Re: [swarm-hackers] Compiling CVS Swarm in ia64 system

[Marcus G. Daniels]

Hi Robson,
> When the make runs emacs in order to modify some files (since I don't
> need GUI and I've disabled with the '--disable-gui' option in
> configure) the compilation gets frozen and a message appears:
> "Activity failed"
>   
Emacs isn't really `editing' files, it is reading a set of interface 
specifications and writing C stubs and Java declarations for those 
stubs.   It's being used like Perl or awk or similar, except that Emacs 
is convenient for traversing whole Objective C expressions in a 
random-access way (as opposed to being line oriented). 

The purpose of the stubs is to provide Java Native Interface wrapper to 
Swarm functions so that Java can call them.
To write the stubs themselves, an optimization is attempted so that 
message dispatch won't be needed.   It's this process that sometimes 
fails, but that failure is expected in cases such as when a -createEnd  
sets the available methods, or else switches the underlying object.

That being said, I think the main problem you will encounter with a port 
of the Java stuff to IA64 is some code called mframe.   That code reads 
a stack frame and some Objective C metadata in the runtime and 
determines what parameter values the caller (Swarm) had made.   It needs 
to do this when, for example Swarm's scheduler sends a message to an 
object in the `classic' Objective C style, but in fact the target object 
will be a Java object.   It needs to take apart the Objective C call and 
re-issue it as a Java call.   This is done through Objective C call 
forwarding, and it uses mframe to this.   mframe is port specific and I 
don't think its been tested on IA64.

With some work Scott Christley is doing, we expect that this mframe code 
will be replaced using standard APIs from MacOS X and GNUstep, but this 
isn't yet done.

Marcus