[IPython-dev] embedding ipython

[Fernando Perez]

Hi Piotr,

sorry for the long delay; the git/github transition swallowed most of
my ipython time for the last few days.  Thankfully we're over that now
:)

On Sun, Apr 25, 2010 at 11:29 AM, Piotr Zolnierczuk
<piotr.zolnierczuk at gmail.com> wrote:
> I have been using an embedded IPython shell (sorry I am still at 0.9.1) in a
> wxPython app for quite some time and I (and my users) like very much.
> I cloned IPython.gui.wx.ipython_view to customize look-and-feel with the
> rest of my app.
>
> My app has a bunch of "GUI" tabs that control physics experiment hardware
> (neutron scattering at Oak Ridge National Lab) and one tab that is the
> IPython shell that allows for "custom" scripts. The main app needs to "pass"
> some objects into the shell, for example an object that is responsible for
> communication with the control hardware,
> so it can be used in the interactive shell. I used user_ns dictionary for it
> and it works for me.

Yes, that's the intent of the user_ns dict.

> Now I have a bunch of questions, so please bear with me:
>
> a) what is the difference between user_ns and user_global_ns  (in particular
> in IPython.gui.wx.ipshell_nonblocking)?

I'm not 100% sure about the wx one, but in general, those two exist
because  in Python,  the exec statement's grammar:

http://docs.python.org/reference/simple_stmts.html#exec

supports *two* dictionaries  to resolve names, one for local names and
 one for global ones.  This lets you control those separately.  I'm
afraid I can't recall exact instances where using both is needed, but
at least you know the purpose :)

> b) the example (wxIPython) as well I my embedded shell do not fill _oh
> dictionary and '_' is always empty so users cannot "recall" the results of
> the previous statement. The ipython shell works fine in that respect. Why
> wxIPython does not?

I think the wx shell code doesn't completely implement the entire
ipython machinery.  Laurent, the author of that code, may be able to
help out better, though he's been fairly quiet as of late.

> c) could "passing" of the objects  be done via 'configuration' file? If yes,
> how?.

In the pre-0.11 api, you could try to populate the user namespace with
objects loaded via this config file:

http://ipython.scipy.org/doc/stable/html/config/customization.html#ipy-user-conf-py

though I have to admit I'm not 100% sure if the wxip loads that at startup time.

In the 0.11  series we completely reworked the configuration system
into something fairly clean and rational where this can be done (and
if it doesn't work, we'll consider it a bug).  I'll announce plans for
0.11 shortly.

> d) could an external app execute a script (function) via IPython shell?

>From another process, not out of the box today.  You'd need to wire
some code inside your process that listens over a port/pipe and
responds to commands.  But writing such bit of code isn't particularly
hard, it just depends on whether you need only local behavior (case in
which a simple named pipe may be enough) or  you need network support.

> e) some users suggested a capability where user "queues" scripts and then
> some kernel executes them one-by-one. Anybody knows a good python solution
> for that?

Well, the ipython distributed computing apis do much of that indeed:

http://ipython.scipy.org/doc/stable/html/parallel/index.html

Cheers,

f