These are chat archives for ipython/ipython

7th
Oct 2014
Sylvain Corlay
@SylvainCorlay
Oct 07 2014 00:05
Another small issue is that in the case of meta classes. __class__.__name__ gives me MetaHasTraits
Thomas Kluyver
@takluyver
Oct 07 2014 00:24
@damianavila JB is trying to install live_reveal - there's both a setup.py and a notebook with installation code. Which one is the recommended way to install it?
Damian Avila
@damianavila
Oct 07 2014 01:35
the setup.py works very well
I think it is the easiest way...
btw... who is JB, I am curious?
Jonathan Frederic
@jdfreder
Oct 07 2014 16:02
@SylvainCorlay sorry I don't understand. What do you think it should be returning instead of MetaHasTraits?
Sylvain Corlay
@SylvainCorlay
Oct 07 2014 16:07
nevermind
:)
Jonathan Frederic
@jdfreder
Oct 07 2014 16:07
:)
Thomas Kluyver
@takluyver
Oct 07 2014 16:48
@damianavila thanks. JB Poline is one of the people we work with here in Berkeley.
Jonathan Frederic
@jdfreder
Oct 07 2014 17:20
@Carreau are you around?
Jonathan Frederic
@jdfreder
Oct 07 2014 17:26
@Carreau just wanted to check if your cm4 branch is rebased on top of master. When I try to rebase my cell tagging branch on top of it, I have to rebase some interact stuff too, which neither of our branches touches...
Jason Grout
@jasongrout
Oct 07 2014 18:31
@jdfreder - in your changes deprecating set_css, you now apply the simplistic css after the _css list. It should be the other way around---simple changes should be overridden by the expert interface.
(at least, it makes sense that the expert interface takes higher priority over the simplistic interface)
Jonathan Frederic
@jdfreder
Oct 07 2014 18:34
I understand what you're saying, but as implemented changing _css after one of the simplistic traits should cause _css to override the simple traitlets. The simple traitlets should only take precedence if set after _css.
Jason Grout
@jasongrout
Oct 07 2014 18:35
right. In the initial after_displayed call, you apply _css first, and then apply the effects of each simplistic trait
i.e., update_css comes before the various update attribute calls here: https://github.com/ipython/ipython/blob/master/IPython/html/static/widgets/js/widget.js#L484
(I think @scorlay also noticed this a few days ago)
uh, @SylvainCorlay
in other words, move line 484 down just below line 500
Thomas Kluyver
@takluyver
Oct 07 2014 18:53
Anyone else want to weigh in on #6642? I'm mildly inclined to reject it as a big cleanup PR, but I could be persuaded otherwise.
Damian Avila
@damianavila
Oct 07 2014 19:08
I think 3 of us is enough... but, let's wait a little bit... IIRC, we have this policy written somewhere, maybe we can link it in the issue as the closed PR for notice....
Min RK
@minrk
Oct 07 2014 19:09
50/50 for me
Damian Avila
@damianavila
Oct 07 2014 19:13
upppsss...Matthias switch sides... jeje...
Thomas Kluyver
@takluyver
Oct 07 2014 19:17
it does change behaviour - I guess the question is: is it more likely to fix subtle bugs we haven't identified yet, or introduce subtle bugs.
I'd guess marginally more likely to fix things - I can't imagine any situation where we want an overridden comparison to let == None be true.
so I guess I've shifted to +0.5
Jonathan Frederic
@jdfreder
Oct 07 2014 20:10
@jasongrout ahh I see what you're getting at.
@jasongrout @SylvainCorlay ipython/ipython#6645
Jason Grout
@jasongrout
Oct 07 2014 20:19
great!
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:19
Hi all - I'm having some trouble starting an ipcluster in the current git master. My engine logs look like this:
2014-10-07 14:15:04.794 [IPEngineApp] WARNING | url_file u'/home/goldbaum/.ipython/profile_default/security/ipcontroller-engine.json' not found
2014-10-07 14:15:04.794 [IPEngineApp] WARNING | Waiting up to 5.0 seconds for it to arrive.
2014-10-07 14:15:09.808 [IPEngineApp] CRITICAL | Fatal: url file never arrived: /home/goldbaum/.ipython/profile_default/security/ipcontroller-engine.json
I get similar complaints about ipcontroller-client.json when I try to attach to the cluster from a notebook
Min RK
@minrk
Oct 07 2014 21:22
@ngoldbaum and your controller log?
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:24
@minrk here's the shell session from invoking ipcluster:
[goldbaum@eudora goldbaum]$ ipcluster start -n 12 --engines=MPIEngineSetLauncher
2014-10-07 14:23:32.574 [IPClusterStart] Using existing profile dir: u'/home/goldbaum/.ipython/profile_default'
2014-10-07 14:23:32.586 [IPClusterStart] Starting ipcluster with [daemon=False]
2014-10-07 14:23:32.587 [IPClusterStart] Creating pid file: /home/goldbaum/.ipython/profile_default/pid/ipcluster.pid
2014-10-07 14:23:32.589 [IPClusterStart] Starting Controller with LocalControllerLauncher
2014-10-07 14:23:33.600 [IPClusterStart] Starting 12 Engines with MPIEngineSetLauncher
2014-10-07 14:23:40.341 [IPClusterStart] ERROR |
            Engines shutdown early, they probably failed to connect.

            Check the engine log files for output.

            If your controller and engines are not on the same machine, you probably
            have to instruct the controller to listen on an interface other than localhost.

