Mailing-List: contact dev-help@apr.apache.org; run by ezmlm
Precedence: bulk
From: "Greg Marr" <gregm@alum.wpi.edu>
Subject: Re: cvs commit: apr/threadproc/win32 thread.c
To: dev@apr.apache.org
Date: Sat, 01 Sep 2001 12:29:53 -0400
Message-ID: <web-411449@dc-mxdb03.cluster1.charter.net>
In-Reply-To: <JLEGKKNELMHCJPNMOKHOEEOKEHAA.striker@apache.org>
MIME-Version: 1.0
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit

On Sat, 1 Sep 2001 11:18:47 +0200
 "Sander Striker" <striker@apache.org> wrote:
> > rbb         01/09/01 00:29:07
> >    APR_DECLARE(apr_status_t) apr_thread_once(apr_thread_once_t
> *control,
> >                                              void (*func)(void))
> >    {
> >   +    /* Quick escape hatch, and bug fix.
> 
> This still isn't correct.  Now we have a race condition that
> could result in the function never being called instead of once.
> This looks better IMHO:
> 
>         if (!InterlockedExchange(&control->value, 1))
>             func();
> 
>         return APR_SUCCESS;

To also include the quick escape hatch:

/* Save the system call if InterlockedExchange can't possibly return 0.
The InterlockedExchange function returns the *old* value of the
variable.  If it is 0, this is the first function to make it 1, and
so the function should be called. */
if ((0 == control->value) && (0 == InterlockedExchange(&control->value,
1)))
    func();

return APR_SUCCESS;

BTW, I looked at doing this with InterlockedIncrement, since it returns
the new value, except that Win95 is broken, and it just returns non-zero
on success, not necessarily the new value.