Re: [RFC] Some new vector procedures (vector-{memq, apply, to-string, ...})

[Alex Gramiak]

Stefan Monnier <address@hidden> writes:

>> I don't see any replacement for vector-apply, but I admit that the worth
>> of that one might not be worth the change at this point.
>
> I'm not sure I like vector-apply, but I am interested in doing something
> along these lines for cl-generic: the use of wrappers of the form
>
>     (lambda (x &rest args) ... (apply f x args))
>
> proves to be too costly in some cases (e.g. cl-print).  But the main issue
> seems to be the allocation of a new list for `args`, so `vector-apply`
> doesn't really help here.

I'm not sure how one would avoid the allocation lambda does without a
new special form (or byte compiler optimization that checks that you
only give args to apply), but perhaps I'm overlooking something.

There could also be a &rest-vector, but that's likely going too far.

>> For vector-memq, the Lisp implementations almost disallow it from being
>> used over memq/lists.
>
> I your `vector-memq` significantly faster than `memq`?

Not hugely; about 15-25% when vector-memq is made into a 2-byte bytecode
(as in my other thread) for a 5 element collection, and about 50% on a
50-element collection.

> My impression is that for small enough "sets", even if a bit faster,
> `vector-memq` wouldn't bring much benefit, and for larger sets if
> performance is an issue you'd likely prefer using a completely different
> structure such as a hash-table.

Right, I'm not under the impression that it would be a monumental
difference (that's why I called it a stupid itch :). Still, I think
there's a middle ground between lists and hash tables that vectors would
fit into nicely enough.