| Anonymous | Login | Signup for a new account | 02-09-2010 12:30 UTC |
| Main | My View | View Issues | Change Log | Docs |
| Viewing Issue Simple Details [ Jump to Notes ] | [ View Advanced ] [ Issue History ] [ Print ] | ||||||||
| ID | Category | Severity | Reproducibility | Date Submitted | Last Update | ||||
| 0004315 | [Squeak] Kernel | minor | always | 07-23-06 20:44 | 10-15-09 20:26 | ||||
| Reporter | wiz | View Status | public | ||||||
| Assigned To | |||||||||
| Priority | normal | Resolution | won't fix | ||||||
| Status | closed | Product Version | 3.9 | ||||||
| Summary | 0004315: [Bug] Should (0.5 - ( 0.5 - 0.1 ) = 0.1 ) be true or false. | ||||||||
| Description |
For this one evauluate (0.5 - ( 0.5 - 0.1 ) = 0.1 ) in a work space compare to (0.5 - ( 0.5 - 0.2 ) = 0.2 ) |
||||||||
| Additional Information |
I am using sq 7046 with SqVm3.8.9b7 on an iMac G3 running OS 9.1 . The edge if equality for addition and subtraction for floats seems to be fail when the operands exceed a factor of 4 of each other in magnitude. Is this a bug? It seems to me a awfully high intolerance. Yours in service, --Jerome Peace |
||||||||
| Attached Files | |||||||||
|
|
|||||||||
 Relationships |
|
|
Notes |
|
|
(0006261 - 890 - 964 - 1468 - 1468 - 1468 - 1468) nicolas cellier 07-29-06 02:17 edited on: 07-29-06 02:20 |
It is not a bug. Float are inexact representation (rounded fractions) and you should never rely on a strict equality test after performing inexact operations, unless maybe for these special floats: integers (small ones) and fractions with denominator being power of 2. Just code the test in C and you'll see nothing better. Unless you are on x86 and use compiler optimisation, in which case intel floating point registers having extended precision you can go a little further with bits... But the result would depend on compiler options. BTW, 0.1 is not 1/10, though current Smalltalk implementation answer true to this test. That's maybe what makes you overtrust Float. (see http://bugs.impara.de/view.php?id=3374 [^] or http://groups.google.com/group/comp.lang.smalltalk/browse_thread/thread/59287ffa46b28a5a/b6a6e19f5975d098?lnk=gst&q=equality+is+not&rnum=1#b6a6e19f5975d098). [^] |
|
(0006263 - 381 - 435 - 435 - 435 - 435 - 435) nicolas cellier 07-29-06 03:35 |
If you want to play with decimal representation, you'd better use ScaledDecimal (0.5s1 - ( 0.5s1 - 0.1s1 ) = 0.1s1 ) (0.5s1 - ( 0.5s1 - 0.2s1 ) = 0.2s1 ) But do not trust what is displayed: ScaledDecimal display only a truncated result to the number of decimals you asked. Internally, it holds an exact fraction. (0.5s1*0.5s1) -> 0.2s1 (0.5s1*0.5s1) asFraction -> (1/4) |
|
(0013366 - 179 - 191 - 191 - 191 - 191 - 191) nicolas cellier 10-15-09 20:26 |
This is the natural behavior of Float operations which are inexact. This is a feature and cannot be solved. Use exact Fraction and ScaledDecimal when you expect exact arithmetic |
|
Issue History |
|||
| Date Modified | Username | Field | Change |
| 07-23-06 20:44 | wiz | New Issue | |
| 07-29-06 02:17 | nicolas cellier | Note Added: 0006261 | |
| 07-29-06 02:20 | nicolas cellier | Note Edited: 0006261 | |
| 07-29-06 03:35 | nicolas cellier | Note Added: 0006263 | |
| 10-15-09 20:26 | nicolas cellier | Status | new => closed |
| 10-15-09 20:26 | nicolas cellier | Note Added: 0013366 | |
| 10-15-09 20:26 | nicolas cellier | Resolution | open => won't fix |
| Mantis 1.0.8[^]
Copyright © 2000 - 2007 Mantis Group
45 total queries executed. 33 unique queries executed. |  |
