Re: [TCLCORE] Coroutines in HEAD

[miguel s. <mig...@gm...>]

Alexandre Ferrieux wrote:
> Notice that here I'm not concentrating on the coroutine/continuation
> difference. One motivation is that I predict that the free-floating
> (as in "unordered") refcounted "stack" frames will be teleported from
> Miguel's mind to silicon in no time (considering the timescale of the
> 8.5 release ;-). So basically I assume that today's anchoring of
> coroutine-sequels to the toplevel won't last long enough to be called
> a limitation of the model !

Much flattered :)

But ... what I may yet do before hell freezes over is running the same kind of 
coroutines in a slightly different way: they now run in [uplevel #0], I think I 
know how to let them "remember" the complete call stack at creation so that 
[upvar] and [uplevel] keep functioning even if the coroutine's creator has 
already returned. That is, so that the complete environment at creation time is 
captured - a real closure.

What I do *not* know (yet?) how to do properly is to marry that approach with 
correct functioning of [info level], [info frame] and error stacks. Access to 
the resumer's environment via [uplevel] and [upvar] would still be barred, or 
else require some new mechanism if the need is felt.

Note that today's coroutine make it look (for [info frame] and error stacks) as 
if whoever resumed a coroutine actually called the coroutine's body from the 
beginning. What should it look like for a continuation (let us call them that)?

IOW, I am more worried about the proper semantics than the implementation, which 
is a SMOP (famous last words).