Yielding to a copied task crashes

[sharpobject]

julia> consume(Task(()->7))
7

julia> consume(deepcopy(Task(()->7)))
Segmentation fault

I don't know what the right behavior is here. If tasks are not meant to be copied, copying one could return the original or throw an exception.

There are significant use cases for copying tasks, though, so I hope that they are meant to be copied. In that case, the bug is that copied ones do not work.

[JeffBezanson]

No, tasks are not currently meant to be copied. I think it could be implemented, but the semantics are a bit fiddly. For example, a task waiting for the original task is probably not waiting for the copy. And values referenced from the task's stack should probably not be deep-copied.

Stefan and I kind of disapprove of deep copying in general, but we were not quite able to talk people out of adding it.

[sharpobject]

These issues are troublesome. In the contexts in which I'm used to being able to copy coroutines, the coroutines are rigidly asymmetric, so the solutions that are enabled by that assumption may not be applicable. I'll look into it a bit more if I have time.

> t = coroutine.create(function()coroutine.clone(t) end)
> print(coroutine.resume(t))
false   attempt to clone uncloneable coroutine
> -- The coroutine cannot be copied while it is running
>
> t = coroutine.create(function()print(coroutine.resume(coroutine.create(function()coroutine.clone(t) end))) end)
> print(coroutine.resume(t))
false   attempt to clone uncloneable coroutine
true
> -- It also cannot be copied while it is waiting on another
>
> t = coroutine.create(function() local t = {0} while true do coroutine.yield(t[1]) t[1] = t[1]+1 end end)
> print(coroutine.resume(t))
true    0
> u = coroutine.clone(t)
> print(coroutine.resume(t))
true    1
> print(coroutine.resume(u))
true    2
> -- It receives a shallow copy of the stack
>
> t = coroutine.create(function() local t = 0 while true do coroutine.yield(t) t = t+1 end end)
> print(coroutine.resume(t))
true    0
> u = coroutine.clone(t)
> print(coroutine.resume(t))
true    1
> print(coroutine.resume(u))
true    1
> -- But certainly not the same exact stack

[pao]

Looks like this should be linked to #4085.