Possible problem with FFI in 3.9-7067 ?

> Andrew Tween wrote:
>> I have been trying to use FFI (the latest on SqueakMap) in  
>> 3.9-7067, on Windows
>> 2000, to get a device context, and do stuff with it. It doesn't  
>> seem to work
>> like it used to.
>
> I'm seeing the same problems.
>
>> Some investigation reveals that everything worked ok up to 7052.
>
> How interesting. Do you know what changed between 7052 and later  
> versions?
>
>> Have I installed the wrong FFI packages for 3.9?
>
> No, this is the right place.
>
>> Or is there a problem in FFI?
>> Or maybe there is a problem in the Compiler?
>
> That's a good question. I'm not sure right now; I'll have to check  
> 7052 and later to see what the difference exactly is.
>
> Cheers,
>   - Andreas
>
>

>
> > Have I installed the wrong FFI packages for 3.9?
>
> No, this is the right place.
>
> > Or is there a problem in FFI?
> > Or maybe there is a problem in the Compiler?
>
> That's a good question. I'm not sure right now; I'll have to check 7052
> and later to see what the difference exactly is.

work after

> Hi,
> I have managed to "cure" the problem by doing the following...
>
> After installing FFI in 3.9-7067, file-in the attached method (ExternalType
> class-initializeStructureTypes.st).
> Then evaluate,
>
>     ExternalType initialize.
>     ExternalStructure compileAllFields.
>
> the "Win32Window win32Draw" example then performs correctly without crashing the
> VM.
> Cheers,
> Andy
>
> ----- Original Message -----
> From: "Andrew Tween" <[hidden email]>
> To: "The general-purpose Squeak developers list"
> <[hidden email]>
> Sent: Wednesday, March 14, 2007 8:38 AM
> Subject: Re: Possible problem with FFI in 3.9-7067 ?
>
>
>> Hi,
>> ----- Original Message -----
>> From: "Andreas Raab" <[hidden email]>
>> To: "The general-purpose Squeak developers list"
>> <[hidden email]>
>> Sent: Tuesday, March 13, 2007 4:47 AM
>> Subject: Re: Possible problem with FFI in 3.9-7067 ?
>>
>>
>>> Andrew Tween wrote:
>>>> I have been trying to use FFI (the latest on SqueakMap) in 3.9-7067, on
>> Windows
>>>> 2000, to get a device context, and do stuff with it. It doesn't seem to
> work
>>>> like it used to.
>>> I'm seeing the same problems.
>>>
>>>> Some investigation reveals that everything worked ok up to 7052.
>>> How interesting. Do you know what changed between 7052 and later versions?
>> The main difference was that FFI was removed from the image.
>> I have just tried removing FFI from a 7052 image, and then immediately
>> re-loading FFI from SqueakMap into the same 7052 image.
>> Having done so, "Win32Window win32Draw" crashes the VM.
>> The removal code I used is from ScriptLoader-md.296, the same as used when
>> transitioning from 7052 - 7053, and is shown below....
>> ---------
>>  | copies namesOfpackagesToUnload |
>>  namesOfpackagesToUnload := #('*FFI*').
>>  copies := MCWorkingCopy allManagers asSortedCollection:
>>   [ :a :b | a package name <= b package name ].
>>  (copies select: [:each | namesOfpackagesToUnload anySatisfy: [:ea | ea match:
>> each package name ]])
>>   do: [:z | z unload].
>>  Smalltalk recreateSpecialObjectsArray
>> -----------
>>
>> So perhaps something isn't being initialised when FFI is installed?
>>
>> It may be a red-herring but evaluating "Win32HDC compiledSpec" before and
> after
>> unloading/reloading, gives diffferent results...
>>
>> Before - Win32HDC compiledSpec a WordArray(100925444)
>> After - Win32HDC compiledSpec  a WordArray(65536)
>>
>>>> Have I installed the wrong FFI packages for 3.9?
>>> No, this is the right place.
>>>
>>>> Or is there a problem in FFI?
>>>> Or maybe there is a problem in the Compiler?
>>> That's a good question. I'm not sure right now; I'll have to check 7052
>>> and later to see what the difference exactly is.
>> Thanks.
>>
>> Cheers,
>> Andy
>>
>>
>>
>>
>> ------------------------------------------------------------------------
>>
>>

> Andrew Tween wrote:
> > I've just tried installing from SqueakMap, but it is delivering the same install
> > and packages as before.
> > I've cleared my SqueakMap cache, but with the same result.
> > When I browse SqueakMap in a web browser, I don't see any recent updates to FFI.
> > Could you check that it is indeed updated on SqueakMap?
>
> Right. I noticed that too. It seems like SM is too smart for its own
> good - you cannot update a file on the server, you have to rename it. Ho
> Hum.

> > Could you check that it is indeed updated on SqueakMap?
>
> Right. I noticed that too. It seems like SM is too smart for its own
> good - you cannot update a file on the server, you have to rename it. Ho
> Hum.
>
> After fixing that, I found that my fix still didn't fix it and your
> actually looked better and better the longer I stared at it. What
> convinced me was when I was running through the debugger to see what
> happens when you don't bootstrap - the structure bit musn't be set or
> else both the pointer and the non-pointer type look like structs to the
> FFI (very, very bad).