integrate() returns 0 for large finite upper bounds but correct result for Inf

help("integrate") already instructs us not to do that:

When integrating over infinite intervals do so explicitly, rather than just using a large number as the endpoint.

And as you've shown, for a good cause:

library(stats)
integrate(dnorm, 0, .Machine$integer.max)
#> 0 with absolute error < 0
integrate(dnorm, 0, Inf)
#> 0.5 with absolute error < 4.7e-05

Well, you can use anything larger than .Machine$integer.max*1e299, but why would you do that?!!

Looking at the source code we can see what's going on:

if(is.finite(lower) && is.finite(upper)) {
    wk