linux "SetEvent" always return 0, won't it fail?

[romanholidaypancakes]

when WaitForEvent is not in a waiting state, using setevent should return failure

[mqudsi]

I’m sorry, can you clarify? Setting the event is completely independent (from the outside) from waiting on the event.

Example code that behaves differently on Linux and Windows would be helpful.

[romanholidaypancakes]

neosmart::while(WaitForEvent(event,-1))){

	// now doing
	do_something();
}

• when the thread1 is executing do_something, it is not in a waiting state(NOT CALLING WaitForEvent), SetEvent should return an error.

neosmart::SetEvent(event);

• linux: always return 0
  

  pevents/src/pevents.cpp

  Line 363 in 1209b1f

  int SetEvent(neosmart_event_t event) {

• windows: an error will be returned when the notification fails
  

  pevents/src/pevents.cpp

  Line 603 in 1209b1f

  int SetEvent(neosmart_event_t event) {

[yinsidij]

@inaryart I tried to repo what you described for SetEvent. However, seems like I didn't see the error. SetEvent for the same event twice would just return 1
Here is my example code
https://github.com/yinsidij/pevents/blob/master/win32/HelloWorld.cpp

Here is the output running in windows (ret = 1 twice, which indicates success)

Thread 14076 waiting for write event...
Thread 18516 sets event...
ret = 1, error = 0
WaitForSingleObject returns 0
Thread 18516 sets event...
ret = 1, error = 0

The output is the same no matter the event is auto-reset or manual-reset or initialState signaled or nonsignaled.
Am I missing anything?

[romanholidaypancakes]

@inaryart I tried to repo what you described for SetEvent. However, seems like I didn't see the error. SetEvent for the same event twice would just return 1
Here is my example code
https://github.com/yinsidij/pevents/blob/master/win32/HelloWorld.cpp

Here is the output running in windows (ret = 1 twice, which indicates success)

Thread 14076 waiting for write event...
Thread 18516 sets event...
ret = 1, error = 0
WaitForSingleObject returns 0
Thread 18516 sets event...
ret = 1, error = 0

The output is the same no matter the event is auto-reset or manual-reset or initialState signaled or nonsignaled.
Am I missing anything?

you are right, setEvent will be return 1, so i think linux should also return the same value(1), so that I can judge whether the wake-up(setevent) operation failed

[yinsidij]

I think neosmart::event uses return value of 0 as success, which is oppose to windows convention (return value of 1 as success). So you might want to update up-level checker.

[romanholidaypancakes]

I think this is normal code, it should be part of the project.

if(setevent(event) == 1){
   // then i known 
    failedCount++;
}

just as my suggestion...

[mqudsi]

I think you are misunderstanding what an error in this context means. The calls are fully asynchronous, and will succeed regardless of whether there is another thread on the other end - a SetEvent() on an already set event will succeed, and it doesn't rendezvous with the thread calling WaitForSingleEvent() before it returns. An error is only something out of the ordinary caused by either an internal bug in the pthreads interfacing or the user providing an invalid argument (e.g. making up a value for the event parameter and passing it in).

The convention on Linux and with the pthreads library (which this library serves as some sort of extension to) is for functions to return error numbers with a return value of zero being no error. Don't check for if (SetEvent(...) == 1) but rather if (SetEvent(...) != 0) because there are no guarantees made that one is (always or the only) error returned.