I got numbers like this as I approached the slave (raw values before converting to meters):
```
Ranged: 8
Ranged: 6
Ranged: 3
Ranged: 3
Ranged: 5
Ranged: 5
Ranged: 5
Ranged: 6
Ranged: 1
Ranged: 1
Ranged: 800000
Ranged: 800003
Ranged: 800003
```
This is because the ToF becomes smaller than the correction factor resulting in a negative number.
This patch performs Sign Extension from 24bits to 32bits.
This result in returning a negative meter value which makes more sense than an impossibly big one.
* Use unsigned long when dealing with millis() and micros().
Although sizeof(uint32_t) == sizeof(unsigned long) on Arduino, this is
not the case on 64-bit Linux, where sizeof(unsigned long) ==
sizeof(uint64_t).
Most timestamp arithmetic and comparisons have been left alone, to
reduce code churn. This is fine, as uint32_t is perfectly wide to store
most timestamp deltas this library will deal with, and C will promote
the integer rather than do a narrowing conversion. The real problem
arises with narrowing conversions being done by assuming timestamps are
32-bit.
No functional changes intended for platforms where sizeof(uint32_t) ==
sizeof(unsigned long) (so most 8/16/32-bit platforms).
Signed-off-by: Elizabeth Myers <elizabeth.jennifer.myers@gmail.com>
* Change most timestamps to use RadioLibTime_t.
This makes it obvious what is and isn't a timestamp.
Not everything has been converted; anything dealing with protocol and
chip-level timestamps has been left alone on purpose, to make it clear
that these functions do require 32-bit timestamps.
No functional changes intended on platforms where sizeof(uint32_t) ==
sizeof(unsigned long).
Signed-off-by: Elizabeth Myers <elizabeth.jennifer.myers@gmail.com>
* Use uint32_t internally in getTimeOnAir.
We need to not overflow the integers with the shifts and
multiplications, so this is correct behaviour.
Signed-off-by: Elizabeth Myers <elizabeth.jennifer.myers@gmail.com>
---------
Signed-off-by: Elizabeth Myers <elizabeth.jennifer.myers@gmail.com>
