Add vector-any, vector-every, string-any, string-every

[gambiteer]

I modeled the new code on the code for vector-for-each.

[lassik]

As a user: thank you. This is the kind of unexciting but very important work that makes people start a given Scheme implementation by default.

[gambiteer]

Marc: I just studied what you did for any and every in lib/gambit/list/list.scm; you recently added lines like

            (or (macro-auto-force (apply pred args))
                (any-n* rests next-rests))

It's not clear to me why you added macro-auto-force, but should I add lines like this to vector-any and vector-every?

[feeley]

"auto-forcing" is enabled by a Gambit configuration option. It is intended to make (force arg) implicit on all strict operations. For example, (+ (delay 1) (delay 2)) is equivalent to (+ (force (delay 1)) (force (delay 2))). In the code you quote, the result of the pred procedure needs to be tested by the or, so the or operation is also a strict operation that needs to force the result of the pred procedure. Otherwise, if the pred procedure is (lambda (x) (delay (< x 0))) then the or would always think the result is non-false.

So yes this needs to be done for vector-any and vector-every?.

When Gambit implements (future <expr>) with promises and threads, it will allow the elegant expression of parallelism with (+ (future E1) E2).

[gambiteer]

I'm looking at this now. In _std#.scm the arguments to proc in both vect-map and vect-for-each are not wrapped in macro-auto-force. Those should be wrapped, too, right?

Edit: The answer is no, sorry, I got confused.

Edit^2: That question didn't even make sense, but this is pretty subtle, you don't autoforce the last call to proc in tail position because it's returned.

[feeley]

The result of the procedure should not be auto-forced, neither for vector-for-each or vector-map, because the vector constructor is not strict. It is a different situation for u8vector-map because the u8vector constructor is strict.

[gambiteer]

(a) I guess I don't know what "strict" means.
(b) I can't find u8vector-map anywhere in the sources.

[feeley]

A function is "strict" on argument x if it needs to know the value of x to compute the result of the function. For example (sqrt x) is strict. A function is not strict on an argument x if it does not need the value of x to compute the result, or it only needs to know the identity of the value. For example (list x) is not strict. The point is that if x is represented as a placeholder for the result of some other parallel computation, then you can just put a reference to that placeholder in the list created by (list x), i.e. you don't need to wait for the other computation to be done before creating the list.

For u8vector-map I was just making an example. If it existed it would need to auto force the result of the function because there's no way to put a reference to a placeholder in a byte (but that's just implementation... it could be designed/implemented differently, for example if u8vectors had a full word of memory for every element of the u8vector).