Older blog entries for obi (starting at number 7)


I don't know if it really solves the problem you're having, but you can try to deal with the problem by using the so-called "nested set model".

It's much faster to SELECT all the results you want (you need only one select for a node including all its children (several levels deep), any way you want). But there's a cost when inserting a node: you need to update all the numbers yourself that get affected by the insert. So this model is good when you read alot, but don't write as much.

Other nice properties is that its extremely easy to count the number of nodes in a (sub)tree, find a level, do a "group by", etc.

As always it's a tradeoff. Check: this link or search google for "worm tree sql" or "nested set model"

... was great! We (hunger and me) were there to see some speeches and talk to some developers, and to give a presentation on Fresco (we were on the "open sessions" track). It was better received than we expected! We thought there would be a lot of criticism ("There's no need for this, we have X", "corba sucks, it's slow", "if you guys would be going any slower you'd go backwards ;)") - but we got none of that. People definately got why we think it's important, grasped the basic architecture and ideas behind it, and understand that it's a hard project to do. Some people even said in the questions afterwards that they thought it "was one of the bravest projects around". That's nice to hear, to say the least.

In general the level of the people at fosdem was just great: if you picked any random person there, in 90% of the cases they would tell you they were working on this or that open source project. Other events like LinuxTag or similar, are interesting for developers too, but it also has this business and/or user focus.

All in all, a great experience, and kudos to the organisation, it seems to be getting better every year.
At first I thought it was hilarious, but when someone sent me more of thesame, I started feeling that sites like this really make me want to become a vegetarian. I mean, why do they feel the need to market meat like this? It's not like all of a sudden a large percentage of their "market share" will become vegetarian or simply stop eating. I wonder if this kind of thing is considered normal or acceptable in the States.

Once again, life imitates a Simpsons episode.
mwh, jaq

The problem with procmail and exim is that its rules are too static to my taste. I'd like to whip up a filtering system for me and my users, where by simply using an imap client to move messages to specific folders you create your "rules".

On regular intervals word lists/tokens would be created for every folder (maildir), and incoming mail would be matched to the most appropriate folder (based on what's already in it).

This would make it simple for my non-technical users who cannot set up a procmailrc (or exim filter)

Additionally, sometimes I put certain private mail together with mail from a specific mailing list. (if it's on thesame topic or whatever). A dynamic/automatic filtering system would take this into account too.

Such a system is currently used for spam for a lot of mailfilters and MUA's - I don't see why it's use should be limited to spam/ham classifying.

mwh:Thanks for the pointer to ifile - i'm going to look into it.

mwh mentioned courier-imap and Maildirs.

In general I'm very happy with the entire courier-mta -imap -pop suite. It was a breeze to setup compared to some other mta's, and it's more than flexible enough for me.

I like the idea of Maildirs, but can't understand why it's not better exploited

One thing I'd love to have is a "categorizer", that, just likes the bayesian spam filters that are so in vogue now, goes over all my (sub)maildirs and finds the best match for a certain incoming message.

That way, I wouldn't merely have a spam filter: The minute I subscribe to a mailing list and I put some of the messages in a certain folder (maildir) through my imap mail client, it would put new mail that matches these messages in the right folder

Another cool little toy is Mairix, a program that can create maildirs popuplated with symlinks to messages in other folders that match a certain search criteria. Server-side vfolders. Very cool.

Uraeus said:

I also saw someone posting about MAS the other day and comparing it to GStreamer. Well we (GStreamer) do not consider MAS to be direct competition even if there are some areas of overlap. In fact we have met up with Mike and Leon on several occassions and are planing on making a MAS plugin as soon as MAS is released. We are also considering to propose MAS as the official sound server for GNOME if it turns out ok.

Judging by what you said, it seems like I misunderstood the scope of MAS or Gstreamer. Can you tell me what Gstreamers design offers that MAS's architecture can't provide? Why do you feel MAS is no direct competition - they also intend to provide plugins, video, audio, etc...


Work has been keeping me very busy. I really wish I had more time to work out some ideas for Fresco. Despite its seemingly slow pace of progress, I do think the project has an important future - I really think the features it provides are very useful.

By the way, hunger is very likely going to give a presentation about Fresco at Fosdem. Hope to see some of you there!


An article about it got posted on slashdot recently. It's a system where the user, instead of organising his/her documents by giving it a file name and path, attaches all kinds of metadata to it - a bit like BeOS's "live queries" I think. There's a lot of situations where a filename isn't really appropriate. While a simple path/filename organisation may sometimes be enough, it puts a lot of the burden on the user to keep things properly organised. And I do think some ways of organising your info really doesn't map to file/paths, even if you have the possibility of using symlinks. I think the time for such systems are approaching rapidly.

