So what's happening underneath that allows for this kind of speedup while still using reflection? Since it still uses reflection, what would it take to get something like this in stdlib? A much faster implementation would probably be a benefit to many.
I'm not sure I understand. This approach seems complicated and prone to issues as the author mentioned. Why not implement this as a middleware and simply emit those metrics from there, to stdout, your monitoring system or add them to a context?
Could someone remove the excessive tagging? Tagging something as go and golang on golangnews seems rather unnecessary and adds no value. Hack and hacking are pretty much the same thing too, one would suffice.