BigDecimal#round raises FloatDomainError regardless of exception mode

[mrkn]

BigDecimal::INFINITY.round shouldn't cause FloatDomainError when BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY) == 0 but it raises the error.

$ ruby -rbigdecimal -e "BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false); BigDecimal::INFINITY.round"
-e:1:in `round': Computation results in 'Infinity' (FloatDomainError)
	from -e:1:in `<main>'

[mrzasa]

1. It happens only when precision is 0 (defailt), it works well when we define a non-zero precision:

> BigDecimal.mode(BigDecimal::EXCEPTION_INFINITY, false); BigDecimal::INFINITY.round(1)
# => Infinity

2. With precision=0, #round converts a decimal to an integer and it's according to the docs

If n is specified and negative, at least that many digits to the left of the decimal point will be 0 in the result, and return value will be an Integer.
https://ruby-doc.org/stdlib-3.0.3/libdoc/bigdecimal/rdoc/BigDecimal.html#method-i-round

3. For the source-divers, the error is raised here:
   

   bigdecimal/ext/bigdecimal/bigdecimal.c

   Lines 2316 to 2318 in 1cb0950

   	if (round_to_int) {
   	return BigDecimal_to_i(VpCheckGetValue(c));
   	}

   
   because BigDecimal_to_i calls with always=1:

   bigdecimal/ext/bigdecimal/bigdecimal.c

   Line 1062 in 1cb0950

   GUARD_OBJ(p, GetVpValue(self, 1));

   
   and it means that the error for infinity is raised always, no matter what mode is set.

I believe it's an expected behaviour. All we could do is add a small note in the docs to be more explicit. WDYT @mrkn ?

[TomNaessens]

Same happens for NaN, looks like another side effect of the same case:

irb(main):029:0> BigDecimal('NaN').round(1)
=> NaN

irb(main):031:0> BigDecimal('NaN').round(0)
(irb):31:in `round`: Computation results to 'NaN'(Not a Number) (FloatDomainError)

irb(main):032:0> BigDecimal('NaN').round(-1)
(irb):32:in `round`: Computation results to 'NaN'(Not a Number) (FloatDomainError)