|
2003-12-31
|
Dave/Dina
I took a week off from work this week to clean up bits of my life and work on Dave/Dina. I've made some excellent progress - I was able to take apart large chunks of Anaconda (the red hat installer), and replace text, images, and functionality. It's my first experience with python, don't know yet what to make of it.
Red Hat also has some scripts included with Anaconda, which I only realized halfway through creating my own. So I switched and started customizing those. They generate all of the boot disk images, cd boot images, and so on by themselves, based on the RPMs in the distribution dir. Basically, the whole process bootstraps itself, which is pretty impressive.
Meanwhile, I also switched to Red Hat 7.3 packages as a base, and reworked the kernel packages, to include a kernel-BOOT rpm which also has the right banner. As I understand Red Hat's policy, they want you to remove all references to Red Hat from your custom distribution so as to not reflect bad on their name. I can understand that, of course, but you have to know this stuff, otherwise you'd think they WANT you to acknowledge you using their stuff.
I have run into a bit of a problem though where I get a GdkWindow error during the install, which doesn't put me in Pdb, but just reboots the machine. That's nasty. So I started adding debug code all over the place to at least get some feedback. I wish I could make these errors non-critical, the way you can for glib, so that at least I can get some sense of where it's at.
I've also resolved myself to using VMware to simulate the install. It is a LOT faster then using a real machine. Basically, you connect the bootnet.img to the floppy drive, which alone saves you a minute in loading time each try.
I'm doing all of this because a friend of mine bought all the parts for his own machine, so I'm guilt-forced into finishing parts of this up. That's great, I needed this kick in the butt. I hope having an actual .iso image that can be installed might get some attention and bring in new people so we can actually start working on planning and implementing some of the features better.
I find it amazing how much you can learn over time. There' s stuff I do today on this project that I wouldn't have dreamt possible last year. It's also fantastic how stuff you do on other projects helps you immensely, and how much you can actually draw from your own experience.
For example, using bitches, the RPM build tool I wrote for GStreamer, helps me so much in Dave/Dina as well. It's simply amazing how much time this sort of thing saves me.
The reverse of this is also true: when I look at scripts I wrote five years ago, or code for my thesis, I'm ashamed ;) So the big question is : What will I think five years from now about stuff I did today ?
GStreamer
Slowly gearing up for a 0.4.1 release. Doing lots of bug fixes. I'm personally concentrating on <hype word>Out-of-box experience</hype word> for end users. There are some really simple-to-fix bugs for first installs which all of us never encounter since we all have GStreamer on our machines since ages. They're embarassing, but very educating to go through. Every project should do this.
wlach, thanks for the good words on my presentation. The latency issues during that demo are totally normal, however. Maybe I didn't make that point clear enough back then. Suffice it to say that GStreamer has everything you need to control latency the way you want it to. Latency is an often misunderstood issue. The demo didn't have any sort of latency control at all. It just sent out data each time the device was able to receive. There was no sort of buffering at all.
Basically, it's always a tradeoff between "hiccups" or underflow and buffering. Increase the buffering and you decrease the risk of underflow. Of course, increasing the buffering can be bad for your responsiveness, unless you take measures. These are all application-level decisions. GStreamer provides stuff to be able to do this, but doesn't enforce any kind of policy on that topic.
So the demo was basically bufferless. It was GStreamer in it's raw mode. I could have added a buffer in the pipeline, but I was just too lazy.
Pukkelpop
Going to the Pukkelpop festival today. Great stuff - going to see a bunch of good bands, including the Breeders, And You Will Know us by the Trail of Dead, Jimmy Eat World, Promise Ring, Jon Spencer Blues Explosion, and more. Even Guns 'n Roses are supposed to appear !
I wish I had applied for an interview with the Breeders sooner. As I said before, Kim Deal is just too damn sexy. Hope I'll get to meet her today if I get there.