ZTEST: Add optional float/double comparison support

[microbuilder]

Currently the zassert_* functions in the ztest API are based on integer values, which can be accurately compared (1==1, etc.).

There are no options for single or double precision floating point value comparisons, which are error-prone to compare againist fixed operands (x==0.5f), particularly with single-precision values.

When CONFIG_FLOAT is enabled, options should be included to compare single and double precision values for equality within a specific margin.

In previous projects, I've implemented the following helper functions to work with floating point values in unit tests ...

bool f_is_equal(float a, float b, float epsilon);
bool f_is_at_least(float a, float b);
bool f_is_less_than(float a, float b);
bool f_is_greater_than(float a, float b);
bool f_is_within(float a, float upper, float lower);

... which allows tests like the following code, with control over the degree of prevision involved in the comparison:

#define F_EPSILON5  1E-5
TEST_ASSERT(f_is_equal(xyy.x, 0.44422f, F_EPSILON5, "2900K -> xyY"));

I understand the allergy to float in an embedded context, but if CONFIG_FLOAT has been set to y and is being used in the application or library, shouldn't the option be available to test said floating point values in both single and double-precision widths, using 'safer' value comparisons?

I can continue to use my own code to test these, but it seems reasonable to at least consider adding these helpers to ztest and ztest_assert.h with an appropriate check for CONFIG_FLOAT?

[aescolar]

@microbuilder : Would you consider preparing a PR with this enhancement?

[microbuilder]

Sure, I just wanted to check if there was strong resistance to including float support, but I'll put a PR together for this.

[microbuilder]

Closing for lack of activity. Will reopen as PR when ready.