Christophe Rhodes <csr21@...> writes:
> Also, feel free to play about with this, come up with useful semantics
> for sequences and funcallable sequences, and report back if it is
> actually of use to you.
Some time later...
... I think I have a handle on what I want to do with this. I would
very much like it if people could test the patch at
buildability and regressions (both functionality and speed), and then
in conjunction with
<http://www.doc.gold.ac.uk/~mas01cr/tmp/sequences.tar.gz> for the
ability to do stuff with non-standard sequences.
So, here's the deal. I actually think that there are useful semantics
to be had from having user-extensible sequences; much more than with,
say, user-extensible numbers; and as such I would like to merge the
patch or something a bit like it. I am agnostic as to whether this is
too destabilizing (or too against the SBCL philosophy of providing
just what the spec says) for SBCL 1.0; I'm basically happy to wait if
that's the consensus. (Additionally, though I haven't had to touch
the compiler /very/ much I'm aware that I might have zapped some
optimizations dear to Bill's heart (the stuff specialized on
simple-vector for COERCE and friends, for instance); the patch really
could do with some careful review: trust me on this one (or rather,
don't trust me :-).
You'll find in the tarball some documentation(!) as well as some code,
but no test cases. The truly enthusiastic could supply me with some
of those, too.