> So we have something like this:
>
> Platforms known to not support any sort of O(1) realloc: JVM
>
> Platforms that may not support O(1) realloc: OS X, others?
>
> Platforms that do (should?) support O(1) realloc: Linux
>
> In any case, I still see that there's value in this feature:
>
> * It would help JRuby and all runtimes that run on non-efficient-realloc platforms.
> * It does no harm and matches Array.new behavior.
> * For folks doing crypto stuff that want to know exactly how big the buffer is right away, this provides a way to do so.
>
> I won't try to argue whether realloc is consistently efficient across platforms or not. It seems like it's not guaranteed to be on any platform.
>
> It's also such a tiny addition...why not?

I don't imagine a lot of people take a string.buffer game for
optimization if it doesn't
have big benefit. now, this feature is unclear how much useful out of
jvm and how
much useful on jvm. afaik, nobody show realistic benchmark result nor
encompassing
affect  platform lists. I'm not incline to agree _guess_ game.

If the benefit is not so much, the feature will be dead and forgotten quickly.