https://issues.dlang.org/show_bug.cgi?id=15881
Issue ID: 15881
Summary: approxEqual Ignores maxAbsDiff
Product: D
Version: D2
Hardware: All
OS: All
Status: NEW
Severity: normal
Priority: P1
Component: phobos
Assignee: [email protected]
Reporter: [email protected]
The current implementation of approxEqual will ignore the maxAbsDiff term in
most circumstances.
The code below compares two floating point numbers. They are sufficiently far
away that on an absolute basis the assertion should fail. However, on a
relative basis, there is less than a 1% difference between the two.
import std.math : approxEqual, fabs;
void main()
{
auto x = 1000.0;
auto y = x + 10.0;
assert(approxEqual(x, y)); //should fail
assert(fabs((x - y) / y) <= 1E-2);
assert(!(fabs(x - y) <= 1E-5));
assert(!(fabs((x - y) / y) <= 1E-2 && fabs(x - y) <= 1E-5));
assert(fabs((x - y) / y) <= 1E-2 || 1E-5 != 0 && fabs(x - y) <= 1E-5);
//this is effectively the current implementation, wrong!
}
The exact line of the approxEqual code causing the problem is
return fabs((lhs - rhs) / rhs) <= maxRelDiff
|| maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;
It mirrors the last assert in the code above. The || operator short-circuits
the last part of the logical statement from being called. Perhaps changing it
to
return fabs((lhs - rhs) / rhs) <= maxRelDiff
&& maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;
will resolve the issue. Alternately,
return (fabs((lhs - rhs) / rhs) <= maxRelDiff)
| (maxAbsDiff != 0) && fabs(lhs - rhs) <= maxAbsDiff;
might resolve the issue.
--
