[Vm-dev] Re: [Cog] reproducable crash with SmallInteger as methods

[Mariano Martinez Peck]

Sorry, I forgot to say that fixing this is important because otherwise you
have random crashes/errors because when using SmallIntegers as compiled
methods, they ARE put in the cache....however, when trying to flush, it
crash. So, it is important if you want to support SmallInetger as methods.

Cheers

Mariano

On Fri, Dec 17, 2010 at 6:54 PM, Mariano Martinez Peck <
marianopeck at gmail.com> wrote:

> And this change in  #primitiveIndexOf:
> also fixes the #flushCache in SmallInteger.
>
> Because if you use SmallInteger as methods, you need to implement
> #flushCache because it is used in MethodDIctionary >> #at:put:
> But before this change, this crashes because of the send to
> #primitiveIndexOf:
> But since now #primitiveIndexOf: is working, #primitiveFlushCacheByMethod
> doesn't fail anymore with SmallInteger  :)
>
>
> Cheers
>
> Mariano
>
>
> On Fri, Dec 17, 2010 at 11:01 AM, Mariano Martinez Peck <
> marianopeck at gmail.com> wrote:
>
>> Hi Eliot. I "fix/patch" it by chaging primitiveIndexOf:   to this:
>>
>> primitiveIndexOf: methodPointer
>>     "Note: We now have 10 bits of primitive index, but they are in two
>> places
>>     for temporary backward compatibility.  The time to unpack is
>> negligible,
>>      since the derived primitive function pointer is stored in the method
>> cache."
>>     <api>
>>     | primBits |
>>
>>     (objectMemory isOopCompiledMethod: methodPointer)
>>         ifTrue:
>>             [primBits := ((self headerOf: methodPointer) >> 1) bitAnd:
>> 16r100001FF.
>>             ^(primBits bitAnd: 16r1FF) + (primBits >> 19)]
>>         ifFalse:
>>             [^ 0 ].
>>
>>
>> at least for my case, it patches the problem. Still, I have no idea:
>>
>> - the implicance/side effect of this change
>> - why in development was crashing but not in deployment.
>>
>> thanks
>>
>> mariano
>>
>>
>>
>> On Thu, Dec 16, 2010 at 5:27 PM, Mariano Martinez Peck <
>> marianopeck at gmail.com> wrote:
>>
>>> Hi Eliot. Today I can reproduce a do-it that crash Cog, when using
>>> SmallIntegers as compiled methods. The problem is actually that IT ONLY
>>> crashes when compiling in XCode with "Development". However, if I compile
>>> with "Deployment", it doesn't crash....
>>>
>>> The code is this:
>>>
>>> SmallInteger compile: 'flushCache "emtpy"'.
>>> SmallInteger compile: 'run: aSelector with: arguments in: aReceiver
>>>     Transcript show: ''something''.'.
>>> TestCase compile: 'foo  ^ ''foo'' '.
>>>
>>> TestCase methodDict at: #foo put: 5.
>>> TestCase new foo.
>>>
>>>
>>> I really don't understand how this can work in "Deployment" but not in
>>> "Development" since the code is the same as I know. Maybe when debugging
>>> certain functions/methods are called that crash?
>>>
>>> Thanks for any tip.
>>>
>>> Mariano
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20101217/7fc799cd/attachment.htm