Custom interactive prompts

[davidhalter]

improves and fixes #741. might fix #177 (not sure though) and probably also helps with #7 (also I don't know if that one is really fully solvable).

There's an implementation difference. The env variable is not named prompt_responses anymore but prompts, because I think it's shorter and a better fit.

For the future we could think about two things:

• Currently the dictionary allows only for strings, I think it would be interesting to allow compiled regex objects, should not be hard to implement.
• @owenn used a small hack to make a prompt react once (and not twice on the same input). This involves removing the old matching output. That means the output of run() doesn't contain the complete text. That's not extremely bad, but we could improve this by counting bytes (and just not trying to check for a new prompt if the number of bytes doesn't change). But I don't really know what side effects that would have, since it's also possible to just answer prompts manually (and use the backward button).

And also I would like to see tox in fabric. It makes testing (especially the setting up part) way easier (if you haven't heard of it: it eases testing against multiple python versions).

Now you can do something like:

with settings(prompts={"remove write-protected regular file `/etc/fuse.conf'? ": "yes"}):
    run('rm -i /etc/fuse.conf')

Not that this example makes sense, but it shows the use case well: interactive prompts.

When could we expect a release containing that code?

[akitada]

+1
This worked great.

[bitprophet]

Was leery about this originally, but the changes this makes to core (aka the bits in io.py) look pretty light, so I will try to play around with it briefly and get it in for the next feature release (~1 week out). Thanks! also @akitada for the prod :)