Fix various issues with handling of floating values within the LWM2M subsystem

[lodup29]

Describe the bug

• Regardless of its representation (TLV, JSON), it's not possible to pass a value > -1.0 and < 0.0 (ex: -0.5). There's no sign field in the float32_value and float64_value so the only way to represent a value such as -0.5 is :

struct float32_value lFloatValue; /* val1  and val2 are integers (s32_t) */
lFloatValue.val1 = 0;
lFloatValue.val2 = -500000;

However only the sign of val1 is considered for TLV while the JSON representation simply formats the value as ("%d.%d",lFloatValue.val1, lFloatValue.val2) ("0.-500000") thus resulting in an exception when attempting to parse the value on the other side.

• Using the JSON representation it is not possible to represent a value with leading zeroes for val2 (ex: 0.05). Indeed as such value should be reprensented as:

struct float32_value lFloatValue; /* val1  and val2 are integers (s32_t) */
lFloatValue.val1 = 0;
lFloatValue.val2 = 50000; /* 0.05*1000000 */

However as previously stated the value will simply end up being formated as ("%d.%d",lFloatValue.val1, lFloatValue.val2) ("0.500000" instead of "0.050000")

• At least some mantissas are one bit too short using the TLV reprensentation. Trying the represent 0.05 (0x3D4CCCCD using IEEE 754 single precision) ends up as 0.049999997 (0x3D4CCCCC) on the other side

To Reproduce

• Use the Leshan lwm2m server demo as the server backend

wget https://hudson.eclipse.org/leshan/job/leshan/lastSuccessfulBuild/artifact/leshan-server-demo.jar
sudo java -jar ./leshan-server-demo.jar -wp 80

• Select Single-value in the top right drop-down menu as JSON or TLV based on the test case
• Use the zephyr lwm2m client example and change the value of the temperature object as described by the following cases:

Representation	val1	val2	Expected value	Value retrieved by Leshan
TLV	0	-500000	-0.5	0.5
JSON	0	-500000	-0.5	Exception
JSON	0	50000	0.05	0.5
TLV	0	50000	0.05	0.049999997

[ioannisg]

@mike-scott, same question as before; wonder if this is still applicable, and, if so, if it could be addressed for 2.0

[mike-scott]

Thank you for the reminder. I will address.

[ioannisg]

Thank you for the reminder. I will address.

Just asking if this has this already been addressed, @mike-scott - few LWM2M Pr were merged lately..

[mike-scott]

@lodup29 @ioannisg I've setup a PR which addresses the first 3 items (and a few other bugs), however item 4 in the table above:

Representation	val1	val2	Expected value	Value retrieved by Leshan
TLV	0	50000	0.05	0.049999997

Is actually not a bug, but reflects a conversion issue going from fixed formatting to binary float format. This can be reproduced using the following website:
https://www.exploringbinary.com/binary-converter/

Enter "0.05" in the Decimal to Binary section at the top, and then click "Convert"
Resulting binary value is:
0.00001100110011001100110011001100110011001100110011001100110011 (repeating)

Copy that binary value into the Binary to Decimal section below and then click "Convert"
Resulting decimal value is:
0.04999999999999999995663191310057982263970188796520233154296875

NOTE: Tool truncates infinite binary fractions to 62 bits. I'm using less than that so value is truncated earlier.