docopt should evaluate sys.argv when being called, not when being imported about docopt

[barberj]

I was just having this conversation with @halst .

I second the idea. If sys is called within the method and not in the method header I can test like a user would call it...

https://github.com/barberj/ForGit/blob/5aedc5afeb3112796cb55f3e5f571d1c4e580938/test_forgit.py (Specifically look at TestCLI)

Above is an example of the tests I'm able to achieve using my forked docopt with argv defaulted to none.

[keleshev]

Why not pass sys.argv[1:] explicitly in case something modifies sys.argv. The benefits of this change are not clear to me.

[barberj]

I'm indifferent now (Line 180 https://github.com/barberj/ForGit/blob/master/forgit.py). I have been doing just as you suggested. It was just frustrating initially when i was trying to write test cases.

[schmir]

The current behaviour is surprising and unexpected. To quote https://en.wikipedia.org/wiki/Principle_of_least_astonishment:

"People are part of the system. The design should match the user's experience, expectations, and mental models."

If you know about the current behaviour, it may be easy to workaround. But that's not a good reason to keep it and I can't think of other reasons to keep the current behaviour...

btw. when can we expect the next release?

[keleshev]

Next release is planned this year.

Let me be devil's advocate for a moment:

What is so astonishing/unexpected when docopt parses the exact same options that were passed from command-line, not those obtained by some side-effect?

[schmir]

docopt doesn't parse the options that were passed from command-line. it parses the value of sys.argv at import time.
given the fact that some people will mock with sys.argv, docopt's behaviour is now also dependent on the import order.

and if you're one of those, you'll have to look around a bit in order to explain docopts behaviour. it isn't obvious that it's parsing something different from the value of sys.argv at call time.

python's behaviour of evaluating default arguments at import time is also a well known gotcha and unexpected to some people (see http://stackoverflow.com/questions/530530/python-2-x-gotchas-and-landmines). you're relying on that exact feature in order to implement more unexpected behaviour.

speaking of side-effects I think I can always just reload(docopt) in case I need to call into a program's main function :)

[keleshev]

I'm sold. This will break some people's code, but I'm implementing that for purity sake. Import order should not influence docopt behavior.

[maximerischard]

Reading sys.argv at runtime also makes it possible to use docopt interactively with ipython's run command.

[keleshev]

As of 0.6.0, argv is None by default, and docopt evaluates sys.argv[1:] when called. Thanks everyone for the fruitful discussion.