Enum Float/Double does not match Haskell Report
Section 6.3.4 of the report says that the Enum instance for Float and
Double should define the enumFromThen* functions such that
enumFromThen e1 e2 is the list [e1, e1 + i, e1 + 2i, ...]
where i = e2 - e1. The actual implementation in base is that of successive
additions (which is approximately the same thing for some types).
Successive additions causes issues with floating point rounding. Changing the definition to something like
enumFromThen_ e1 e2 = let i = e2 - e1 in map (\n -> e1 + n * i) [0..]significantly improves the accuracy of floating point ranges, as well as **matching what the report says**.
Prelude> last $ take 101 $ [0,0.1..] :: Float
10.000028
Prelude> last $ take 101 $ enumFromThen_ 0 0.1 :: Float
10.0Trac metadata
| Trac field | Value |
|---|---|
| Version | 8.0.1 |
| Type | Bug |
| TypeOfFailure | OtherFailure |
| Priority | normal |
| Resolution | Unresolved |
| Component | libraries/base |
| Test case | |
| Differential revisions | |
| BlockedBy | |
| Related | |
| Blocking | |
| CC | |
| Operating system | |
| Architecture |