RTC_DS3231: correctly handle negative temperatures

[edgar-bonet]

As reported in issue #287, the method RTC_DS3231::getTemperature() fails on negative temperatures. This is because the conversion from raw bytes to a float-typed temperature assumes the value is always positive, wheres the datasheet specifies it is stored as a signed, two's complement integer.

This pull request fixes the method by changing the conversion code to this:

  int16_t temp = uint16_t(buffer[0]) << 8 | buffer[1];
  return temp * (1 / 256.0);

Some points worth noting:

• Casting buffer[0] to an unsigned type is meant to avoid signed integer overflow, which is undefined behavior in C++.

• Assigning the result to a signed integer makes numbers with the most significant bit set be interpreted as negative. There is no extra work to do, as all current processors use two’s complement representation internally.

• Multiplying by 1 / 256.0 (rather than 1 / 4.0) obviates the need for the 6-bit shift. Also, 1 / 256.0 being a compile-time constant, there is no division performed at run-time (which would be expensive).

• This code also reduces the number of floating-point operations to only two (int-to-float conversion and multiplication), v.s. four in the original code (two int-to-float conversions, one multiplication and one addition).

Tested down to −27.00°C by @i440bx.

Fixes #287.

[BillyGriffiths]

We ran into the same problem with negative temps. Do you have any idea when this will be merged?

[edgar-bonet]

@BillyGriffiths: No idea. This repo seems practically unmaintained. The last change to the source code of the library was the merge of pull request #257, on 2022-08-04. Since then, it has only had minute changes that do not touch the source code of the library itself.