enjoying salad since 1978.

Thursday, May 17, 2007

Google Reader's Birth!

Wetherell is determined to start a low-tech folklore-a-like about Google Reader. He drug me into the mix in Chapter 1. I would like to clairfy a few points:

I didn't say he sucked.

Chris does not suck. He said to me: "I have this framework for parsing atom feeds in javascript but I want it do more so I'm going to start parsing RSS in it."

I reeled at the thought of parsing 9 incompatible versions of RSS some of which were so badly generated by tools of the day that XML parsers threw up when they ran across them. "Are you freaking nuts?" I asked him.

It took me about two seconds to tell Chris: "Let me just make a tool that gives you everything as Atom."

I think I did make him watch me write it.

I can't remember why. It wasn't more than 20 lines of Python. I had already done the bulk of the work a few weekends earlier, I had written atomfeed.py which generates atom 0.3 feeds so I slapped that together with Mark Pilgrim's Universal Feed Parser, made it a CGI, did the dirtiest thing possible and put it on the company intranet webserver. I later made a fancy scalable version in a fancy proprietary google webserver (thanks to greg stein) but that CGI ran the prototype for the first 100 in-house demo users, I think.

I didn't make Chris dance the jig.

Sometimes I'll do a little dance when something goes right. Everybody should dance! Chris agrees!

Sunday, May 06, 2007

We want a new layer!

Van Jacobson's 'A New Way to Look at Networking' is one of the more interesting talks I've seen this past year. He essentially goes one step farther than P2P pundits by saying that not only should content live in the cloud but that TCP's conversational model is harmful to this vision by allowing third parties to get in the way forcing debates like Net Neutrality and QoS.

[Update]After thinking about this, it's not correct to say 'farther than P2P pundits' as some p2p systems like khashmir use protocols other than TCP. Khashmir, in particular, uses Airhook and p2p hackers, for a while, were getting rather infamous for trying to bolt lots of tcp's features onto udp with each group choosing their own favorite set of tcp features to badly emulate in udp.

I'd like to be able to toy with building protocols directly on top of IP or even just on bare link layer but that doesn't seem too feasible. I'd like to say I have better things to do with my time... but I don't, really.

Friday, May 04, 2007

So Novell sells off NetMail and Hula and Hula forks as a result.

Robert Love just left Novell as their Chief Architect of Linux Desktop.

Nat hasn't blogged in nearly a year.

I wonder what is going on at Novell. I wish they blogged as well as Sun.

Wednesday, May 02, 2007

A few weeks ago, when Peter Norvig wrote his great article on writing a spell checker, I ported it to Haskell. Unfortunately, it was super slow and ugly. So ugly and slow that I was too embarassed to put it in my semi-public sandbox. I figured this is because I simply don't write that much Haskell. Sadly, it seems I'm not the only one to having trouble. This thread on fa.haskell is a great way to get some fresh ideas if you wrote your own slow port. Of course, theirs is still slow but less ugly than mine.

Next, I'll port it to Erlang and see if I can get it faster. One immediate upside is that mnesia gives me persistence and distribution for cheap. It won't be a fair comparison since I have far more experience with erlang than haskell.

As an aside, I haven't had time to put up my rome photos and diary yet as I've been fighting an ear infection I came down with after a jetlag-induced cold.

Another nice thing about not working: I don't have to feel guilty for taking sick days.