new_audit: bring back redirects-http with passive https check

[brendankenny]

hot-take/needs-discussion PR: Bring back the redirects-http audit, but without the extra pass.

• If requestedUrl is on http and
  • finalUrl is on https, pass
  • finalUrl isn't on https, fail
• if requestedUrl is on any other scheme, notApplicable

redirects-http was removed in #12643 due to a history of debugger protocol issues, the massive slowness of an extra pass, and the fact that Chrome switched to https by default if a user doesn't supply a scheme.

However there are still sites out there that should be redirecting to https and Lighthouse should continue to nudge the remaining 10-20% of sites people spend time on to do so if it's feasible.

We already have this audit in git history ready to go with just a few changes, and switching to only alerting when someone is purposefully testing their http site should take user complaints down to zero (for instance, the old check required the http version of a site to respond, if only to redirect), which also allows making the audit essentially zero-overhead by passively observing what the URL artifact is collecting anyways.

Thoughts? :)

[connorjclark]

Sorry for late approval.

[connorjclark]

We have a few redirects smoke tests, should it go there instead?

Otherwise, please add an exclusion to these expectations so we can still run these smokes in devtools runner.

[connorjclark]

@brendankenny branch updated

[connorjclark]

🤯

[adamraine]

🤯

[adamraine]

🤯

[connorjclark]

It's been two years but it seems like this would still be useful.

[adamraine]

Doesn't appear to be any downsides to this. I'll update for 12.0.

[adamraine]

Aaaaactually I'm pretty sure is-on-https already checks that the main document request is secure...

So maybe we just close this?

[brendankenny]

Aaaaactually I'm pretty sure is-on-https already checks that the main document request is secure...

Conceptually they're different checks. is-on-https is "was everything, end to end, sent over https", while redirects-http was specifically about the redirect. Separating them allows the redirect to be communicated more clearly ("you had an insecure request to http://example.com" vs "you didn't redirect from http://example.com to https://example.com").

The context of this PR is that there is a difference, and it would be little effort (and none of the complexity that prompted deleting the audit in the first place would be required) to keep the audit, so why not keep it.

On the other hand, yes, a lot of overlap, and the path to being able to pass both if you start a LH run on an insecure URL is very narrow (just preloaded/cached HSTS these days? Any other approah?).

In #3417 there was a proposal to relax is-on-https basically as #15907 does, then introduce a new "susceptible to a downgrade attack" audit. Now I don't know why adding a third audit seemed like a good idea, but that was a long time ago :)

My quick opinion two years later: I think it's still good to warn users when they're not using HTTPS or not upgrading, both of which #15907 communicates (albeit less clearly without a dedicated description). I wish we could do more with giving advice about the connection upgrade etc, but that's not the direction best practices is going, so it's fine.