task_done_hook has no InterruptException handler

[quinnj]

Case 1:

function try_no_segfault()
    while true
        try
            sleep(1)
            println("trying...")
            # @async 1+1
        catch e
            if typeof(e) <: InterruptException
                println("caught Interrupt")
                return
            end
        end
    end
    return
end

Case 2:

function try_segfault()
    while true
        try
            sleep(1)
            println("trying...")
            @async 1+1
        catch e
            if typeof(e) <: InterruptException
                println("caught Interrupt")
                return
            end
        end
    end
    return
end

In the successful case, run try_no_segfault(), then CTRL+c to see a successful interrupt and return to the julia prompt.

To reproduce the segfault, run try_segfault(), then CTRL+c to see a segfault and end of the julia session.

[yuyichao]

Which version are you using? And is it actually a segfault?

[quinnj]

julia> versioninfo()
Julia Version 0.5.0
Commit 3c9d753 (2016-09-19 18:14 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin13.4.0)
  CPU: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, haswell)

No, I'm not sure it's a segfault, but it does end my julia session instead of just exiting the function, and it seems to be purely because of the presence of the @async block.

[amitmurthy]

This may help in identifying the problem. Adding a yield() before the sleep() makes the problem go away.

        try
            yield()
            sleep(1)
            println("trying...")
            @async 1+1             
        catch e

Issue exists on 0.4 too.

[yuyichao]

The yield simply makes the window this can happen much smaller.

[amitmurthy]

There may be something more. I couldn't crash it with the yield() before the sleep() even after many tries. A yield() after the sleep() has no effect....

[yuyichao]

Well, it shortens the window from 1 second to few micro second so you'll have a pretty hard time triggering it.

[quinnj]

@yuyichao, any ideas on how to fix this?

[vtjnash]

This would be fixed by #14032

[StefanKarpinski]

Does changing the behavior of ctrl-C count as breaking production code?

[StefanKarpinski]

This doesn't seem like it needs to be in the 1.0 milestone. Speak up otherwise.