Michael Sperber:
>I just doubt it's worth the tradeoff.  Is there hard data that the
>optimizations you envision actually give significant performance
>gains?  I've always found non-shared strings plenty fast.

I know it's not hard scientific evidence, but at the company I currently
work for, we have a Scheme-based engine generating web pages, which uses
string-append in a fairly naive but straightforward way. It was way too
slow, and that slowness was largely accounted for by time taken up in
string-append. I implemented string-append/shared, switched code to use it
instead of normal string-append, and got something like a 100x speed
improvement, just from that change. Note that the obvious way to have made
the code in question use strings in a less naive way would have basically
been to implement the equivalent of string-append/shared, but at a higher
level.

-dan