Re: Poor lockBeforeDate: implementation

[Adam Fedor]

On Thursday, July 17, 2003, at 08:54 AM, Richard Frith-Macdonald wrote:

> I was looking at this code, and it fairly horribly retries at 1 second 
> intervals.  There is an old comment in there that it uses sleep(), 
> (which has a 1 second granularity) because usleep() is 'not thread 
> safe under all architectures'
>
> Does anyone know if this is actually true for the architectures we 
> actually support?
> I'd like to rewrite this code to use [NSThread+sleepUntilDate:] (which 
> I have just updated to use nanosleep() if availalble) to start 
> sleeping for very small intervals, (perhaps starting with a 
> millisecond) increasing up to a maximum of 1 second intervals.  That 
> should make a big increase in responsiveness of applications using 
> lockBeforeDate: ... but I don't want to do it if it will break things 
> because the sleep mechanism used is not thread safe in some way.

I've heard that usleep does not work well on Solaris, but Solaris has 
nanosleep, so if you use that in preference, then it should be ok.