And the followup is, when do you remove fallbacks from your code? Every time I think I should remove something I end up thinking, "well, it still works just fine, does it really matter if it stays for a little longer?"
The cost is in latency and complexity for users of your head browsers. How much do you want to reduce the user experience of your users on modern browsers to support users on old ones?
There isn't a one-size-fits-all answer for this, but you should have some idea of what your traffic breakdown by browser is, and ideally what your cost in conversion rate is for each additional ms of latency (this varies by industry). I don't remember offhand what the cost per byte in latency is, but IIRC we measured something like 1ms per 1K bytes at Google (this would work out to a 1 MBps = 10Mbps connection, which seems around right for typical cable/residential fiber households these days).
I'm not sure what you mean about "complexity for users of your head browsers". The idea of the fallback/polyfill is that it doesn't even come into effect for the modern browsers.
As far as latency is concerned, I hear you and that makes sense, but the things I'm talking about are not very big, so the latency gain isn't going to matter much for our purposes.
Yeah, we definitely look at the analytics. But even if it the old browser users are sub 1 percentage point, I see the number and think, "I could take this 5 line polyfill out, but then these 1000 uniques wouldn't work."
Polyfills are generally a good idea - they cost basically nothing for users of modern browsers, and they go away.
I thought that the issue in this thread is JQuery, which is a layer on top of the modern browser API (probably because the modern browser API didn't exist when it was created, and is in a large part a native implementation of JQuery functionality), and so incurs the cost of all its legacy compatibility support even if you don't need it.
