nim cpp treats Nan as 0.0 (during compile time)

[timotheecour]

nim cpp -d:release -o:bin/nim_cpp compiler/nim.nim
bin/nim_cpp cpp -r main.nim

CT
nan
nan
nan
true
nan
true
nan
RT
0.0
nan
nan
true
nan
false
nan

import math

proc main()=
    let a = Nan
    let b = $a
    echo b # 0.0 BUG

    echo $Nan # nan
    let a2 = $Nan
    echo a2  # nan
    echo 0.0/0.0 == 0.0 # true BUG
    echo 0.0/0.0 # nan

    let a3 = 0.0
    echo a3/0.0 == 0.0 # false
    echo a3/0.0 # nan

static: echo "CT"
static: main()
echo "RT"
main()

this explains oddly failing CI tests in cpp mode

FAIL: tests/float/tfloatmod.nim C++
Test "tests/float/tfloatmod.nim" in category "float"
Failure: reExitcodesDiffer
Expected:
exitcode: 0

Gotten:
exitcode: 1

Output:
tfloatmod.nim(129)       tfloatmod
tfloatmod.nim(126)       main
tfloatmod.nim(106)       testSpecialCases
system.nim(3978)         failedAssertImpl
system.nim(3971)         raiseAssert
system.nim(3011)         sysFatal
Error: unhandled exception: /Users/travis/build/nim-lang/Nim/tests/float/tfloatmod.nim(106, 13) `(r == expected) or (r.isnan and expected.isnan)` -inf mod -inf == nan but expected 0.0 [AssertionError]

marking as high priority as it affects running cpp mode in CI normally (right now it's being run as allowed failure, which means if new regressions happen with cpp mode, we won't know): indeed it seems like root cause for 3 failures in cpp mode out of ~6:

• tests/float/tfloatnan.nim
• lib/pure/math.nim
• tests/float/tfloatmod.nim

EDIT:

• looks like the bug is during const folding and compile time