Semaphore-count

[Craig Allen]

So I have this snippet pinched from Rosetta Code:

(define-syntax-rule (define/atomic (name arg ...) body ...)
  (define name
    (let ([sema (make-semaphore 1)])
      (lambda (arg ...)
        (dynamic-wind (lambda () (semaphore-wait sema))
                      (lambda () body ...)
                      (lambda () (semaphore-post sema)))))))

But I would like to be able to ask whether I will have to wait, before I am in fact waiting (so I can print a message). I notice the fsemaphore package exports semaphore-count, but semaphore does not. Is this an oversight, or is there a better way to do this?

Thanks,

Craig

[Matthew Flatt]

There's no `semaphore-count`, but you can use `semaphore-try-wait?` to
poll a semaphore instead of blocking.

[Craig Allen]

I saw that function, but was scared off by its documentation:

Like semaphore-wait, but semaphore-try-wait? never blocks execution. If sema’s internal counter is zero, semaphore-try-wait? returns #f immediately without decrementing the counter. If sema’s counter is positive, it is decremented and #t is returned.

I'm not sure why I'd want it to decrement the count, I really just want to see if my critical section is running. I will try it though, thanks.

[George Neuner]

On 9/4/2018 8:31 AM, Craig Allen wrote:

I saw that function, but was scared off by its documentation:

Like semaphore-wait, but semaphore-try-wait? never blocks execution. If sema’s internal counter is zero, semaphore-try-wait? returns #f immediately without decrementing the counter. If sema’s counter is positive, it is decremented and #t is returned.

I'm not sure why I'd want it to decrement the count, I really just want to see if my critical section is running. I will try it though, thanks.

'wait' takes the semaphore when it is available.  'try-wait' takes the semaphore if possible, or fails immediately if it can't. 

AFAIK there is no way to simply 'check' if the semaphore is available.  In a multithread environment, such checking doesn't do much good anyway ... a positive counter value doesn't mean you actually can take a semaphore - some other thread(s) may beat you to it.

The canon [language independent] methods of "optimistic" locking are either to spin until 'try' returns true, or to combine one or a few initial probes using 'try' with a fall back to a normal wait if the probes fail.

George

[Jos Koot]

Hi Craig Allen

Would something like the following work?

Jos

 

(define (make-sema n)
 (define-syntax-rule (with-counter-sema expr ...)
  (begin
   (semaphore-wait counter-sema)
   (define result (begin expr ...))
   (semaphore-post counter-sema)
   result))
 (define counter-sema (make-semaphore 1))
 (define counter n)
 (define sema (make-semaphore n))
 (define (wait) (with-counter-sema (semaphore-wait sema) (set! n (sub1 n))))
 (define (post) (with-counter-sema (semaphore-post sema) (set! n (add1 n))))
 (define (count) (with-counter-sema n))
 (values wait post count))

 

(define-values (wait post count) (make-sema 3))
(count)
(wait)
(count)
(post)
(count)

---

From: racket...@googlegroups.com [mailto:racket...@googlegroups.com] On Behalf Of George Neuner
Sent: 04 September 2018 17:47
To: Craig Allen
Cc: racket users
Subject: Re: [racket-users] Semaphore-count

--
You received this message because you are subscribed to the Google Groups "Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to racket-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[George Neuner]

On 9/4/2018 11:46 AM, George Neuner wrote:
>
> AFAIK there is no way to simply 'check' if the semaphore is available.
>

I was wrong - there is a way using events: semaphore-peek-evt  can check
whether a semaphore is ready without affecting its counter.

https://docs.racket-lang.org/reference/semaphore.html?q=semaphore-peek#%28def._%28%28quote._~23~25kernel%29._semaphore-peek-evt%29%29


George

[Jos Koot]

Ignore my email on this subject, please.

It is wrong.

Jos

[Craig Allen]

Thanks, I will look into this.