fix: Casting to float lose precision in some cases

[EnascutaRazvan]

What does this PR do?
I reproduced the case and it seems like when casting to float the precision is lost
Related to -> #10391

Motivation
Because i use this database everyday

Related issues
#10391

Additional Notes
I don't know exactly why that cast to float was used, but it seems generates some problems

[tglman]

Hi,

Thank you, just triggered the ci build on this, if all pass will merge it and port it to 3.2.x

Bye

[tglman]

Hi,

Tests are failing, so we do have cases where we rely on this conversion, looking in the code is possible to avoid this conversion specifying the double type at the end of the number, so something like: 35.927675D would this work for your use case ?

Regards

[tglman]

Hi,

Maybe another possible approach is replace the check: if (Math.abs(returnValue) < Float.MAX_VALUE) with a reconversion to the double value, to see if the value still matches, like:

float floatValue = (float) returnValue;
double check = (double) flatValue;
if(check == returnValue) {
 finalValue= flatValue;
} else {
finalValue = returnValue;
}

Can you verify if a check like this works in your case and update the PR?

[tglman]

One last thing (hopefully ...)

in case you update this PR can you add a test case for the case conversion error in OInsertStatementExecutionTest or in OUpdateStatementExecutionTest so we avoid to get regression in future.

[EnascutaRazvan]

I ll try it tomorrow, ty

[tglman]

Hi,

Tried to run the tests, they fail for formatting, you can run mvn spottless::apply to fix all the formatting issues.

(waiting for the test as well )