Re: Wrapping code
A Divendres 25 Novembre 2005 13:55, Goswin von Brederlow va escriure:
> Leopold Palomo-Avellaneda <lepalom@wol.es> writes:
> > Hi,
> >
> > I have a question that maybe could be an off topic but I think that it
> > could be interesting for the amd64 users.
> >
> > I have a library code with some executables of a device. This device uses
> > the parallel port and provides a library to use it with some more
> > functionalities (provides some .so and some .h to link with).
> >
> > Of course I don't have the source code and worst the library is compiled
> > for RedHat 7.2 :-( (gcc-2.96 ...)
> >
> > I have though that maybe it could be possible to develop a wrapper to use
> > the library in the way that I can update compiler and distro and also
> > arch amd64) but really, I don't know if is this possible and how to
> > begin.
> >
> > Is this possible or I'm asking something stupid?
>
> It is possible. Nothing is impossible. But it is also quite stupid.
>
> You would need a very god reason, like you get paid for it, to do this.
>
> > If it's possible, how I can do that? rewriting the headers and a cpp code
> > calling the library functions? and the compiler, have I to make some
> > special flag? can I reuse a .so as a .a?
> >
> > This idea could help to the people that have devices that have libraries
> > to develop with them but not the source code and for example want to
> > change of distro or upgrade an a amd64. Of course that if the library is
> > compiled in a 32 bits it will be 32 bit, but it's possible to mix 64bit
> > and 32bits code?
>
> You need a library that mmaps the old library into a 32bit address
> space, provides a wraper function for every function that uses 64bit
> calling convention, translates those into 32bit variables, copies all
> pointer targets into the 32bit address space and adjusts them, calls
> the origianl function with the gcc-2.96 C calling conventions (which
> differ miniscule from current gcc) and in 32bit mode, and reverses all
> that with the results. Especialy for the pointer conversions you need
> detailed knowledge of every function, so no automatic conversion
> possible there.
>
> On the other hand you might just be able to just use the library as is
> in 32bit mode, unless some glibc symbols clash or they use one of the
> very rare construct with abi changes. And thats why the wraper is stupid.
I forgot to mention that it's not c but that it's c++ with ABI from the redhat
7.2 (gcc-2.96) . :-( If it's possible to use swig to generate wrapper and
then use the generated wrapper on another machine.
Regards,
Leo
Reply to: