Revert "add stream shutdown and support half-duplex operation"

[DilumAluthge]

Reverts #40783

This broke the tests on FreeBSD 12.

[vtjnash]

IIRC, someone said on Slack the failure is a bug in the FreeBSD libunwind library, not this PR?

[DilumAluthge]

If I understand correctly, those are two separate issues. The discussion on Slack was about the segfaults during the FreeBSD tests.

But separately from the segfaults, there are test failures in the Sockets tests on FreeBSD that are caused by this PR.

@ararslan Is that correct?

[ararslan]

As far as I know, that is correct. In the Sockets tests, the pipe is not being closed when the tests are expecting it to, so !isopen(P) is false. The segfaults, which show up as that unhandled task error, seem to be due to some issues with DWARF emission and unwinding and are unrelated to this PR.

An alternative to reverting this is to do something like

diff --git a/stdlib/Sockets/test/runtests.jl b/stdlib/Sockets/test/runtests.jl
index 328ea929f4..0186677726 100644
--- a/stdlib/Sockets/test/runtests.jl
+++ b/stdlib/Sockets/test/runtests.jl
@@ -521,7 +521,12 @@ end
     @test readuntil(P, 'w') == "llo"
     Sys.iswindows() && wait(t)
     @test eof(P)
-    @test !isopen(P) # eof test should have closed this by now
+    # FIXME: eof test should have closed this by now, but it seems not to on FreeBSD
+    if Sys.isfreebsd()
+        @test_broken !isopen(P)
+    else
+        @test !isopen(P)
+    end
     close(P) # should be a no-op, just make sure
     @test !isopen(P)
     @test eof(P)

and open an issue for tracking that. I've not yet looked into it, as I've prioritized investigating the segfaults.

[vtjnash]

that seems okay too

[DilumAluthge]

Between the revert (this PR) and a @test_broken, which do you prefer?

[DilumAluthge]

Closing in favor of #41983