Hitting ^D in an 'interact()' causes endless printing of 'exit'

[pexpect]

Hitting ^D in an 'interact()' causes endless printing of 'exit'

[takluyver]

I think this comment points to the issue:

# This is for Linux, which requires the blocking form
# of waitpid to # get status of a defunct process.
# This is super-lame. The flag_eof would have been set
# in read_nonblocking(), so this should be safe.

In interact, it's not using read_nonblocking(), so flag_eof is not sent, and it doesn't discover that the process has exited. If I'm right, fix coming up.

[takluyver]

Oh, no, it's simpler than that. You're using fdpexpect, so the fdspawn object doesn't know about the child process, only the file handle you've given it. The pty continues to exist after the process has died, so it has no way of knowing to stop reading.

Is there a reason you (or rather shutit) are replicating the functionality of pexpect? The easiest way to fix this would be to use the main spawn() class, which creates a pty, starts a new process in that pty, and tracks both the pty and the process. i.e. reduce that code sample to:

import pexpect
c = pexpect.spawn(['/bin/bash'])
c.interact()

[aidanhs]

Good point. I think this behaviour is probably historical, I'll need see if I can remember why I wrote it like that...
It may well be that just using plain pexpect is the best way to do this.

However, it doesn't change the fact that it terminates correctly in 2.4. This contradicts "[...] it has no way of knowing to stop reading."
Specifically, running on 2.4:

Traceback (most recent call last):
  File "test.py", line 21, in <module>
    c.interact()
  File "/usr/lib/python2.7/dist-packages/pexpect.py", line 1492, in interact
    self.__interact_copy(escape_character, input_filter, output_filter)
  File "/usr/lib/python2.7/dist-packages/pexpect.py", line 1520, in __interact_copy
    data = self.__interact_read(self.child_fd)
  File "/usr/lib/python2.7/dist-packages/pexpect.py", line 1510, in __interact_read
    return os.read(fd, 1000)
OSError: [Errno 5] Input/output error

[aidanhs]

Turns out the reason I did this initially was to connect pexpect to paramiko (which happens to expose a fd for the ssh terminal). I'm going to do it the way you suggested (thanks!) for shutit.

I'll leave this issue open, but I wouldn't call it a huge priority.
I'm not sure if it's possible to reproduce the issue for the original use case inside paramiko, but I do think it'd be preferable if pexpect have the behaviour it had before 3.0.

[takluyver]

Well, I wouldn't quite call throwing an error 'terminating correctly', but I take your point. Pexpect now catches that error, but it should probably stop looping when it does.

[jquast]

Yes, this bug was introduced by c7a180a -- @takluyver, see my pull request -- how does this match to what you intended with this comment?

[takluyver]

#72 is now merged, closing this.