Thank you for the fix 🙏

I don't think the PR fixes the behaviour; or at least, the provided test does not check for the panic I encountered.

Try running gno run on this file, from master:

package main

func main() {
	SmallestNonzeroFloat64 := 0x1p-1022 * 0x1p-52 // 4.9406564584124654417656879286822137236505980e-324
	DividedByTwo := SmallestNonzeroFloat64 / 2

	println(DividedByTwo)
}

You'll notice that the result is 0, as expected. Because this is not currently problematic in Gno code: the problem is not in floating point division per se (which works correctly), but rather in relation to constants, which have to use Bigdec for their underlying representation (in this scenario, when the division occurs, both numbers in the operation are already floats). Which is why, as in the OP of #1150, it is important that we instead test them as constants:

package main

const (
	SmallestNonzeroFloat64 = 0x1p-1022 * 0x1p-52            // 4.9406564584124654417656879286822137236505980e-324
	DividedByTwo           = SmallestNonzeroFloat64 / 2
)

func main() {
	println(DividedByTwo)
}

(To be clear, the expected behaviour here, matching what Go is doing, is to have DividedByTwo == 0).

They can be declared as constants within the function if scoping was your issue, I don't think that matters

So in the GH review today, I noticed something odd: the code wasn't actually duplicated to the float32 counterpart. In fact, a different test (float7) was still panicking.

Funnily enough, it wouldn't work with SmallestNonZeroFloat64 (as the code didn't have the same logic to check whether the number was actually zero, as it was for float64). Anyway, I fixed the issues.

Now only waiting on @piux2 's review.

The code appears to be well-implemented. However, it seems that the usage of panic in the original code is intentional. It serves as a mechanism to handle the conversion of untyped BigDec constants. The function db.IsZero() seems to be utilized to distinguish a BigDec with a value of zero. This might act as a placeholder, awaiting the completion of the BigDec implementation to handle actual db values. I'll leave this for @jaekwon to further review and comment.

db:

*github.com/cockroachdb/apd.Decimal {
Form: Finite (0),
Negative: false,
Exponent: -1075,
Coeff: math/big.Int {
neg: false,
abs: math/big.nat len: 40, cap: 44, [2178133438652175037,13358389938759785811,2327513931309058258,13281776902381683126,12428501436096707604,11167376665378208913,7533419439590254135,11111329854973851249,381100326086554668,743925905065200102,16358649908316238435,7965390482101853370,8683260283480532003,10397040223970786542,13141442883861695791,13481094988647533917,13769060133574195735,4357625168117272887,2481288102022783458,15672420522825115297,242328635129153437,8057563909054409931,17290527069613993211,8205538365884795466,5757056363039532081,13174075699421158699,5499449782126738648,17921724191600867845,4561879847826661174,12842551368738208122,8993367378774161191,14552919816609462385,15503360519369613010,4353283772766715668,1833667960781563132,4601466743984949394,1077832616916714926,4715285661293679468,953411222142837528,1],},}

Here is the reference:
#306 (comment)