timing of the inner loop not well defined

[bomilkar]

Looking at the inner loop (a bit simplified):

		// main data acquisition loop
		while(sock_err >= 0)
		{
			usleep(12500);
			pressure_measurement_handler();
			sock_err = NMEA_message_handler(sock);
		}

So the time per loop is definitely more that 12.5 msec. I can't say how much time is spend outside of usleep() but there is some I2C I/O involved which takes time, etc.

Have you thought about using a system timer to control the loop? Or is it not so important how much time passes between sensor reads? Think about vario=delta_h/delta_t !

[bomilkar]

I have quantified the issue. Maybe it helps making a decision.
I have sensord running along with variod, and a FLARM.
Counting the sentences in the nmea log from XCSoar I get the following numbers:
GPRMC = 1149
POV,P = 11634

So that means for 1149 GPRMC sentences I find 11634 POV,P sentences.
That means on average we get 10.125 POV,P sentences per second.
(Of course I cleaned up the log to the portion where it had a stable GPS fix and double checked the time stamps of the GPRMC sentences. 1149 GPRMC sentences means precisely 1149 seconds elapsed.)

So with usleep(12500) I would have expected 8 POV,P per second, and even less if we take into account that pressure_measurement_handler() and NMEA_message_handler(sock) take some time, too.

The bottom line: we are really far from a predictable timing.

[hpmax]

The PR I put in should address the timing issue by keeping track of the system time when the conversion is initiated and using the difference between the conversion times.