Re: bug 828

[Bill Allombert <Bill.Allombert@math.u-bordeaux1.fr>]

On Mon, Jul 06, 2009 at 03:40:36PM +0100, Jason Moxham wrote:
>> On Mon, Jul 06, 2009 at 02:22:01AM +0100, Jason Moxham wrote:
>>> Hi
>>>
>>> I've been looking at bug 828  dynamic loading of a function in a static
>>> build of pari.
>>> Here is a guess based on my debug below , note the line numbers will not
>>> match as I added a bit debugging stuff
>>> FILE ../src/language/init.c LINE 836
>>>      2 [main] gp-sta 5304 _cygtls::handle_exceptions: Error while dumping
>>> state
>>> (probably corrupted stack)
>>> Segmentation fault (core dumped)
>>>
>>> It look like that when you load up the function addii from the dynamic
>>> library it also loads up another copy of the varibles avma and bot ,
>>> which are then accessed and of course the values they have are garbage.
>>
>> Interesting. What are exactly the linker flags used ?
>> There are probably some linker flags that could fix that.
>>
>> Could you try to rebuild everything with -fPIC ? This cahnges the way
>> global variables are linked, so it might make a difference here.
>>
>> Cheers,
>> Bill.
>>
>> -------------------------------------------------------
>
> If I'm reading this correctly there is no solution. Windows(and 
> apparently some form of BSD) does not support functions in a dll reading 
> global varibles in the original executible. Therefore our static build 
> will never work in windows and we should disable that part of the test 
> for windows. It works in the shared case as a dll can read a varible in 
> another dll.
>
> follow the the long threads below
>
> http://sourceware.org/ml/cygwin/2004-04/msg00184.html
>
> http://sourceware.org/ml/cygwin/2004-04/msg00351.html

Thanks for your link and research. However this does not explain why this works
with gp-dyn. I still wonder what happen when gp-sta is built with -fPIC.

Cheers,
Bill.