On Fri, Dec 4, 2015 at 6:52 AM, Alex Shinn <xxxxxx@gmail.com> wrote:

As I understand, the SRFI is attempting to do two orthogonal things:

1. provide a cursor-oriented API for strings (to account for non-O(1) indexing)
2. provide a cheap substring API (to remove the optional start/end-arg API hell)

Actually, at this moment, we're no more sure if substring is expensive than

if string-set! is cheap. I mean, we no longer count on string-set! being cheap

on all implementations. Then why do we need to stick to the conception that

substring is expensive? I think it should work reverse----if useful code can be

expressed concisely with cheap substring, that would motivate implementations

to adopt such strategy.

We can't immediately drop string-set!, but we can say mutating a string

may invalidate certain structures derived from it (e.g. cursors), or,

"it is an error to mutate a string after it is passed to certain APIs".

Note that the same loose restrictions are already there, e.g. it would be

undefined to mutate a string which is used as a key of hash tables.

We still need separate APIs for charcter-index access and cursor-based

access, but at least we can define them on a single opaque object

which is called string.