Nicolas Cellier wrote:
> Take this example:

Ah, good point. (it works fine with o1 = 'g' and o2 = o3 = o4 = 'd'). 
The thing is that the test in WeakSet>>add: was misleading - it tests 
for nil where scanFor: searches only for flag.

Cheers,
   - Andreas

> ws := WeakSet new.
> array := WeakSet new instVarAt: (WeakSet allInstVarNames indexOf: 'array').
> o1 hash \\ array size + 1 -> 1.
> o2 hash \\ array size + 1 -> 2.
> o3 hash \\ array size + 1 -> 1.
> ws add: o1; add: o2; add: o3.
> 
> o1 := o2 := nil.
> Smalltalk garbageCollect.
> "o1 and o2 are reclaimed."
> (array at: 1) -> nil.
> (array at: 2) -> nil.
> 
> o4 hash \\ array size + 1 -> 1.
> ws add o4.
> (array at: 1) -> o4.
> 
> (ws includes: o3) -> false.
> 
> I will try and build a test case. Should be easy with Fractions.
> 
> 2009/5/6 Andreas Raab <andreas.raab at gmx.de>:
>> Nicolas Cellier wrote:
>>> Isn't that dangerous if you don't fixCollisionsFrom: index ?
>> Why would it? All this does is avoiding to update the tally ivar if we're
>> replacing a previously reclaimed entry with a new value. And I don't think
>> that #fixCollisionsFrom: even looks at tally.
>>
>> Cheers,
>>  - Andreas
>>
>>> Nicolas
>>>
>>> 2009/5/6 Andreas Raab <andreas.raab at gmx.de>:
>>>> Adrian Lienhard wrote:
>>>>> I haven't had time to look into this very closely yet but maybe somebody
>>>>> else already knows what is going on. My guess is that the NewSymbols
>>>>> weak
>>>>> set is repeatedly grown since tally is not decreased if the GC removes a
>>>>> weak element.
>>>> That's an excellent guess. Try changing WeakSet to e.g.,
>>>>
>>>> WeakSet>>atNewIndex: index put: aValue
>>>>   "Add newValue at the given index. Don't increment tally
>>>>    if current slot is nil since a previous value has been collected"
>>>>   (array at: index) ifNil:[^array at: index put: aValue].
>>>>   ^super atNewIndex: index put: aValue
>>>>
>>>> Cheers,
>>>>  - Andreas
>>>>
>>>>
>>>>
>>>
>>
>>
> 
>