICT, there are currently four different types of fetch attempts... you can see them in PageResourceLoader.js's _dispatchLoad() in the devtools source.

Here's the umbrella crbug.

You can also see some of this in the new "Developer Resources" drawer panel in devtools:

[adamraine]

CSP sensitivity.. I didn't test, but I believe a frame-src directive would introduce problems for this technique

It does #12044 (comment)

[connorjclark]

Network.loadNetworkResource was added in M87, stable is currently M88. We might consider checking the chrome version and using the current implementation as a fallback for awhile.

[brendankenny]

Network.loadNetworkResource was added in M87, stable is currently M88. We might consider checking the chrome version and using the current implementation as a fallback for awhile.

What are y'all thinking for a timeline for this? Stable is now m89, earliest #12199 could land is m91, so the main reason to hold back is CLI users using an old stable Chrome/specific Chromium build?

[connorjclark]

yea, exactly. It seems people using docker images are eternally a few more versions behind than you'd expect..

v8?

[paulirish]

content-type sensitivity. In #12064 we see that if a map is served with application/octet-stream

https://proxx.app/ also reproduces this behavior.

[connorjclark]

We're using the new protocol method to fetch resources now, but we need to keep the old one around for LR: #13006

Let's track in #13394 – next month we should be able to drop the iframe fetcher completely.