SYSTEM WARNING: Creating default object from empty value

SYSTEM WARNING: Creating default object from empty value

Mantis - Squeak
Viewing Issue Advanced Details
1602 Collections minor always 08-03-05 23:05 09-10-08 17:40
BGaertner  
KenCausey  
normal  
closed  
fixed  
none    
none 3.10  
0001602: [ENH] Interval method indexOf:
"Change Set: IntervalIndexEnh
Date: 15 June 2004
Author: Boris Gaertner

Interval inherits #indexOf:startingAt:ifAbsent: from SequenceableCollection.
The inherited method enumerates all elements until it finds a hit. Try

 (1 to: 100000 by: 0.01) indexOf: 99999

to see that this can be extremely time consuming. For Intervals, we can do
better.

This enhancement assumes that the elements of an Interval are Numbers.
It is tempting to use intervals that contain instances of DataAndTime, but
this would require the generalization of some methods. "
related to 0001603closed nicolas cellier [BUG][FIX] Interval method includes: 
related to 0006456closed andreas [BUG] Interval of Float do: infinite loop 
child of 0007002new  The Mother of all issues involving interval 
 IntervalIndexEnh.1.cs.gz [^] (749 bytes) 08-03-05 23:05
 IntervalIndexEnh.3.cs.gz [^] (785 bytes) 08-03-05 23:10
 Interval-indexOf-Test.1.cs [^] (454 bytes) 04-29-07 19:29
 IntervalIndexEnh.4.cs [^] (1,488 bytes) 04-30-07 17:08
 Interval-IndexOf-M1602-Patch-nice-BG.1.cs [^] (2,154 bytes) 02-03-08 20:25
 Interval-indexOf-M1602-Test-nice.1.cs [^] (1,160 bytes) 02-03-08 20:36

Notes
(0002100)
KenCausey   
08-03-05 23:09   
"Boris Gaertner" <Boris.Gaertner@gmx.net>:

"The code that I posted earlier today is wrong:
For
   (1 to: 10 by: 0.1) indexOf: 1000
the correct answer is 0, but my code answers 9991,
This is a stupid mistake, because the check that is
needed here, is available - I simply did not use
what is available. The method
  indexOf:startingAt:ifAbsent:
has to check:

    (self rangeIncludes: anElement)
      ifFalse: [^0].

The attached change set contaisn a method that performs
that check.

In a separate mail, I will post two additional tests.

My apologies for my mistake"

(attaching IntervalIndexEnh.3.cs.gz)
(0002101)
KenCausey   
08-03-05 23:11   
I was able to load both changesets without errors, but I did not test them further.
(0010516)
edgardec   
04-10-07 12:28   
In a 3.10alpha.7083

Before the fileIn
MessageTally time: [(1 to: 100000 by: 0.01) indexOf: 99999] 53772
After 0

I collecting in a 3.10 update
(0010540)
edgardec   
04-13-07 21:03   
This now is 7084CollectionEnh.cs and was in updates for 3.10
Thanks Boris !
(0010628)
nicolas cellier   
04-29-07 19:18   
As explained by german at 0001603 , this patch is wrong

(100000000000000 to: 500000000000000 by: 100000000000000)
   indexOf: 250000000000000
should answer 0

See uploaded test provided.

German said code should be (val fractionPart abs * 100000000 < 1)

This trick is there to answer true for Float that fall near one of the elements.
As explained by me at 0006455 ,
1) This is QUESTIONNABLE
2) This is still wrong

Code should rather test

(val - val rounded) abs * 100000000 < 1.

This is programmed in patch number 4.

And there, why 8 decimals? or 10 or whatever?

Interval-indexOf-M1602-Test-nice.1.cs contains tests for 0001602 and 0006455 .

Last thing, Boris patch must not be dissociated from 0001603. So Interval>>#includes: is also removed by Interval-IndexOf-M1602-Patch-nice-BG.1.cs.
Otherwise, you will have different results with includes: and indexOf:

If arithmetic is exact (Integer, Fraction, ScaledDecimal...), then no fuzzy inclusion test takes place.

Note that super indexOf: would behave better if 0006456 is applied, but with strict equality, as done by self asArray indexOf:.

(0011733)
nicolas cellier   
02-03-08 19:40   
"fix begin"
Installer mantis bug: 1602 fix:'Interval-IndexOf-M1602-Patch-nice-BG.1.cs'.
"fix test"
Installer mantis bug: 1602 fix:'Interval-indexOf-M1602-Test-nice.1.cs'.
"fix end"

(0012392)
nicolas cellier   
07-19-08 20:52   
One more thing about Interval of Float: we know they are nasty things due to inexact arithmetic and should better be avoided.

Here, i exhibit an example of Float Interval that does not includes its elements, whatever the efforts to make the includes: test fuzzy...

| eps i |
eps := (1.0 timesTwoPower: -52) * 1.25.
i := (1 to: 1+(99*eps) by: eps).
^(i includes: (i at: 2))

How many elements are not in the collection?

| eps i |
eps := (1.0 timesTwoPower: -52) * 1.25.
i := (1 to: 1+(99*eps) by: eps).
(i count: [:e | (i includes: e) not]) / i size asFloat.

You see, 75% ...
Fuzziness efforts are vain...
You won't find a generic threshold deserving to be written in the Kernel stone.
Should be up to application to decide about the threshold and the policy if they really insist on using Interval of Floats...
(0012603)
KenCausey   
09-09-08 14:58   
Would someone familiar with this issue please consider opening a new report based on it's status now (post-3.10.2) or point to existing issues that cover it so that this issue can be closed? It gets very confusing to have long histories on these issues and we really need to change to a system where when an issue is harvested the issue is immediately marked resolved and discussion stops (with the exception of adding relationships and pointing out those relationships). And when a release is issued the issue should be immediately closed.
(0012620)
nicolas cellier   
09-09-08 20:26   
IMO, better a long history than no history.
However, yes, what a tedious job to read verbose notes!
A new issue has now be opened at 0007180 :
- because patch uploaded in 3.10 does not solve the problem.
- so that this issue can be closed.
(0012627)
KenCausey   
09-10-08 17:40   
Thanks again Nicolas. Partially harvested in update 7084, released with 3.10. The saga continues at 0001603, please continue discussion there.