[IPython-dev] ipython notebook installation instructions

[Fernando Perez]

On Thu, May 31, 2012 at 10:33 AM, Brian Granger <ellisonbg at gmail.com> wrote:
> We have started tho think about things like that, but haven't quite
> gotten there.  I think the main question I have is whether or not this
> transformation should be applied by the frontends or the kernel.  The
> problem with the kernel is that there could be multiple frontends
> connected to the kernel and it would be very confusing for those users
> to all of a sudden to switch modes.  Also, I don't like the idea of
> having stateful magics like this.  We have tried that with the %autopx
> magic is it tends to be very confusing.  I guess as I write, I am
> leaning towards this being a frontend transformation.

My thinking is actually looking a bit different right now, let me try
to explain.  For a long time we had the notion of a user-configurable
prefilter function, that would be given every line of input to
transform.  That's becaue IPython didn't have a concept of cells or
much else beyond the current line.  Sage (and others) plugged their
syntactic transformations via this mechanism, as was intended.  It
worked, but it's not ideal.

Now that we have a concept of cells, I'd like to see all users of the
idea of custom IPython-based environments to move to a cell-based
approach.  The transformations pipeline is now reasonably well
encapsulated, so we can come up with a clean API for user-defined
transformations that work at the cell level.  Users should be able to
define in a profile a function to be applied in the pipeline, that
gets a cell and returns a cell.

But that means that this would happen at the kernel level, not the
client level, and I think that's OK: these would be highly customized
kernels, and such that you'd want them to behave identically no matter
what client you're using to control them.

The above doesn't prevent someone from writing a client that does
extra stuff to its input before sending it to an IPython kernel.
Since the kernel is agnostic to the clients, people can already do
that and we'd never know.  Someone could for example subclass our
Qtconsole and write one that has additional transformations on all
input typed before it sends it out.

But I think that we should think up an official, clean way of updating
the old mishmash of prefiltering transforms for future use.  Sage
isn't the only project doing this kind of thing, so having an official
API for it is, I think, a good idea.

Cheers,

f