Newdocms is a good first step. I'm not sure its design for the "tree of concepts" is general enough. But it's really important to have some real code "out there", that people start to play with, to see what tradeoffs are appropriate for a system like that, and also important, try to find out some good ways to allow a user to interact with such a "document management system". After all, people are so used to folders/files, for such a system to be accepted it has to offer some very compelling advantages, and be at least as easy/fast to work with.

If you're interested you can find more info about newdocms here.


I'm a bit dissapointed with the whole "graphic drivers on Linux" issue.
  • Xfree86 is very good, but they seem very inflexible in exploring other directions - and as a community it seems they're very closed, as mharris points out in his latest diary entry.
  • Since the 3D drivers are quite linked to xfree, I wonder if we can hope to see 3D without X in the foreseeable future (there's fbdri or Mesa + Glide2 - but neither is really practical). I remember bringing this up to one of the xfree developers at one of the LinuxTag's - his response: "Why would you want to run without X?"
  • GGI/KGI seems a bit more alive lately, but they have quite some catching up to do.
  • I'm not against things like NVidia's binary drivers, but they make matters really harder. Recent example: I was really happy seeing that Apple brought out a nice new, small laptop - it has everything I need and a good price. But I do alot of my work in Linux, and since the machine has NVidia graphics, it means no 3D under Linux. I guess I could try to do all my work under OS X, but I'd like to have the choice at least.
  • ATI seems a bit better about releasing info to DRI. I cannot help but wonder if that will continue now that they have their own closed driver. The juicy bits, like the programability of their cards, may never be released in the open.
  • The other hardware makers just aren't providing proper information or driver support to the linux(/bsd) community. Matrox has 2D drivers for their newest cards where you need a binary-only module for even basic features (like DVI). SIS doesn't provide anything (which is bizarre, since they are probably the ones with most to gain from good Linux drivers). PowerVR are DRI but binary-only. Not a big deal in this case since the only non-x86 platform with powervr hardware I've seen is the dreamcast. 3DLabs... well, who knows what's going to happen there.
If a hw-provider _has_ to distribute binary-only drivers for Linux/BSD, it would be nice if they could at least:
  1. make sure they respond well to bug reports
  2. allow for flexibility in platform choice (distro, BSD, PowerPC etc)
  3. try to provide the same features as they do on Windows
  4. if at all possible, a sunset clause would be nice: "If we don't provide support anymore for some reason (company goes bankrupt, change of focus, we suddenly don't care anymore, whatever) - then we'll open the source"
Under those conditions, I don't think anyone would reasonably object to binary-only drivers.

Back in the real world, I have to decide what hardware we are going to work on. Not an easy choice.


I'm looking forward to seeing MAS (the media application server) opened up. It's something like GStreamer, but network transparent (distributed schedulers, ouch) cooked up by the X.org people. It will be released under an X license. They're going to release the source end of january. I wonder why they've kept in under wraps for so long (years?). I've seen some demonstrations of them, and it's clearly quite powerful. But, maybe it's too late - there's already a lot of other projects that do similar things, and maybe they're good enough. Then again, there's no reason why several cannot coexist successfully (Gnome/KDE come to mind) - at which point it becomes a matter of taste.

sej: I'm always glad to see people interested in Fresco :-).

You said: I'm quite sure that Fresco (and its renderers) have to rely on an intermediate pixmaps for that as well. And as long as the pixmaps are local to one process (or in shared memory), "grabbing" them is as simple as reading local memory

Actually, no. There's no "grabbing of intermediate pixmaps". Everything is composited in the DrawingKits, and if you're using the openGL DK this can happen on your graphics hardware.


sej: I agree with you on the general coolness of Fink. Being a debian user myself, I wish it would resemble it more, but I understand it's not always a very good fit with MacOS X. It definately makes my life easier though.

lilo: funny, i've been through exactly this procedure (to update contact information, using fax form) just last week. It succeeded but it was a royal pain. NSI was, as ever, extremely unhelpful.

To make a long story short, they've been making it very hard to do things like changing domain ownership, changing contact information, transfering a domain, etc etc. I genuinely think their policy is to make things as difficult as possible to discourage their customers to make changes. The only thing that is very easy to do is (surprise, surprise) make payments.

I've been moving all my domains to Gandi too - I've been using them for about two years now and they've been extremely easy to deal with. I'll be very happy when the last of my domains will be gone from NSI.

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!