Has anyone considered whether copy-on-write semantics would be useful for safeBuffers? Under a profiler, it looks like my application spends an astonishing amount of time in the memcpy() in safeBuffer::operator= (const safeBuffer &). If safeBuffer strings are typically written after being copied, the buffer copy is unavoidable, but if they're generally just read, it might be more efficient to keep a reference count and just copy the pointer. Making a new copy of the buffer would be deferred until the string needed to be updated, if ever.
I'm not particularly familiar with the code base, so I'm not sure how invasive this would be, and whether it is likely to yield any real benefit. Before I contemplate undertaking an analysis, I'm wondering Berin or anyone else has any thoughts as to whether it's likely to be worthwhile.
