redirects audit perhaps too lenient

[connorjclark]

The redirects audit currently always passed when there is a single redirect present, even if significant:

It's been this way since it was introduced:

https://github.com/GoogleChrome/lighthouse/pull/3308/files#diff-dd1c921c5fb418fc98a727bbbfca7e0ca696b0441fd5c9b5ebf6353df72b2168R46-R50

The impact of hiding this audit in this case is not so bad, since there will be a very visible warning at the top of the report about a redirect. But, in PSI when use_original_url=true is set, the information is hidden away. Also, if we end up grouping audits by LCP phase we're going to want this audit to not show as passing when there is a real problem.

I couldn't work out why an allowance was made for a single redirect, does anyone know?

[brendankenny]

I couldn't work out why an allowance was made for a single redirect, does anyone know?

I ran into this when writing tests for #14838 and was very surprised.

The only hint I see is

Looking at psi it allows 1 redirect (not sure if we should only consider subdomains as allowed redirects).
https://example.com -> https://m.example.com

in #3308, so I assume it was some ancient allowance in old PSI for mobile site redirects back when that was more of a thing and it's carried through otherwise unquestioned.

Maybe worth more discussion but if the time wasted redirecting is significant then it's significant, regardless of if it was 1 or 2+ redirects. Seems like the count check should be dropped.

[adamraine]

With our focus on audit impact instead of arbitrary scoring criteria in primo prio, we should definitely do this.