On Tuesday, 15 November 2016 at 20:46:59 UTC, Steven Schveighoffer wrote:

> From experience, this smells like a race condition.

And I have found the error! And you are right! It is really race condition.
The problem was that I had not one receiveTimeout() but 3 receiveTimeout() for each thread. I don't know how I expect that each send must go to his own receiveTimeout, but of course it was false.

Big thanks to all!

On Tuesday, 15 November 2016 at 16:36:00 UTC, unDEFER wrote:
> The parent thread has the next code to receive it and handle:
> receiveTimeout( dur!"hnsecs"(1),
>                 (Tid tid)
>                 {//the message from child thread handler} );

 I just have to wonder every time I look at this... how long is 1 hnsecs? I'd assume not very long. Either blocking unless there's no more messages, or having a longer timeout period seems a better bet.

 Might make it 1-5ms.

On Wednesday, 16 November 2016 at 10:54:32 UTC, Era Scarecrow wrote:
>  I just have to wonder every time I look at this... how long is 1 hnsecs? I'd assume not very long. Either blocking unless there's no more messages, or having a longer timeout period seems a better bet.
>
>  Might make it 1-5ms.

Really I have wanted to write 0ms, but don't sure that it is possible.

On 11/16/16 4:35 AM, unDEFER wrote:
> On Tuesday, 15 November 2016 at 20:46:59 UTC, Steven Schveighoffer wrote:
>
>> From experience, this smells like a race condition.
>
> And I have found the error! And you are right! It is really race condition.
> The problem was that I had not one receiveTimeout() but 3
> receiveTimeout() for each thread. I don't know how I expect that each
> send must go to his own receiveTimeout, but of course it was false.
>
> Big thanks to all!

Glad you figured it out!

-Steve

On Wednesday, 16 November 2016 at 13:10:06 UTC, unDEFER wrote:
> On Wednesday, 16 November 2016 at 10:54:32 UTC, Era Scarecrow wrote:
>>  I just have to wonder every time I look at this... how long is 1 hnsecs? I'd assume not very long. Either blocking unless there's no more messages, or having a longer timeout period seems a better bet.
>>
>>  Might make it 1-5ms.
>
> Really I have wanted to write 0ms, but don't sure that it is possible.

receiveTimeout(0.seconds,
    /* ... */
);

On Wednesday, 16 November 2016 at 20:20:07 UTC, Anonymouse wrote:
> receiveTimeout(0.seconds,
>     /* ... */
> );

Thanks!

On Thursday, 17 November 2016 at 11:37:37 UTC, unDEFER wrote:
> On Wednesday, 16 November 2016 at 20:20:07 UTC, Anonymouse wrote:
>> receiveTimeout(0.seconds,
>>     /* ... */
>> );
>
> Thanks!

You can also do -1 for instant receive.