test_rational() fails

[Tey']

One of the unit tests in testdmath fails with HEAD version:

#!
.\testdmath.exe
src\tests\testdmath.cpp[70]       Rational(HNumber("-1235000")/HNumber("78950000")).toString().toStdString()      [NEW]
        Result   : -343/21927
        Expected : -247/15790
38 total, 1 failed, 1 new

I lack of time to find out which commit added this regression.

[Pol Welter]

WOW...

This is just becoming weirder....

The constructor of Rational uses HMath::floor. It literally comes down to this very line (hmath.cpp[979]) again:

if (HMath::abs(n - nearest_int) < HNumber("1e-70") * HMath::abs(n + nearest_int))

Hadrien changed the tolerance to 1e-70 again (for good measure). That is what breaks Rational! In fact, we seem to require some value smaller than the resolution of SC. If I set it to 1e-100 it's fine.

I give up. I have literally not the slightest idea what's going on there.

[Tey']

On a side note, reverting to 1e-1000 "fixes" this issue, which probably explains why the test failure didn't show up when we initially reviewed pull request #47. Still, that doesn't make sense.

[Pol Welter]

Fixes issue #616

Also added some bits that should have been in commit b7b2fc6.

→ <<cset f3a379af3261>>

[Pol Welter]

Exactly. That's what I'm going to push now.