sh.py and ipython for python 3.3

[amoffat]

Someone using sh.py with ipython ran into the following issue, and I was hoping someone with better working knowledge of ipython's internals could shed some light on what's happening: amoffat/sh/issues/97

sh.py has some magic going on internally, where any attribute access on the module can resolve dynamically to a specific object. This issue only seems to be in ipython for python 3.3, python 2.7 works fine. Anyone have an idea of what could be happening?

Thanks

[takluyver]

It seems Python 3.2 changed hasattr() to only catch AttributeError, rather than any exception. Here's the mailing list thread and the bug report for it.

I think we should catch this in IPython, but I consider that 'protecting against misbehaving code'. To be a good Python citizen, I think sh should:

• Catch any special attribute lookups (__x__), and, if there isn't a sensible return value, just raise AttributeError without trying to look them up as program names. This would deal with the first case reported.
• If attribute access fails because a command can't be found, ideally the error raised should be a subclass of AttributeError.

That would help other tools that try to do introspection as well.

[amoffat]

Thanks for the feedback @takluyver, I'll take care of it in sh

[takluyver]

Reopening, because we'll fix this in IPython as well. Even if code is doing bizarre stuff, it shouldn't be able to crash IPython. And if there's one bit of code in the wild doing this, chances are there are others.