interpreter: unexpected floating-point behavior

[rvanasa]

Output when running floatTests.mo from the base library using the moc interpreter:

nan:0x8_0000_0000_0000 should be 'nan' (positive)
-nan:0x8_0000_0000_0000 should be '-nan' (negative)
-nan:0x8_0000_0000_0000 should be '-nan' (negative)
nan:0x8_0000_0000_0000 should be 'nan' (positive)
nan:0x8_0000_0000_0000 should be 'nan' (positive)
-nan:0x8_0000_0000_0000 should be '-nan' (negative)

copySign/keep positive NaN failed:


copySign/negate positive NaN failed:


copySign/keep negative NaN failed:


copySign/negate negative NaN failed:


copySign/second argument negative NaN failed:
1.2 was expected to be -1.2

copySign/NaN and positive infinity failed:


copySign/NaN and negative infinity failed:


copySign/positive infinity and negative NaN failed:
inf was expected to be -inf

copySign/negative infinity and negative NaN failed:
inf was expected to be -inf

9 tests failed.
test/.vessel/matchers/v1.1.0/src/Suite.mo:93.13-93.26: execution error, assertion failure

Repro command for convenience:

cd motoko-base
$(dfx cache show)/moc -r test/Float.test.mo --package matchers test/.vessel/matchers/v1.1.0/src --package base src

[rvanasa]

It also appears that Float.format and Float.toText behave differently in the interpreter compared to the replica.

Interpreter:

assert Float.toText(1.2) == "1.2";
assert Float.format(#exp 3, 123.0) == "1.230e+02";

Replica:

assert Float.toText(1.2) == "1.200000";
assert Float.format(#exp 3, 123.0) == "1.230e02";