check pipe initialization using a status code

[bicycle1885]

Checking initialization of a pipe with its handler can cause false-positive assertion error when STDIN is piped (ref: #8529).
This seems to be caused by immediate release of a handler of the pipe after reading all data from it (ref: #8529 (comment)).
The .status field of a Pipe can distinguish "once initialized" pipes from "never initialized" ones.
This issue may be related to #6957 as well.

[Keno]

I guess this will end up passing /dev/null to the child process? I guess that's not unreasonable if all the data has been read.

[bicycle1885]

No the same as passing /dev/null. But this will have the same effect if a pipe is closed.
I think it is problematic that behavior changes depending on the sizes of an input and a pipe buffer:

pipe.jl

line = readline(STDIN)
@show STDIN
print(line)
println("-- "^20)
run(pipe(STDIN, `head`))

large input:

~/v/julia (fix-pipe-check|…) $ cat base/*.jl | ./julia pipe.jl
STDIN = Pipe(active, 271306 bytes waiting)
module Dates
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
# get and set current directory

function pwd()
    b = Array(UInt8,1024)
    len = Csize_t[length(b),]
    uv_error(:getcwd, ccall(:uv_cwd, Cint, (Ptr{UInt8}, Ptr{Csize_t}), b, len))
    bytestring(b[1:len[1]-1])
end

small input:

~/v/julia (fix-pipe-check|…) $ cat base/build_h.jl | ./julia pipe.jl
STDIN = Pipe(closed, 492 bytes waiting)
# This file is automatically generated in base/Makefile
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
```.

But I think this is a separated problem.

[vtjnash]

this should be using the test isopen(stdios[1])

[JeffBezanson]

fixed by 322ff51.