I started trying to use Notebook and showing it into a QtWebKit view, but I decided to change that in the approach I did for a desktop Sage app for these reasons: - First of all, the Notebook is very very big in memory. And what we really need is not a full Twisted server, just a way to talk to Sage and get the results back. Using Sage inside the Python process allows this (as using exec() in Python allows to have custom environments, that is what I'm using when two sheets are run in parallel in Lorien Labs), and also avoids the neccessity of marshalling data from and to a different process, making fast to get things like the variables list. - Also, the Notebook allows to save, control versions, share... things that in a local app should be handled by the local app (in my opinion) - Finally, the Notebook code was a bit difficult to adapt for using custom widgets from Qt (for example, in Lorien Labs the text inputs are not normal HTML fields, but complete Scintilla editors). So I took the parts neccesary to execute the code (basically sage_preparse) and used it to create the new application.
It would be great if the Notebook could be separated in two parts: the part that shows the web view to the user and the part that actually makes the job in the worksheets. But for now this is not done, so I think executing the code using a combination of sage_preparse + native exec() on Python on a custom environment is the best way to go. Each of the worksheets have a separate worker thread where this execution is done, so the entire UI is not blocked, and endless loops may be stopped, very much like in Notebook. On 11 ene, 05:56, ghtdak <gl...@tarbox.org> wrote: > On Jan 9, 7:50 pm, ghtdak <gl...@tarbox.org> wrote: > > > On Jan 9, 2:20 am, Alejandro Serrano <trup...@gmail.com> wrote: > > > > Hi, > > > I'm currently developing something very similar as the thing you are > > > proposing. I sent it to the list about two months ago > > > (http://groups.google.com/group/sage-devel/browse_thread/thread/53b3d6... > > I took a look and its interesting. But I have some questions. > Perhaps, though, it might be easier to describe what I was trying and > why so you could contrast your approach. > > First, it seemed that exploiting the existing notebook capabilities > was going to yield the best reward / effort. Knowing that QtWebKit > and the browser demo which comes with the Qt distro works with Sage, I > figured that the easiest first cut would be to build a simple > application with an embedded browser window and hook into the > underlying Sage Notebook server process. This way, all the notebook > code could be exploited and I could focus on whatever specific > weirdness I wanted without needing to do major surgery. > > Unfortunately, this approach wasn't possible with the Sage Notebook > process as it existed primarily because it wasn't an asynchronous > process. It blocks on the pipe to the main Twisted Server process. I > had some luck using Python threads for some workarounds but, > unfortunately, this made nailing the Qt main loop tricky because Qt > really wants a single thread for gui applications... Since I was using > PyQt I didn't know what tricks I would need to try when integrating > with Cython etc... so I kinda punted. > > Some things have changed. First, it is more straightforward to create > a custom python process using Cython. While I haven't tried it, it > seems that one could start up all the Qt stuff before invoking the > notebook process code... this approach requires understanding how the > Twisted server worked and getting in the middle of the notebook > process launch... I don't think its very hard but the few times I > poked around didn't yield much success. Even with this approach, > however, threads would be needed meaning synchronization etc and I > think it would be architecturally incorrect. > > There was occasional back and forth on making the notebook process > asynchronous. I haven't been following so don't know if asynchronous > is part of the current notebook rewrite effort. > > In my perfect world, I would have an asynchronous sage notebook > process supporting the browser window... which would require no code > change to Sage, and I would simply add components to the event loop to > support other activities. For example, Mayavi2 could work alongside > the notebook sharing the underlying data structures with no concern > for synchronization which could get tricky. Extensions could be > written to support notebook invocations of Mayavi2 but they could come > in time... custom widgets would initially be used that would have no > impact on the Sage baseline code. > > I previously built a Twisted Qt Reactor to do this for PyQt but it > requires that the supported processes be asynchronous... which the > sage notebook wasn't and may still not be (again haven't been > following) > > In looking at your code, it "appears" that you're providing some > capabilities which might be more easily handled in the notebook. I > really liked the variables being shown and I believe you're thinking > along some of the lines I was originally. But you mentioned you were > using threads and I wonder why and what kind... this will require more > looking on my part. > > I should say that my primary interest in having the notebook be > asynchronous was to interact with external systems. This is a > requirement which isn't shared by many in the Sage community so it > hasn't been a priority. It is also a non-trivial adjustment to the > notebook itself... the combination may preclude its development. > > -glenn
-- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
