Libtask waits for ever is function body has errors.

[trappmartin]

See the following minimal example:

using Libtask

function f()
    t = 0;
    while true
        t[2] = 0
        produce(t)
    end
end

t = Task(f)
consume(t)

[yebai]

@xukai92 @trappmartin I think we need to catch exception/error in consume / produce functions, and re-throw them when appropriate.

[trappmartin]

I tried to do that but didn’t manage so far. See commit below.

[trappmartin]

@yebai could you have a look at this? I tried to resolve this issue on the branch mt/broken_function but with almost no success so far.

[yebai]

@trappmartin @xukai92

There seems to be a bug in function Base.schedule_and_wait, which does not throw Exception as expected. Here is an MWE

function f()
    t = 0;
    while true
        t[2] = 0
    end
end

Base.schedule(Task(f))
Task (failed) @0x0000000113da86d0
MethodError: no method matching setindex!(::Int64, ::Int64, ::Int64)
f() at ./REPL[4]:4

Base.schedule_and_wait(Task(f)) # never returns. 

Ps. I have filed an issue in the JuliaLang repo: JuliaLang/julia#29869

[mohamed82008]

schedule(t) passes t to the scheduler for execution and moves on. It doesn't throw anything:

julia> f() = error("Ignored error")
f (generic function with 1 method)

julia> g() = (schedule(Task(f)); 1)
g (generic function with 1 method)

julia> g()
1

wait will dismiss the current task until rescheduled or yieldto is called. When you do this in the main task, and no other task is going to schedule it again, your program will freeze. schedule_and_wait just calls schedule then wait so the freezing effect of wait is still there, when called in the main task.

[yebai]

Related:

• https://github.com/TuringLang/Turing.jl/issues/594
• Program bug is silent for coroutines Turing.jl#586