service stop DDC; service start OLS
[I'm dropped out from
Planet GNOME again, for the
same old reason. Seems like it's sensitive to conferences, last time it was on GUADEC 6's first day, this time DDC'05's first day.]
Random comments that I got during my talk:
Keith Packard (keithp) noted that X contains a couple hundred kilobytes worth of locale data, parallel to glibc's, and that probably can be replaced by a central locale data library too.
Daniel Veillard (DV) is seeking a lightweight (<30kb) way of having locale-aware collation in libxslt that supports ascending/descending and lower-first/upper-first/locale's-default options. So far he has only found that in ICU, but ICU is multiple times bigger than his whole library, so that's no option.
Anyway, DDC second day went pretty smooth. DV and I got together to find a solution for his problem, and I started hacking up a solution based on strcoll functionality of the libc. The main problem is supporting that ascending/descending and lower-first/upper-first/locale's-default thing.
here is the part of the XSLT spec about sorting. Doing all these in a not-that-slow way was a torture that I managed to survive after a few hours of hacking. Good hacking for a day. The code is available
here and is in public domain. The algorithm is described in the comment at the beginning of the file. I would like to hear from people whether what I'm doing makes sense, is there any easier way to do this, etc.
Later in the afternoon we headed out for dinner with
David Schleef (ds),
Chris Lahey (clahey), and a few other people. Caught
Michael K. Johnson (mkj) in the elevator and had a long conversation about
their packaging system and sitro. Then ran into
Colin Charles of the Fedora Project fame. Ah, he recognized me as a Summer of Code student of them, but I still told him that I've not started the C code yet. So we all had a good dinner and left for the final party at Final Party at Vineyards Wine Bar & Bistro. This time we spotted
H. Peter Anvin (hpa) wearing a cool t-shirt (got photos, will post later) and his wife Suzi. While talking standing still an hour later there,
Dave Jones (djones) joined us and another half an hour passed before we left them and headed for the party.
The party was awesome. Lots of good drinks, chatted with
Hubert Figuière again, and more with
Jon Phillips (rejon) and Bryce Harrington, two
InkScape developers. Also got photos of Colin Charles and David Zeuthen (davidz) taking photos of eachother! Will post them all really soon. Also met
Soeren Sandmann, which was good. I finally figured out why
planet is called planet (apart from the obvious reason): After hooking my weblog on
p.g.o, all the
planet knows that I work on Persian support, I didn't make it to
6UADEC, and even that I finally found a place to stay for
DDC!
OLS started pretty good too. I was in
bert hubert's talk about
On faster application startup times: Cache stuffing, seek profiling, adaptive preloading (he mentioned me working on
preload too) when DV wrote on IRC that he's looking at keithp's eye-candies. So I walked to the next room and got some of the eye candy at
Keith's talk. I've been wondering whether is possible to do a decent desktop in say 32MB, and decided no, but this talk started changing my mind and thinking what should we do to get an order of magnitude (not really, but) smaller.
Got a chance to talk to
Seth Vidal, the hero behind
yum. Seems like latest yum is doing all items in my wish list. So for example you can say "yum install /usr/bin/mencoder" instead of the nasty "yum list *mencoder*" (in
livna they have packaged it separately as mplayer-mencoder, girrrrrrr.) There's also a yum shell, yes! No more "yum list this"...takes 30sec..."yum list that"...takes 30sec... And they are aiming for transactions, so you can run the shell, say "remove sendmail", then "install postfix", and then "run"... Beautiful.
If I find some time tonight at the
welcome reception, I would like to hack on
dasher. They need some help with autotools, I will fix all that, and if time remains, work on fixing the cpu-usage problem. Currently dasher in a stopped state still eats ~40% of my 2.4GHz CPU! It should not. After that is to cache the language model so it doesn't eat 100% CPU for 10 seconds on startup. Way to go...
And finally, in a few minutes is the keysigning that has brought
lots of signatures to my
PGP public key.
