arg_readers Process tests broken on windows

[vtjnash]

When run properly (for example, JuliaLang/julia#39544) the tests on Windows fail, as they neglect to properly drain the process output pipe, as is mandatory for correct and successful operation on all operating systems (though this condition is currently only tested for and enforced on Windows, unless we merge JuliaLang/julia#39574).

An example test failure may either look like the following, or simply will deadlock the process until CI time expires:

Error in testset Tar:
Error During Test at C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\stdlib\v1.7\Tar\test\runtests.jl:492
  Got exception outside of a @test
  failed process: Process(`cat 'C:\Windows\TEMP\jl_9FC0.tmp'`, ProcessExited(1)) [1]
  
  Stacktrace:
    [1] pipeline_error
      @ .\process.jl:526 [inlined]
    [2] open(::Main.Test30Main_Tar.var"#118#136", ::Cmd; kwargs::Base.Iterators.Pairs{Union{}, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
      @ Base .\process.jl:401
    [3] open
      @ .\process.jl:392 [inlined]
    [4] #10
      @ C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\ArgTools\src\ArgTools.jl:187 [inlined]
    [5] (::Main.Test30Main_Tar.var"#114#132"{String})(tar::ArgTools.var"#10#20"{String})
      @ Main.Test30Main_Tar C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\stdlib\v1.7\Tar\test\runtests.jl:512
    [6] arg_readers(body::Main.Test30Main_Tar.var"#114#132"{String}, path::String, type::Type)
      @ ArgTools C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\ArgTools\src\ArgTools.jl:230
    [7] arg_readers(body::Function, path::String)
      @ ArgTools C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\ArgTools\src\ArgTools.jl:228
    [8] macro expansion
      @ C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\stdlib\v1.7\Tar\test\runtests.jl:493 [inlined]
    [9] macro expansion
      @ C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Test\src\Test.jl:1152 [inlined]
   [10] macro expansion
      @ C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\stdlib\v1.7\Tar\test\runtests.jl:493 [inlined]
   [11] macro expansion
      @ C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Test\src\Test.jl:1152 [inlined]
   [12] top-level scope
      @ C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\stdlib\v1.7\Tar\test\runtests.jl:485
   [13] include
      @ .\Base.jl:386 [inlined]
   [14] macro expansion
      @ C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\test\testdefs.jl:24 [inlined]
   [15] macro expansion
      @ C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Test\src\Test.jl:1152 [inlined]
   [16] macro expansion
      @ C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\test\testdefs.jl:23 [inlined]
   [17] macro expansion
      @ .\timing.jl:356 [inlined]
   [18] runtests(name::String, path::String, isolate::Bool; seed::UInt128)
      @ Main C:\buildbot\worker-tabularasa\tester_win64\build\share\julia\test\testdefs.jl:21
   [19] (::Distributed.var"#106#108"{Distributed.CallMsg{:call_fetch}})()
      @ Distributed C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Distributed\src\process_messages.jl:278
   [20] run_work_thunk(thunk::Distributed.var"#106#108"{Distributed.CallMsg{:call_fetch}}, print_error::Bool)
      @ Distributed C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Distributed\src\process_messages.jl:63
   [21] macro expansion
      @ C:\buildbot\worker\package_win64\build\usr\share\julia\stdlib\v1.7\Distributed\src\process_messages.jl:278 [inlined]
   [22] (::Distributed.var"#105#107"{Distributed.CallMsg{:call_fetch}, Distributed.MsgHeader, Sockets.TCPSocket})()
      @ Distributed .\task.jl:406

[vtjnash]

Bump. Should this be considered an issue with Tar.jl mis-using the package, or with ArgTools mis-handling the difference between streams and files? This is causing CI reliability problems, so we may need to hear from the designers here on what they intended.

[StefanKarpinski]

Could probably add code to drain the process. Would that fix the problem?

[StefanKarpinski]

Of course, if the operation in question needs to drain the input in order not to error, then it should do that. Which suggests changing Tar instead.

[vtjnash]

Yes, those are likely the two main possible courses of action. I think draining the output is the most robust solution for Cmd, in isolation, but then may raise questions about how to handle Process or Pipe, which suggests possibly Tar should be changed instead. (and doesn't rule out just making both changes either)

[vtjnash]

bump

[StefanKarpinski]

I think this PR JuliaIO/Tar.jl#114 might fix this issue but it's hard to tell since I can't reproduce the problem in Tar.jl even using Julia nightly.

[vtjnash]

It should be tested on JuliaLang/julia#39544, not master

[StefanKarpinski]

Will that fail on all platforms or do I need to test on a specific one?

[StefanKarpinski]

Should be fixed on the latest Tar.jl