| Type | Bug | Status | reviewed | Date | 1-Jul-2009 10:03 |
|---|---|---|---|---|---|
| Version | alpha 65 | Category | Datatype | Submitted by | meijeru |
| Platform | All | Severity | minor | Priority | normal |
| Summary | Treatment of INF (and NaN) in 64 and 32 bit decimals |
|---|---|
| Description | Decimals are stored according to the IEEE standard 754-2008. This foresees special encoding of positive and negative infinity, as well as "not a number" (NaN). For the type decimal! these encodings cannot be stored, e.g. through conversion from binary. However, I have found at least a way to get +INF through defining a vector with 32 bit decimals and storing too large a number. This value is then printed out as 1.#INF (!). |
| Example code |
>> to-decimal #{7FF0000000000000} ;; official encoding of +INF in 64 bits
** Math error: math or number overflow
>> make vector! [decimal! 32] ;; cannot hold more than ca. 1E38
>> v/1: 1E39 ; push it beyond the limit
== 1.0e39
>> to-binary v/1
== #{7FF0000000000000} ; to-binary operates AFTER conversion of 32 bit +INF to 64 bits +INF!
>> v/1
== 1.#INF ; this is how it is formed/molded
|
| Assigned to | n/a | Fixed in | - | Last Update | 4-Aug-2009 20:07 |
|---|
| Comments | |
|---|---|
|
(0001170)
BrianH 1-Jul-2009 14:00 |
Sounds like this line:
>> v/1: 1E39 == 1.0e39 should have done some bounds-checking during the assignment. |
| Date | User | Field | Action | Change |
|---|---|---|---|---|
| 4-Aug-2009 20:07 | BrianH | Code | Modified | - |
| 4-Aug-2009 20:07 | BrianH | Category | Modified | => Datatype |
| 4-Aug-2009 20:07 | BrianH | Status | Modified | submitted => reviewed |
| 4-Aug-2009 20:07 | BrianH | Type | Modified | Issue => Bug |
| 1-Jul-2009 14:00 | BrianH | Comment : 0001170 | Added | - |
| 1-Jul-2009 10:03 | meijeru | Ticket | Added | - |