#5118: Numeric.floatToDigits fails for large floatRange
--------------------------------+-------------------------------------------
Reporter: claudiusmaximus | Owner:
Type: bug | Status: new
Priority: normal | Component: libraries/base
Version: 6.12.3 | Keywords:
Testcase: | Blockedby:
Os: Linux | Blocking:
Architecture: x86_64 (amd64) | Failure: Incorrect result at runtime
--------------------------------+-------------------------------------------
Changes (by daniel.is.fischer):
* cc: daniel.is.fischer@… (added)
Comment:
You get underflow with `floatRange _ = (minBound, maxBound)`:
{{{
floatToDigits base x =
let
(f0, e0) = decodeFloat x
(minExp0, _) = floatRange x
p = floatDigits x
b = floatRadix x
minExp = minExp0 - p -- the real minimum exponent
-- Haskell requires that f be adjusted so denormalized numbers
-- will have an impossibly low exponent. Adjust for this.
(f, e) =
let n = minExp - e0 in
if n > 0 then (f0 `quot` (expt b n), e0+n) else (f0, e0)
}}}
So with 64 bits of precision, you try to compute
{{{
(771*2^54) `quot` 2^(maxBound-63-(-54))
}}}
which will take some time and a lot of memory if it succeeds at all,
resulting in a 0 where a positive integer is required, leading to an
infinite recursion a few lines later.
Even with the halved `floatRange`, for `HugeFloat`s of very small or very
large modulus, `floatToDigits` will perform abysmally bad if it succeeds
at all.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/5118#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