            You can set this by adding "--ip='*'" to your ControllerLauncher.controller_args.

            Be sure to read our security docs before instructing your controller to listen on
            a public interface.

2014-10-07 14:23:40.342 [IPClusterStart] ERROR | IPython cluster: stopping
2014-10-07 14:23:43.343 [IPClusterStart] Removing pid file: /home/goldbaum/.ipython/profile_default/pid/ipcluster.pid
Min RK
@minrk
Oct 07 2014 21:24
And the ipcontroller log?
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:24
If you were referring to something else, let me know
I only see engine logs in the log folder
Min RK
@minrk
Oct 07 2014 21:25
The log of ipcontroller
Try starting ipcontroller separately
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:25
Hmm, I've never heard of ipcontroller before, I'll try that
Min RK
@minrk
Oct 07 2014 21:25
When ipcluster doesn't work, troubleshooting step one should always be: stop using ipcluster, and see if you can get it to work by manually calling ipcontroller and ipengine
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:26
ok, are there any examples of using ipcontroller to launch a cluster via MPI?
Min RK
@minrk
Oct 07 2014 21:27
starting engines with MPI is just mpiexec -n 4 ipengine
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:27
ipcluster was the only way I could figure out how to get this to work properly on this system when I initially looked into this a few months back
Min RK
@minrk
Oct 07 2014 21:27
ipcluster is just a fancy wrapper that does: ipcontroller; for i in range(n): ipengine
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:30
@minrk ok, I can get it to work if I do ipcontroller and then in another shell mpirun -n 12 ipengine. Is there a way I can invoke that in a single command? I guess that's basically what ipcluster is doing.
Min RK
@minrk
Oct 07 2014 21:31
can you do ipcluster start --debug?
Does it work if you don't use MPI?
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:33
@minrk very strange, it seems to work now - victory I guess?
Min RK
@minrk
Oct 07 2014 21:34
might have been a race condition
is this a VM?
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:34
the file system on this cluster has been super laggy today, that could be it
Min RK
@minrk
Oct 07 2014 21:34
ah, ok
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:34
if the json file doesn't get written before a timeout
anyway, thanks for your help debugging this
Min RK
@minrk
Oct 07 2014 21:34
so with MPI, the engines may not be on the same host as the controller
yeah, a laggy filesystem could be it
you can increase the two relevant times: IPClusterStart.delay is the time between starting ipcontroller and starting ipengine
EngineFactory.timeout is the time the engine waits for the connection file to appear
Nathan Goldbaum
@ngoldbaum
Oct 07 2014 21:36
in this case they are running on the same node - I'm using a viz node that has half a terabyte of memory and it's much more straightforward for me to parallelize using yt/mpi4py for this.
awesome :)
really appreciate your help
Min RK
@minrk
Oct 07 2014 21:36
sure thing
Sylvain Corlay
@SylvainCorlay
Oct 07 2014 22:49
hello - quick question, widget-hbox had a margin of 5px and widget-text of 0px (in widgets.less)
does any one know if there is a reason for this difference ?
margin : 0px !important;
Kyle Kelley
@rgbkrk
Oct 07 2014 23:43
vimtermjs.gif
Thomas Kluyver
@takluyver
Oct 07 2014 23:45
well done
though take it to the last word saloon ;-)
Jessica B. Hamrick
@jhamrick
Oct 07 2014 23:45
But vim is so horrible! :)
Thomas Kluyver
@takluyver
Oct 07 2014 23:45
and if you're going to start editor flamewars, we'll make a whole new room for you
Jonathan Frederic
@jdfreder
Oct 07 2014 23:46

@sylvaincorlay

widget-hbox had a margin of 5px and widget-text of 0px (in widgets.less)

Don't remember, probably has something to do with the appearance when the widget is displayed inline with other widgets with descriptions (i.e. interact)

Sylvain Corlay
@SylvainCorlay
Oct 07 2014 23:48
ok. i set everything to 0 by default in my own less override to align Text and DropDown Correctly in an HBox. The containing stuff may have some padding or margin
Kyle Kelley
@rgbkrk
Oct 07 2014 23:48
alright, alright. I'll tone down my editor :fire:
Emacs is fine when I'm using Atom
In all seriousness though, there was plan to use codemirror for editing a file?
In the case of images for demos on tmpnb, nano could be installed for now.
Not that the terminal stuff is merged. I only have that on an experimental branch.
Jessica B. Hamrick
@jhamrick
Oct 07 2014 23:54
I think there was some plan for doing file editing
Thomas Kluyver
@takluyver
Oct 07 2014 23:54
yes, there will be a simple text editor - basically fullscreen codemirror plus a save button
Kyle Kelley
@rgbkrk
Oct 07 2014 23:58
And if anyone wants vim or emacs mode, we point them at /terminals/new
Thomas Kluyver
@takluyver
Oct 07 2014 23:58
exactly
Kyle Kelley
@rgbkrk
Oct 07 2014 23:58
Love your work @takluyver
Thomas Kluyver
@takluyver
Oct 07 2014 23:59
well, they can use the vim/emacs modes available in codemirror
Kyle Kelley
@rgbkrk
Oct 07 2014 23:59
+10 fanboy points right here
oh that's a thing?
Thomas Kluyver
@takluyver
Oct 07 2014 23:59
woo, I'll start a collection
codemirror certainly has at least a vim mode
obviously for real vim diehards it's a poor mockery of the real thing
but that's what Paul's vimception is based on