possibly fix a libuv assertion failure on shutdown

[JeffBezanson]

On Keno's branch kf/stderrpipe (#11919) I reliably see this assertion failure:

jeff@gurren:~/src/julia/test$ ../julia runtests.jl file examples
    From worker 2:       * file                 in   9.90 seconds
julia: src/unix/stream.c:1166: int uv_shutdown(uv_shutdown_t *, uv_stream_t *, uv_shutdown_cb): Assertion `((stream)->io_watcher.fd) >= 0' failed.

signal (6): Aborted
gsignal at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
abort at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x7f5a200f9b86)
unknown function (ip: 0x7f5a200f9c32)
julia: src/unix/stream.c:1166: int uv_shutdown(uv_shutdown_t *, uv_stream_t *, uv_shutdown_cb): Assertion `((stream)->io_watcher.fd) >= 0' failed.

signal (6): Aborted
unknown function (ip: 0x7f5a20862deb)
jl_close_uv at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
jl_atexit_hook at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
jl_exit at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
exit at client.jl:39
gsignal at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
abort at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x7f86420d0b86)
unknown function (ip: 0x7f86420d0c32)
unknown function (ip: 0x7f8642839deb)
jl_close_uv at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
jl_atexit_hook at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
jl_exit at /home/jeff/src/julia/usr/bin/../lib/libjulia.so (unknown line)
exit at client.jl:39
    From worker 3:       * examples             in  19.12 seconds
    SUCCESS

This change makes it go away. cc @vtjnash

[vtjnash]

this change doesn't look valid

[JeffBezanson]

Agreed; I strongly suspect this is not the right thing. All I know is it fixed this particular assertion failure.

[vtjnash]

is the handle object value when you reach that point?

the symptoms of this looks like #4229 and #7275

[JeffBezanson]

By that do you mean handle->data is non-NULL?

[vtjnash]

s/value/valid

does the rest of it (including but not limited to ->data look like a well-formed object?

[Keno]

The thing that's failing here is actually the simple cluster manager test. The reason the output is duplicated is because both of them are failing.

[Keno]

This is fixed on that branch and was caused by initializing pipes that server as unix domain sockets as readable&writable. This isn't the right way to use that libuv API however, which expects both of those to be unset and will set them itself once the accept/connect has gone though.