I was going to disagree with this because I thought "but what about the tests!", but in the linked video of Rob Pike's talk he says (paraphrased) "but then of course there's a test, so that every time it is tested, it guarantees that the library and the copied code agree on their definition. The test has a library dependency but the copied code doesn't".

That's actually a really clever way to do things and I think I'll adopt it.