ichthyo/LUMIERA.clone
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Library: investigate what could be tested with floating-point

Browse source 
The floating-point variant turns out to be difficult to test,
since it is never ''actually exact'' -- due to the nature of floating-points.

We can not expect reasonable behaviour at the domain boundaries,
even while in some cases the computation works (and in others it
goes infinite and is sanitised to `minVal`).

The limit for rejection is itself rather a heuristic;
thus it is difficult to show "wrong" behaviour beyond the limit,
while the limit itself is typically already so large that it is
hard to handle in test code -- at least when expecting visible results.
This commit is contained in:
Fischlurch 2026-09-15 17:32:05 +02:00
parent 078aa48879
commit 0c69db19d7
2 changed files with 101 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/lib/util-quant.hpp
View file

@ -216,7 +216,7 @@ namespace util {
offset += PERIOD;
if (not (offset < PERIOD)) // can happen due to ε or non-finite values
offset = 0;
F res = minVal + offset; // could be rounded to maxVal
F res = minVal + offset; // could be rounded to maxVal (or become infinite)
return res < maxVal? res : minVal;
}
@ -245,6 +245,7 @@ namespace util {
limit_cyclicWrap (F period)
{
REQUIRE (std::isfinite (period));
REQUIRE (F(0) < period);
F limit = period / std::numeric_limits<F>::epsilon();
if (not std::isfinite (limit))
limit = std::numeric_limits<F>::max();

101
tests/lib/cyclic-wrap-test.cpp
View file

@ -27,7 +27,7 @@ using ::Test;
//using util::isnil;
//using util::_Fmt;
using util::isLimited;
using std::floating_point;
namespace util {
namespace test {
@ -42,6 +42,9 @@ namespace test {
template<typename NUM>
constexpr NUM _MIN = std::numeric_limits<NUM>::lowest();
template<floating_point FLO>
constexpr FLO _EPS = std::numeric_limits<FLO>::epsilon();
} // (End) Test helpers
@ -75,6 +78,8 @@ namespace test {
verifyIntegerTypes<uint32_t>();
verifyIntegerTypes<int64_t> ();
verifyIntegerTypes<uint64_t>();
verifyFloatTypes();
}
@ -105,7 +110,7 @@ namespace test {
const I LOWER(-100); // note: expected to wrap for unsigned types
const I UPPER(LOWER+5);
auto cyc = [&](I val){ return cyclicWrap (I(val), LOWER,UPPER); };
auto cyc = [&](I val){ return cyclicWrap (val, LOWER,UPPER); };
CHECK (LOWER == cyc(LOWER+0) );
CHECK (LOWER+1 == cyc(LOWER+1) );
@ -134,6 +139,98 @@ namespace test {
CHECK (LOWER + offset == cyc(_MIN<I>) );
}
void
verifyFloatTypes()
{
double maxDom = _MAX<double>;
double super = _MAX<double> - _MIN<double>;
SHOW_EXPR(maxDom)
SHOW_EXPR(super)
SHOW_EXPR(_EPS<double>)
SHOW_EXPR(limit_cyclicWrap<double>(1))
SHOW_EXPR(limit_cyclicWrap<double>(1024))
SHOW_EXPR(limit_cyclicWrap<double>(1.0/1024))
SHOW_EXPR(limit_cyclicWrap<double>(_EPS<double>))
SHOW_EXPR(limit_cyclicWrap<double>(maxDom))
double limit5 = limit_cyclicWrap<double>(5);
double atLimit = limit5 * (1-_EPS<double>);
double stepLim = std::floor (limit5 / 5) * 5;
SHOW_EXPR(limit5*_EPS<double>)
SHOW_EXPR(limit5)
SHOW_EXPR(stepLim)
SHOW_EXPR(atLimit)
SHOW_EXPR(cyclicWrap (10.0 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (limit5 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (atLimit ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim-5 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+1 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+2 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+3 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+4 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+5 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+6 ,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+10,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+20,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+30,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+40,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+50,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+100,0.0, 5.0))
SHOW_EXPR(cyclicWrap (stepLim+101,0.0, 5.0))
SHOW_EXPL(cyclicWrap (limit5 ,10.0, 15.0))
SHOW_EXPL(cyclicWrap (limit5+1 ,10.0, 15.0))
double epsStep = _MAX<double>*_EPS<double>;
double toxicMin = _MIN<double> + 2*epsStep;
double toxicMax = epsStep;
double toxicPeriod = toxicMax - toxicMin;
SHOW_EXPR(epsStep)
SHOW_EXPR(toxicMin)
SHOW_EXPR(toxicMax)
SHOW_EXPR(toxicPeriod)
SHOW_EXPR(cyclicWrap<double>(0, toxicMin,toxicMax) )
SHOW_EXPR(cyclicWrap<double>(1, toxicMin,toxicMax) )
SHOW_EXPR(cyclicWrap<double>(1*epsStep, toxicMin,toxicMax) )
SHOW_EXPR(cyclicWrap<double>(2*epsStep, toxicMin,toxicMax) )
SHOW_EXPR(cyclicWrap<double>(3*epsStep, toxicMin,toxicMax) )
SHOW_EXPR(cyclicWrap<double>(0, 1, _MAX<double>) )
SHOW_EXPR(cyclicWrap<double>(1, 1, _MAX<double>) )
SHOW_EXPR(cyclicWrap<double>(2, 1, _MAX<double>) )
SHOW_EXPR(cyclicWrap<double>(-1,1, _MAX<double>) )
SHOW_EXPR(cyclicWrap<double>(-epsStep,1, _MAX<double>) )
SHOW_EXPR(1 == cyclicWrap<double>(1, 1, _MAX<double>))
SHOW_EXPR(2 == cyclicWrap<double>(2, 1, _MAX<double>))
SHOW_EXPR(1 == cyclicWrap<double>(-1,1, _MAX<double>) )
double wrapped = cyclicWrap (-epsStep,double(1), _MAX<double>);
SHOW_EXPR(1 < wrapped)
SHOW_EXPR(wrapped < _MAX<double>)
auto cyc = [&](double val){ return cyclicWrap (val, double(-10),double(-5)); };
CHECK (almostEqual (double(-10), cyc(0)));
CHECK (almostEqual (double( -9), cyc(1)));
CHECK (almostEqual (double( -8), cyc(2)));
CHECK (almostEqual (double( -7), cyc(3)));
CHECK (almostEqual (double( -6), cyc(4)));
CHECK (almostEqual (double(-10), cyc(5)));
CHECK (almostEqual (double( -9), cyc(6)));
CHECK (almostEqual (double( -8), cyc(7)));
CHECK (almostEqual (double( -7), cyc(8)));
CHECK (almostEqual (double( -6), cyc(9)));
CHECK (almostEqual (double(-10), cyc(105)));
CHECK (almostEqual (double(5)/2, limit5 * _EPS<double>));
CHECK (limit5 + 1 == limit5);
CHECK (1 == cyclicWrap<double>( 1, 1,_MAX<double>) );
CHECK (2 == cyclicWrap<double>( 2, 1,_MAX<double>) );
CHECK (1 == cyclicWrap<double>(-1, 1,_MAX<double>) );
CHECK ( 1 < wrapped );
CHECK (wrapped < _MAX<double>);
}
};