Comment for t.test.test_process.SignalMixin is egregiously wrong

[jml]

    # XXX: Trial now does this (see
    #      twisted.trial.runner.MethodInfoBase._setUpSigchldHandler)... perhaps
    #      this class should be removed?  Or trial shouldn't bother, and this
    #      class used where it matters?
    #        - spiv, 2005-04-01

MethodInfoBase no longer exists. Trial doesn't have any direct calls to signal.signal.

[exarkun]

SignalMixin should just be deleted.

[exarkun]

I deleted it in ctrl+backslash-2371+2123-3

[exarkun]

(In [19640]) Merge ctrl+backslash-2371+2123-5

Author: exarkun Reviewer: jknight, radix, therve Fixes #2371, #2123, #2378

Remove the long-since obsolete test helper SignalMixin, which set up and tore down special signal handling state. For some time, trial has taken care of this detail. (#2378)

Change the process file descriptor "connection lost" code to reverse the setNonBlocking operation done during initialization, leaving the file descriptor in the blocking state (what it probably was before it was given to the process code). This works around a bug with PTYs on OS X where the last write made to a non-blocking PTY might be lost if the process exits before the data is read from the other end of the PTY. This also corrects a (perhaps unreasonable) assumption that blocking writes may be made to stdout after twisted.internet.stdio has been used. (#2371)

Change ConsoleManhole so that it waits for connectionLost notification before stopping the reactor. This avoids the possibility that the reactor will stop before its transport has had all of its output flushed. Combined with the fix for #2371, this fixes #2123.