RealFloat Zero Padded Exponent

[BebeSparkelSparkel]

Some dxf files have doubles encoded in scientific notation with the exponent padded with zeros.

https://github.com/LibreCAD/LibreCAD/blob/8e10245994d8307a76cedc3cbffe471de88c8325/librecad/support/library/algoritm/alg27.dxf#L1560

I suggest adding a Bool to FGeneric and FScientific to allow for padding the exponent with zeros to the maximum precision of the exponent.

[clyring]

Perhaps the way to support this is a higher-order interface that takes displayExponent :: Int -> Builder. Then you could write such a function that prints an exponent of 0 as E+000 or one that dislays it as e0. That seems a bit less nasty than having a bunch of flags for 'is e capitalized' and 'should a + be written for positive exponents' and 'should the exponent be zero-padded'.

Another idea is to provide a function of type Double -> (Builder, Int) where the Builder prints just the mantissa and the Int is the exponent.

[BebeSparkelSparkel]

This is much more flexible. Thank you for the idea.

[BebeSparkelSparkel]

@clyring I have been unable to make your suggestion performant. I have tried several ways but am consistently getting results that are 150-300% slower. Could you make some recommendations on how modify the overhaul-realfloat branch to implement your suggested change?

[BebeSparkelSparkel]

This seems to be going beyond the intended capabilities of the intended formatting capabilities.

[clyring]

@clyring I have been unable to make your suggestion performant. I have tried several ways but am consistently getting results that are 150-300% slower. Could you make some recommendations on how modify the overhaul-realfloat branch to implement your suggested change?

I'd expect there is a practical way to get an appropriately general interface for this without requiring such a performance hit. Perhaps using BoundedPrim instead of Builder helps? Or maybe we need to get the right amount of inlining to happen? But I don't have the spare focus to dig into this yet.

But it would be nice to provide an interface flexible enough to support this. Let's leave the issue open for now.

[BebeSparkelSparkel]

I have been thinking about this and perhaps the most flexible option for printing both the mantissa and exponent would be to extract F2S, D2S, and their dependencies that calculate the digits word and exponent integer to a new package. This would allow the most flexibility for custom printing of the mantissa and exponent and would allow other text types to use it for their own printing implementations.

[BebeSparkelSparkel]

Also, a problem with your proposed approach. FStandard does not use this so formatFloating cannot return (Builder, Int).