For future reference, yes, this is what I want, specifically. The author should not be extending a Base method if the base documentation is generic and goes against the behavior they want to implement. At very least, if they do this then they should create a documentation for the specific method (over their data structure) with a warning.
I want package authors to not ignore performance guarantees set by the Base, so I can get a datastructure of some package and pass it to some algorithm that I know the complexity without needing to check if the author has violated any of the performances guaranteed. If it works then it is the right complexity. Simple. If I really do not care about the performance degradation I can carefully extend Base.last myself for the third-party data structure.
You are basically saying it is misleading the fact that package authors are not perfect programmers and may make mistakes that compromise API guarantees… there is no way to guarantee many things except by believing package authors will try to be faithful to the API, and disregarding packages that you discover to fail to do so.
I opened a PR in LinkedLists.jl to get this corrected, if possible.
And when the programmer hits that point, it will be clear to them what is the expected overhead. They may decide to not support iterators, or spend an extra minute to do a loop that avoids the extra allocation, or save that value when it was readily available before, or even rethink their datastructure. It seems to me a problem of bad programmers, not bad API.