[redland-dev] Win32 DLL's / shared heap limitations effect on Redland

[Dave Beckett]

On Thu, 2005-03-03 at 07:25 -0800, Brian Holdsworth wrote:
> I found a past post from last August that eluded to a
> problem with running Redland on Windows.  Like that
> poster, I am using all Redland libraries, which I have
> built as 3 separate DLL files.  It mostly works fine,
> until I start to create and destroy multiple models
> using multiple instances of the parser objects. 
> Without getting too specific, it seems that the three
> DLL's might not be playing nice together.  I may be
> seeing what the earlier poster claimed, which is that
> the librdf.dll is trying to free heap memory that was
> allocated by the raptor.dll.  This causes exceptions
> under Windows, as the required practice is to have the
> same DLL both allocate and free blocks.  This is due
> to the fact that each DLL has its own heap manager.

Yeah, I heard about that.

> I would like to know if it is realistic to assume that
> the above is occurring - i.e. are the librdf and
> raptor libraries intertwined in this way?

They are intertwined but not intended to have this problem - i.e. this
is a bug if you find where it is happening.

> If so, I think I will pursue creating one giant
> librdf.dll for Windows to see if that solves the
> problem.  Can anyone think of a reason not to solve it
> this way?

Because it's horrible? :)

Somebody else solved it by making rasqal and redland (librdf.dll) use
raptor's memory alloc/free routines so all the memory is alloced inside
raptor.dll.  #define malloc raptor_alloc_memory and
#define free raptor_free_memory kinda thing.

Dave