Older blog entries for cpw (starting at number 9)

I've been wondering why Linux distributions all end up being Unix-like, but I suppose it's obvious. Compatibility, and that fortune that goes 'Those who do not understand Unix are compelled to reinvent it... poorly.' And who really understands Unix?

I'm occasionally tempted to implement a distribution which keeps its packages in packages (much like the nextstep/macosx 'bundles') and has a shell that can find stuff in them. It should be easier to deal with upgrades and version skew this way. The problem is that I don't know what bits will be harder this way.

init is another candidate for reworking. It won't start daemons for ordinary users, it spawns some things from inittab and some from the init.d scripts... It's all terribly ad hoc, I feel.

I've been trying to get used to click-to-focus recently, seeing as focus-follows-mouse is relatively unpopular across GUIs and I've not seen any evidence that either is better.

The most annoying thing about it is that new windows (Netscape, for example, which takes a while to start) steal focus from the terminal I'm typing in. One can turn off focus for new windows, but that's not quite right when you're not doing something in another window. The best solution would give the window focus if typing wasn't currently occuring in another, I would think. Opinions?

raph was wondering how to handle discussions in diary items. I guess the Wiki approach would work, but there are others.

• In-Reply-To field in diary entries, and a threaded presentation option
• Diaries could become like personal instances of the front page. The diary owner would be the only one able to post top-level articles, but anyone could post replies.

We may be able to rely on the trust metric to keep people from editing what they shouldn't, which is the factor that's unfamiliar to those of us from Usenet and other BBSen, and might allow a Wiki approach

I wonder how Wikis handle people screwing up and blowing away entries. Do they implement revision control on the entries? I guess I should go look. Later: OK, some of them do. Good.

I wish someone would solve the 'endless maze of twisty little categorizations, all alike' problem ;)

Xfree 4.0.1 fixes the VC-switching problem. Excellent.

I've decided I like Sawmill. Easy to configure. Will let me fit three terminal windows in 1024 vertical pixels with minimal tweaking. Pretty, but that one's not part of the acceptance test.

Mandrake as a whole - it's OK, but I'll try Red Hat next time.

Thanks to jbowman, chromatic and strlen for the certification props.

OK. Mandrake 7.1. Kinda pretty. Some odd dependancies and package categorizations, some of which I can think of explanations for. Their XFree86 4.0 configure program is a little on the pathetic side, but then again it's very new. (It wouldn't believe my 32MB TNT2 Ultra could do 1280x1024x32, when it actually goes way, way beyond that.)

XFree86 4.0 is a little different and a little broken. Changing back to text VCs doesn't (fortunately, changing back from that to X does), and I've had to tweak my DPI settings again. Does anyone know if the VESA DDC standards include a way for the monitor to tell the system its physical size? mail cpw at catenoid dot com if you know the answer (or just to say 'Hi!').

Fortunately, this appears to be 4.0, so maybe I'll get VCs back if I can upgrade to 4.0.1.

I still think that the kernel should be able to arbitrate access to the video hardware, so you can run multiple buggy graphics systems and have the kernel wrest the video hardware from one and give it to the other. I suspect the GGI people said similar things, though, and Linus didn't like it much. Clearly he needed to play more svgalib games, which would quite frequently do nasty things if you tried to change VC while they were running.

Quality vs. quantity: glade's separation of XML-encoded UI presentation and program code is the kind of thing that encourages not-confident-coders to participate. Embedding data in the code is Bad, because code is Hard to Read.

I'd like to know how these mentorships get started.

The thing I keep coming back to tonight is this: If developers didn't have massive egos, why would they develop?

Picked up Mandrake 7.1. Looking forward to direct- rendering OpenGL. It should be good practice for the installfest, too.

I wonder if Compaq can be persuaded to open-source VMS? It has the kind of rabid fans that could really make such a project work.

Skud's Quality vs. quantity article was interesting. I wonder if people are afraid to hack other people's code because they once tried to comprehend their own code after a two week break, and figure that it'd have to be that much harder with other people's code. If this actually is part of the problem, I guess the solution is literate programming.

Skud's article also reminded me of something I've been thinking about for a while - The Trouble With Freshmeat. If you look at the entries one after the other it looks like a million monkeys wanking on typewriters, which is depressing as hell to watch.

I reckon that free software development should look like that - unless you can filter based on topic (which fm.net allows to some extent) and other criteria such as stability, installed base, user-friendliness and empirical reliability metrics.

You might also want to check the metrics for each project's dependancies and work those into the evaluation. For example, project foo may look OK, but it uses barlib which is a bitch to install, so I don't really want to know about it right now.

I suspect that group-trust-metric-like objects might be useful in evaluating scores for the more warm and fuzzy criteria above.

The front page of useit.com has a quick screed on the merits of careful manual line breaking for small-screened devices. It suggests _ you should attempt to _ break the text _ into meaningful units.

For a device-independance Nazi like me, inserting BRs seems daft.

I wonder how hard it would be to do this automatically using an approximation of an English language parser. Here's the end of a noun phrase - good place to break!

Alternatively, perhaps some more aspects of the text structure could be encoded at authoring. Authoring, however, is already too much work.

Damn natural language problem.

Today is my day at Interact+IT for Linux Users of Victoria. I was hoping that the several years I've avoided such things would make this more fun, but judging from Skud's report it won't be enough. Oh well. Perhaps the LUV contingent will show up. Perhaps people will ask good questions. Perhaps the stand will have a machine with Unreal Tournament installed.

I really need to get XFree86 4.0 up and running on my home machine. That will probably be a good opportunity to rant about nvidia for a bit.

Later: LUV's stall was hopping with LUVsters and punters alike - a success, I'd say, and I had a good time. The rest of the show was lamer than I'd expected, though. Oddity of show goes to the two Acorn True Believers hawking ARM/RISC OS point-of-sale boxes, showing off their custom motherboard and talking about ATX power supplies that couldn't cope with it sucking as little power as it does. It struck me as sad in an Amigaesque kind of way - those Archimedes boxes were so damn cool in 1988, but x86 has triumphed over clever design with clever implementation driven by huge volume sales.

Amigas, Archs and NeXTs are gone, and MIPS, PPC, and Alpha seem to be going. ARM, MIPS and perhaps PPC will survive in portables and embedded systems. Alpha, the only performance contender left, has no such bolthole.

I worry.