readHexRational missing from Numeric

The HexFloatLiterals extension was merged as result of https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0004-hexFloats.rst

As part of this proposal showHFloat was added to Numeric to output floating points in hexFloat format, unfortunately, no one seems to have remembered to add readHexRational or a similar function to parse hexFloats to Numeric. This leads to the awkward scenario that I can easily produce hexFloat values, but have no way to parse them from String/Text data...

All the code already exists in compiler/util/Utils.hs, someone just needs to cherry pick the implementation add add it to Numeric in base.

Trac metadata

Trac field	Value
Version	8.6.3
Type	FeatureRequest
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	libraries/base
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tfeature request Trac import core libraries labels

As carter pointed out on IRC, (and asked me to echo here) this would also improve our IEEE754 compliance.

I've started work on this here. I've tried to copy readFloat as closely as possible.

mentioned in merge request !268

added Pnormal label

i had some discussions about this on irc today... theres some issues in how show instances for Float work at the moment? (perhaps there issues in Read too, but the following should demonstrate issues in how floating point notation is printe)

> show $  (+ 0x0) $ (realToFrac :: Rational  -> Float) $ (fromRational  0x1.fffffe000981dp127 :: Rational )
"3.4028235e38"
> show $  (+ 0x0) $ (realToFrac :: Float  -> Double) $ (fromRational  0x1.fffffe000981dp127 :: Float )
"3.4028234663852886e38"

so theres something wrong with the Show machinery in showHFloat and Show instance of Float!

some better examples are

Prelude Numeric> showHFloat  (0x1.fffffdp127 :: Float) ""
"0x1.fffffcp127"

which is straight up data corruption

Prelude Numeric> (0x1.fffffdp127 :: Float) ==  0x1.fffffep127
False
Prelude Numeric> (0x1.fffffdp127 :: Rational) ==  0x1.fffffep127
False
Prelude Numeric> (0x1.fffffdp127 :: Rational) ==  0x1.fffffdp127
True
Prelude Numeric> (0x1.fffffdp127 :: Double) ==  0x1.fffffdp127
True
Prelude Numeric> (0x1.fffffdp127 :: Float) ==  0x1.fffffdp127
True

@bgamari @RyanGlScott @yav ... i think showHFloat and the show instance of Float have some data corrupction going on :( ... which maybe should be a different ticket?

to be clear, i think the issue is specifically with showHFloat... not Show?

eg:

3.4028235e38 == (0x1.fffffep127 :: Float)
True
``

Yes, this sounds like a distinct issue.

There might very well be a problem here, but I am not sure which if these is wrong. In particular the number 0x1.fffffdp127 cannot be represented exactly in a Float. Tbh, I don't recall which rounding mode we use in such situations.

#18440 (comment 287332) i had opened a ticket on the other issue, though subsequent comparisons left me a touch confused and less confident that theres a bug. @yav and @bgamari i closed it, but would appreciate a second look there

mentioned in merge request !12695