Mantissa retrieval?

[jordanlewis]

Hi,

We're revisiting potential use of this library for https://github.com/cockroachdb/cockroach. https://github.com/cockroachdb/apd has served our needs well so far, but we're taking another look here because of performance. apd.Decimal always uses an "inflated" representation, which is good for code simplicity, but not for performance. Now that this library is tested against GDA (#117), we'd like to reconsider using it for the improved performance offered by the decimal.Big compact representation.

One difficulty (that I think was raised long ago, back in 2016) is that we rely on retrieving the mantissa of a decimal in order to encode it on disk. I'm not sure how to retrieve the mantissa of a decimal.Big when its either in compact or inflated form.

Would you be able to offer some advice, please? Perhaps this is a simple question - I'm by no means an expert on decimal representation.

Here's the exponent/mantissa retrieval code that we'd need to implement against decimal.Big, for reference:

https://github.com/cockroachdb/cockroach/blob/master/pkg/util/encoding/decimal.go#L96-L148

[ericlagergren]

Commenting here to remind myself to read this when I get home today.

[ericlagergren]

Immediately, you could use Raw. If the uint64 is not 1<<64-1, then it's "valid." Otherwise, the big.Int is. Or even Decompose could work.

However, like I mentioned before I am open to a more specific API.

[jordanlewis]

Thanks. After spending time playing with the code, I now understand something I didn't before, which is that the compact and inflated representations are both "just" the mantissa, without any special encoding. I think the Raw access method is perfectly fine for our use case, so I'll just close this.