Re: Patch to trunk: merge common code in hourglass impl

["Lennart Borgman (gmail)" <lennart.borgman@gmail.com>]

Jason Rumney wrote:
> Eli Zaretskii wrote:
>> Yes, atimers work by sending SIGALRM, which isn't supported on
>> Windows.  While Windows has equivalent functionality, no one stepped
>> forward to emulate atimers, probably because their use in Emacs is
>> quite limited, and in areas where Windows offers alternate ways of
>> doing the same.
>>   
> 
> What is the equivalent functionality? Threads are not the same as 
> signals, as they do not block other threads from executing. This is a 
> problem when any Lisp code is run, as a GC can occur in the main Lisp 
> thread while the signal emulation thread is running Lisp code, leading 
> to memory corruption and crashes. So any system timers need to be 
> handled in the Lisp thread, which effectively means they are 
> synchronous. Since the hourglass code doesn't actually need to run Lisp, 
> I used a Windows system timer running in the UI thread, which is not 
> generally busy enough to have delays like the Lisp thread, but I don't 
> think a general atimer implementation can be done that way.


Is it not possible to use critical sections (which is like a low cost in 
process semaphore)?