Re: program error (exception access violation) in windows

[Martin Krischik <krischik@users.sourceforge.net>]

ldb schrieb:
>> raised PROGRAM_ERROR : EXCEPTION_ACCESS_VIOLATION
> 
> We have tried, in numerous ways, to come up with an ada library that
> can be used by c programs. The ada program needs to return a pretty
> large and ugly dynamic structure. For the sake of discussion, think of
> it as an array of structs (with pointers to dynamic sub-arrays).
> 
> I've gotten it to work flawlessly under linux using shared libraries. I
> use a combination of importing malloc() and the
> Ada.Address_To_Access_Conversions package. In this way, I can malloc
> the correctly sized eventual c data-structure, and use the created
> access from that package to populate the data structure. When I return
> the address of that to C, everything appears to work fine. I can free
> it like any other C data structure.
> 
> It seems to fail (sometimes) when I move it to windows (using a c
> program and my ada-based dll), resulting in the above exception. When I
> remove the free calls, in c, everything works fine (and memory leaks).
> I have a 'minimal' program that uses a simple array of integers, and it
> seems to work fine, but for our main program, it still crashes.
> Tommorow, I'm going to rip apart our program to see if I can find a
> minimal example of the crashing behavior.
> 
> We've tried numerous different ways of allocating the memory in ada,
> and free'ing it in C. We even got a smiliar error when we tried
> allocating it in Ada, and free'ing it in Ada through another C call to
> the ada library. We are about to change our interface from Ada to
> single elements, only, and have C wrapper functions that do all the
> allocating and query for each single value to populate the data
> structure.
> 
> Has anyone heard of issues like this? Or have a possible guess as to
> what is going on?

I once worked with C. Now I used gnat wich internaly uses malloc and 
free anyway. Still I went thrue the trouble of localising all the memory 
management on one side. Also I did not just provide a malloc and free 
c-likes but OO "contructors" and "destructors" constructs.

Ok, you end up with lots of Xxxx_New and Xxxx_Delete functions but all 
worked fine in the end.

Martin