[ModbusSerialClient] On higher baudrates _silent_interval is too small to be enforced

[vermut]

At least say if I'm right or wrong.

Code at: https://github.com/riptideio/pymodbus/blob/master/pymodbus/client/sync.py#L367

Baudrate: 57600
_silent_interval: 0.000668402777778
typical ts: 1499640168.84
typical _last_frame_end: 1499640168.83

It appears quite often ts == self._last_frame_end + self._silent_interval due to float round error. Even if it's 0.00001 smaller.

[vermut]

P.S. I'm not sure if time.sleep(0.000668402777778) will do something meaningful instead of being rounded up to 0.

[dhoomakethu]

I agree , its been on my plate for sometime. from modbus serial specification Modbus_over_serial_line

Remark :
The implementation of RTU reception driver may imply the management of a lot of interruptions due to the t1.5 and t3.5 timers. With
high communication baud rates, this leads to a heavy CPU load. Consequently these two timers must be strictly respected when the
baud rate is equal or lower than 19200 Bps. For baud rates greater than 19200 Bps, fixed values for the 2 timers should be used: it is
recommended to use a value of 750µs for the inter-character time-out (t1.5) and a value of 1.750ms for inter-frame delay (t3.5). 

[xtofl]

Can these precisions even be guaranteed by the (non-realtime!) python runtime? Should perhaps some 'self-check' be added to estimate how many times a timer will not be exact, on a specific system, under specific load?

[dhoomakethu]

Thats a good thing to have but not a high priority item to be considered for development. Though the timings on python is not guaranteed, it should work for most of the cases considering the fact that majority of the hardwares/OS combination on which pymodbus is being run is good.

[xtofl]

Ok. Thanks for the quick reply!

[dhoomakethu]

fixed with 1.3.2