Act on compass read return value

[martinbudden]

Stops ignoring the return value of the compass read function. Takes action if read returns false.

[stronnag]

Very interesting. I've just rebuilt the quad with the dodo to eliminate the GPS/compass, so it'll probably be Monday before I try this.

[digitalentity]

Good catch! HMC5883 is extremely picky. A few corrupt readings in self calibration and compass will be totally off. Next thing would be signalling this to IMU and navigation to take appropriate action, but that's a task for another PR.

[digitalentity]

We still need to return zero vector here so IMU can detect invalid compass and act accordingly.

[martinbudden]

I've updated to return a zero value in sensors/compass.c if the mag.read fails.

One other thought: in initialisation it tries to collect 10 samples, if any read fails there will be less than ten samples. Should I change it to collect 10 valid reads? And put in some protection so that if the reads continually fail it does not get stuck in the loop?

[digitalentity]

Yes, I think we should collect ten valid readings and if it times out just return false. Or maybe (for safety reasons) if we got at least one failed read/write during detection process - report no compass.

[martinbudden]

OK, I've improved the handling of invalid readings in initialisation.

The init function is void not bool - I think that should be changed as well, but that is more pervasive, so I'll wait until @stronnag has tested this before I make that change.

[digitalentity]

I agree to both:

• we need to test and merge this first
• init function should be able to fail as well and the result of failure should be a non-present sensor indication.

[digitalentity]

I think it's safe to merge this right away. @martinbudden is this ready to merge?

[digitalentity]

See #514 for further discussion.

[martinbudden]

Yes, you can merge this now.