March 11, 2011

PyCon Lightnings, 17:30 March 11, 2011

Session starts almost 20 minutes late - looked to me like logistical problems with the stage change, but I am so far out of the loop and the people in charge are so competent that the best thing I can do is start the post and wait patiently like everyone else.

I am sitting with Titus Brown and Peter Conerly, who are discussing matters of great technical moment. It's Peter's first PyCon, and the opening announcement is the every lightning slot in the conference is allcoated.

The opening talk is a request for PyCOn to raise $10,000. If 1 in 5 gave $10 each ... Text REDCROSS to 90999, then tweet that you gave with hashtag #pythoncares. End of talks on Sunday we will get a count.

Next speaker's display isn't working. Awkward pause. Silent audience. I am hoping things soon run more smoothly for Jacob (Kaplan Moss, the evergreen sprint chair who is here for the first time as a gentleman farmer).

Jeff Daily talks about Global Arrays Parallel Programming Toolkit, that added Python bindings to it toolkit in the last year. It gives access to a high performance computing array, giving access to distributed numpy-style ndarrays. Compared with other models, Global Arrays is less general than some techniques, but it scales well to at least 200,000 cores. This finally achieves highly parallel programming with true Python programming ease. Much simpler than message-passing interfaces, the technique is extensible and scalable, but requires that algorithms acknowledge the need for data consistency. If you're at PyCon, see Jeff at Poster #13.

Matt Harrison talks about his geek inferiority complex. He bought a computer at Fry's because computer class at school was too full. For years now people have been declaring "the year of Unix". Linux people run stuff like Gnome and KDE and a tiny set of other window managers. This guy inhabits a pretty specialized point in the Linux desktop fractal space, a rarified atmosphere in which indexing by 1 is awesome. And still we get more specialized. Matt's wrists forced him into this (in)sanity by screaming when he tried to pick the mouse up. Impressive demonstration of driving a window manager from the keyboard ensues, and lo! relevance: the QTile window manager is written in Python and is scriptable: now windows are flashing on and off in four different virtual screen. 2 further users will amount to exponential growth in the user population. Please join him.

Pete Fein wants us to hack for freedom. Video of protesters, under attack by flame throwers, getting pictures out through cellphone. Pete explains how he has helped freedom by getting stuff out on the street in 105 cities in a week. He worked with telecomix.org to help keep sites up and services running. They are leaderless and fanatically devoted to freedom. When the net was up they provided encryption and other services, mirrored censored sites and manually relayed tweets for egyptian users with no network access later. With Anonymous they set up a network to send information by fax, and by injecting information into logs of a few servers they found still up. It can be exhausting working for freedom, but we have the power to help people achieve it.

Next speaker (fumanchu at aminus.org(?)) not using his own laptop, another pause but we are now seeing images. No, gone again. Back again. Go. Logging. Really? Yes, because all the modules in your application can use the logging model independently, but this isn't applicable for things like performance statistics so that independent modules can log performance data and report on it. Solution: log into Python dictionaries, with package names as the top-level namespace. You can define record structures, and (for reporting only) can insert functions into the dict to be used when reports are required (writing the log should be high speed, so no function calls). This allows people to build their own reporting tools, or to user ready-made ones such as the one built into CherryPy. Open Space tomorrow.

More AV glitches as the next speaker arrives, then departs so I have no insight into the problem. Ah, they need a VGA dongle for a mini-display port. Strange, PyCon normally has this stuff buttoned down. I hear someone say "waht are the odds" and don't know whether this is a good sign or a bad one. BVideo at last appears, and Nathan Nichols wants to tell us some lessons he learned writing automated content generation systems using DSLs. Clients send data, they create quality content for it. Journalists are needed but hey they don't know Python. So instead use DSLs. Rules: 1. Keep Parsing Simple, Stupid (PyParsng was overqualified for the job, regexes and YAML serve better); 2. Keep things on rails (no, not Rails - Nathan knows his audience) and remember that while we think Python is awesome journalists don't and would be overwhelmed by its complexity; 3. Tools are super-important, and DSLs allow you to write very specific tools. Good talk.

Another pause, this time to load the next presenter's files ont he same laptop. It's Giles Thomas wants to tell us why Resolver Systems switched from IronPython back to CPython. Resolver are known for their eponymous spreadsheet, which is quite clearly a windows application. They now have a new project, Dirigible, was written for the web to provide similar services. The initial strategy of marketing to financial companies, and Windows was mandated bh the conservative nature of the client base. So Resolver 1 was a full .NET application. For Dirigible, CPython was the natural choice since none of the .NET stuff was required.

The SingPath competition is about to start somewhere close by. More lightnings in the morning. Awesome day 1!

No comments: