SR-7608: Pipe() crashes when the underlying system call fails.

[spevans]

• Pipe does not have a failable or throwable initializer but currently
  calls fatalError() for all errors returned from pipe().

• If pipe() returns EMFILE or ENFILE then set the fileDescriptor
  to -1 for both .fileHandleForReading and .fileHandleForWriting.
  For all other errors call fatalError().

• TestFileHandle.swift: Remove duplicate test_pipe().

• TestPipe.swift: Add extra test to try and exhaust the process's pipes.

• Test the file descriptor is valid and open for certain methods
  and properties to match Darwin.

[spevans]

@swift-ci test

[spevans]

@swift-ci test

[spevans]

@swift-ci test

[spevans]

@swift-ci test

[parkera]

Do we need to provide some new API on Darwin that avoids throwing an ObjC exception?

[spevans]

On Darwin Pipe() doesn't actually throw an exception but erroneously sets the pipe ends to be 0.

eg:

var pipes: [Pipe] = []
while true {
    let pipe = Pipe()
    print (pipe.fileHandleForReading.fileDescriptor, pipe.fileHandleForWriting.fileDescriptor)
    if pipe.fileHandleForReading.fileDescriptor == 0 {
	print("Created \(pipes.count) pipes")
	break
    }
    pipes.append(pipe)
}

Will create about 1266 pipes on 10.13.6 before returning 0 for the file descriptors. Maybe this initialiser should be deprecated and a failable one introduced?

[parkera]

Interesting. I'll track that with a radar for us. rdar://problem/44605663

[millenomi]

I'm worried about this, because it means that the test process cannot use pipes meaningfully anywhere else.

Can we move this code to a separate process we spawn?

[spevans]

All of the pipes should be closed when the array is cleared. test_MaxPipes is run before test_Pipe to test this very case.

[spevans]

@swift-ci test

[millenomi]

@swift-ci please test and merge