🚀 go-pugleaf

In article <715992030snx@cotswold.demon.co.uk> Nigel Roles, nigel@cotswold.demon.co.uk writes:
>BC++ 3.1 reference manual on longjmp states it is sufficient for implementing
>coroutines.

Well, I tried that, but it didn't quite work -- can someone tell me
what dumb thing I did wrong?

To create a thread, I malloc space for the new stack.  Then I build
a return frame on the new stack by switching the sp and ss to the address
of the new stack, and doing a setjmp (I owe the structure of this code
to the Awesime package by Dirk Grunwald).  Later on, I can switch to
running the new process with a longjmp.  Here is the code:

static PROCESS process;		/* set by calling function */
static jmp_buf originalContext; /* static so not ref'd off the stack */

/*
 * buildContext() creates a return context on the process' new stack.
 * The return frame is stored in the jmp_buf process->context.
 * To switch to the new proces, a longjmp(process->context) is done.
 */
void buildContext()
{
	if (setjmp(originalContext) == 0) {	/* save where we came from */
		_SS = FP_SEG(process->stackBase);	/* switch to new  */
		_SP = FG_OFF(process->stackBase);	/* stack and then */
		if (setjmp(process->context) == 0) {	/* save context */
			longjmp(originalContext);	/* jump back */
		} else {
			/* switched to new process for 1st time */
			/* so call the process' entry point */
			process = activeProcess();
			(process->entryPoint)(process->arg);
			killProcess(process);
			/* not reached */
		}
	}	
	/* return frame built successfully */
}

After switching to the new stack and doing a setjmp there, I can't
seem to get back to the original context with the longjmp. 
If anyone can see what's wrong, please let me know!

stan