Stochastic/data-dependent control flow broken

[yebai]

See here. This will be fixed in new versions of Libtask - hopefully soon. We should warn the user when making a new Turing release - examples depend on this feature will return incorrect results but won't throw errors.

Also, unbounded loops are not supported but hopefully will be fixed in Libtask soon.

[yebai]

cc @devmotion @KDr2 @FredericWantiez @cpfiffer

[cpfiffer]

Wait, what exactly is the behavior?

[KDr2]

We can throw an error if there's control flow in the task function, maybe we'd better add this to Libtkask first?

[KDr2]

We can throw an error if there's control flow in the task function, maybe we'd better add this to Libtkask first?

Oh, I am sorry, it is not easy to predict the control flow is a dynamic one or a static one.

[yebai]

Fixed by TuringLang/Libtask.jl#116