Re: [AD] proposed patch for nicer asm symbols exports

[Peter Wang <tjaden@xxxxxxxxxx>]

Sam Hocevar wrote:

>   AFAIU (I was not here at the time Allegro was designed), the
> alleg_unsharable serves at least two purposes:
>
>   - provide the main() function on Unix, because the main() from the
>     original program has been diverted to __mangled_main_address.
>  

Well, this part is not right.  main() is in liballeg.so as far as I can see.

>   - provide assembly functions that are not position independent code
>     (PIC). When building a Unix shared object (library), one needs to
>     use the -fPIC flag to create position independent code, or the
>     library will not be properly loaded in a shared address space. PIC
>     code stores a special pointer in a reserved register (EBX on x86),
>     which means that the compiled code cannot use EBX and will run at a
>     slight performance loss. Since the asm code in alleg_unsharable does
>     not reserve EBX, it is built as a static library instead of a
>     shared one.
>  

Thank you for the explanation.

>   However, on x86 library code need not be compiled with -fPIC because
> the architecture allows linking non-PIC code with PIC code in the same
> object. And we do not provide assembly routines on platforms that do not
> allow this linking (such as Alpha, IA64, HPPA).
>  

Ok, great.  I suppose the linking of non-PIC and PIC code only works if 
your patch is applied?  Sorry if it's a dumb question.

>   I would like to get rid of alleg_unsharable because its presence
> means that you can only upgrade liballegro in a way that works with all
> previous versions of alleg_unsharable your binaries were compiled with,
> or you have to recompile all your applications, which kinda defeats the
> purpose of shared libraries.
>  

If you send me a patch for misc/deplib.sh and other files for getting 
rid of liballeg_unsharable completely, I'll apply it unless there are 
objections.  Assembly routines won't be provided for non-x86 machines 
any time soon.

Peter