On 28 February 2010 19:56, Levente Uzonyi <leves at elte.hu> wrote:
> On Sun, 28 Feb 2010, Jerome Peace wrote:
>
>>
>> Hi Levente,
>>
>> Thank you for your suggestions.
>>
>> I am not aiming myself at revising my tests. If you wish to add to them
>> great. I suggest adding specific suggestions and code to the mantis report.
>>
>> For some dictionaries "nil" has become a valid key. The tests in the
>> report were designed for 3.9 - 3.10 where Dan Ingalls innocent suggestion
>> that #= replace #==  caused a coder to break #comeFullyUpOnReload. I had to
>> design a test that proved dictionaries we not getting hashed correctly just
>> to prove somebody had caused a problem.
>
> nil was always a valid key for Dictionary and IdentityDictionary. Sets were
> the collections which couldn't contain nil.
>
except WeakKeyDictionary, for obvious reason.

>
> Levente
>
>>
>> The difficulty with dictionary problems is they appear like a
>> will-of-the-wisp causing very hard to track down mischief. Trust me on this.
>>
>> The problem may be recurring so its important to have tests IN THE IMAGE.
>> That can be run to find them. For Identity dictionaries these test have
>> proven to work if used. I haven't experimented with the new dictionary
>> flavors or even with other dictionaries, just the one class.
>>
>> The thing I am asking for help with is getting the test from change sets
>> on a mantis report to live code in an image. I don't have any skill with MC.
>> I am a Luddite with regards to it and have not taken the first steps to
>> mastery. This suits me as a debugger. I can find, analyse, and patch bugs
>> with the old methods. It leaves a gap in getting code committed. So I either
>> need a book on MC for Luddites or someone to help by committing the code to
>> resolve the issue. That's what I asked for.
>>
>> Once the tests are where they can be used, improvements will be welcome.
>> The existing tests are good enough to start with. Without something the
>> image is likely to have bugs way to hard to chase.
>>
>> More help in devising and PROVING tests that will find other problems is
>> welcome. It was not the aim of my post however.
>>
>> Cheers --Jer
>>
>>> Levente Uzonyi leves at elte.hu
>>> Sun Feb 28 13:19:40 UTC 2010 replied to:
>>
>>> On Sat, 27 Feb 2010, Jerome Peace wrote:
>>>
>>>> Hi all,
>>>>
>>>> I could use some help getting some crucial tests into trunk.
>>>>
>>>> The patches are in the mantis report of the title.
>>>> http://bugs.squeak.org/view.php?id=6721
>>>>
>>>
>>> nil is a valid key in Dictionary and IdentityKeyDictionary. A better test
>>> would be something like this:
>>>
>>> { Dictionary. IdentityDictionary } do: [ :dictionaryClass |
>>>        dictionaryClass allInstancesDo: [ :dictionary |
>>>                dictionary keysAndValuesDo: [ :key :value |
>>>                        self assert: (dictionary at: key) == value ] ] ].
>>>
>>> Or, if you want to use lower level (private) methods:
>>>
>>> { Dictionary. IdentityDictionary } do: [ :dictionaryClass |
>>>        dictionaryClass allInstancesDo: [ :dictionary |
>>>                dictionary array doWithIndex: [ :association :index |
>>>                        association ifNotNil: [
>>>                                self assert: (dictionary scanFor:
>>> association key) = index ] ] ] ]
>>>
>>>> The ill hashed dictionaries have been detected in Fun Squeak.
>>>>
>>>
>>> The tests above can be used to track down dictionaries which have an
>>> invalid state.
>>>
>>>
>>> Levente
>>>
>>>> I am worried that there is something in the main image that causes
>>>> conditions to occur so that dictionaries are not rehashed when needed.
>>>>
>>>> The problem is that a mis-hashed dictionary will show up only as other
>>>> symptoms. Code that seemingly doesn't work as expected. Without a ready test
>>>> to run tracking it back to the true problem is difficult.
>>>>
>>>> I have no MC skills so I am asking for someone to take over harvesting
>>>> from the report.
>>>>
>>>> Thank you in advance.
>>>>
>>>> Yours in curiosity and service, --Jerome Peace
>>>
>>
>>
>>
>>
>>
>
>



-- 
Best regards,
Igor Stasenko AKA sig.