check_complete doesn't take automagic into account

[minrk]

With automagic on:

In [1]: ip.input_transformer_manager.check_complete('cd ~')
Out[1]: ('invalid', None)

In [2]: ip.input_transformer_manager.check_complete('%cd ~')
Out[2]: ('complete', None)

Which are equivalent after prefilters, transformers have been applied.

[minrk]

I noticed this since the qtconsole now relies on this for implicit execution.

[takluyver]

I think that is actually by design, though perhaps that design doesn't make sense any more. Input transformers are applied line-by-line, but prefilters are only applied when the code is run. So 'invalid' triggers an attempt to execute, and then prefilters step in and may transform it into valid code.

This distinction was useful when we were using InputSplitter in the Qt console frontend process: prefilters can use the IPython user namespace to determine what should happen, so they can't run in the frontend. Input transformers work without reference to the namespace, so they could be in the frontend. Now that we're doing the 'is complete' checks in the kernel, there's probably no need for that distinction to exist.

[minrk]

If there's a good reason, I'm okay wht that. I was just a bit surprised when I ran into it. Pretty low priority, I think.