Hi,

Marc-André, I'm all for "real-life" benchmarks if someone proposes some.
Until that we have to live with micro-benchmarks, which seems to be the
method used for other CPython optimizations by the way.

You are talking about slicing optimizations but you forget that the
"lazy slices" idea has been shot down by Guido and others when proposed
by Larry Hastings (with a patch) some time ago. Also the "lazy slices"
patch was for 8-bit strings, which *are* variable-size objects, which
seems to counter your argument that variable-size Unicode objects would
prevent making such optimizations.

As I said the freelist changes actually have mixed consequences, and in
general don't improve much (and that improvement is just backed by
micro-benchmarks after all... why would it be more convincing than the
far greater improvement brought by making Unicode objects variable-size
objects?).

Why wouldn't you express your arguments in the python-3000 thread I
launched one month ago, so that at least there is a clear picture of the
different arguments for and against this approach?