Older blog entries for sab39 (starting at number 25)

Advodiary

Maybe having advodiary to use will help me update more frequently... I've wanted to make more timely entries for a while...

"Killer App"

tk: I had read your last diary entry out of context, in that it mentioned wanting a "killer app" but didn't mention Delitalk explicitly. So the suggestion was just a "killer app" in general, not a killer app for Delitalk explicitly. Sorry.

Just for posterity's sake, I'll post the idea here also...

I have a number of cassette tapes that I don't have CD equivalents to, and most of these I simply cannot get CD equivalents to for any amount of money - either because the tapes pre-dated CDs being widely available, the band has split up and CDs are no longer available, or the tape was made by a "friend of a friend of a friend"'s band who I am no longer able to get in touch with (and who probably don't have the resources to make a CD anyway).

But I'd still like to be able to rip and encode these cassettes into digital form (Ogg, of course). It's relatively easy to use sox or equivalent to record the whole of one side of a cassette into a single wav file (using a walkman and a headphone-out-to-line-in cable). But the difficult part is then going through the cassette and identifying where one track starts and the next ends, and encoding those parts into ogg individually (and, of course, ID3 tagging them and auto-creating a playlist in the same way that abcde does for CDs).

My theory is that it should be possible to hook into xmms or something and let the user use xmms's position control to jump to the end of each track, hit a button indicating "track 1 ends and track 2 starts here" and, after repeating that process for all the tracks and typing in the track (and artist for multi-artist tapes) name, generate an ogg file for each track and a playlist.

Bonus points for making it easy to process two sides of a cassette into a single playlist, applying some kind of filter to the sound to eliminate tape hiss, and for coming up with an inventive UI for noting the exact point in the sound where one track ends and another begins. You could also support cases where tracks overlap slightly (eg the band starts playing the next song before the last chord of the previous one has died down) or where there's some dead space between tracks that you want to skip entirely.

Personal

Life is good...

mjg59: Interesting the way cam.ac.uk people are instantly recognizable by the format of their preferred nicks :)

Shame the RH/AOL merger rumours are apparently false, because I just thought of this AOLinux one-liner:

You've got mail(1)!

Well, I'm pleased with myself. I just wrote an actual useful feature for mozilla - or at least, an actual useful enhancement to someone else's feature.

The new link toolbar in bug 87428 is absolutely awesome (and not in any way written by me, all credit goes to what appears to be a pretty large group of people hacking on this patch). But I found it frustrating that a toolbar that's only useful a tiny fraction of the time (because essentially 0% of sites currently support it) should take up screen real-estate all the time.

I'm not the first person to have suggested that the toolbar auto-hide, but I am apparently the first person to do something about it. So the latest version of the link toolbar comes with a three-way visibility option: Show, Hide, or Show As Needed. And for the first time there will actually be some lines of code in mozilla written by me, which makes me extremely happy (I feel like a little kid running around to show everyone what I did - this feeling will probably pass, but I'm enjoying the thrill of it right now :) )

Oh, and in case anyone here wants to try it, the latest XPI in the bug works on windows (although you have to save the file to disk with an .xpi extension and *then* load it in mozilla) and there are linux instructions about 4/5ths of the way down the bug (from me dated 2001-07-09 10:28). Sites I've found that support the feature are w3.org/TR (if you actually go inside any of the recommendations) and surprisingly, the blackdown.org java-linux FAQ. Having a "Next" button is great for navigating these things.

Now if only lwn.net would reply to my letter asking them to support it... :)

I can't reply to the moneyflow article because I'm no longer an apprentice, so I'll comment here.

I recently read "The Case Against Micropayments" and found myself persuaded by it - mostly because I find myself that I'd much rather pay a flat rate for something than pay a few pennies repeatedly. I know how much "a few pennies at a time" can add up. And whenever any site (such as wsj.com) I encounter has a subscription model, I think instinctively that "they don't get it".

So what might happen instead? I can think of one scenario, combining a couple of ideas I've read elsewhere. I've heard many people say they'd gladly pay to get rid of ads on the pages they view. I can certainly imagine being prepared to pay X dollars a month for the privilege of ad-free surfing.

So how would such a thing be implemented? Well, DoubleClick could probably do such a thing unilaterally, or it could be implemented by a cooperative group of sites that use advertising. The principle would be that every time you visit a page on one of their sites, it first checks a cookie to decide whether you have paid or not. If you have, it doesn't serve you any ads.

The payment would have to be made to an external organization, and should ideally be someone who can be trusted not to track personal information. This entity can probably tell every site within their "network" that you visit, so you would need to trust them to not use that information for anything other than distributing the money.

Since each page access (or at least each browsing session) would have to involve a round-trip to the external organization's server, they can check how often you visit the site and give that site a corresponding proportion of the proceeds (ideally at the same rate as a single ad view would - or an appropriately weighted average of the rates for a clicked-through vs non-clicked-through ad).

Just a thought...

Ooh, I'm back down to observer. Fair enough, I haven't posted a diary entry here since January...

Anyway, here's my rant for the day: Linus' and Alan's changelogs. Suck. When they merge changes with each other, they just say "Merged with Alan" or "Merged with Linus" - they don't say *which* of Linus' or Alan's changes they took. With Alan's you can at least safely assume he took almost everything, but I have *no* idea whether 2.4.2 contains the Matrox G450 changes from 2.4.1ac1 or not. And I need them, because (guess what?) I have a Matrox G450. And Linus' changelog only says "sync up more with Alan", which is no bloody help at all.

Challenge to all perl gurus out there... can you make the following script any shorter?

Original (80 char) version:

$|=++$x;for(;;){$i+=$x;$x=-$x if$i>77|$i<1;print" "x$i."* \r";select$u,$u,$u,.1}

Current (58 char) version:

select$u,$u,$u,.1while$|=print" "x(abs 78-++$i%156)."* \r"

I'll permit changing the phase (starting at a different point in the cycle) because I already had to do that to get to 58, and changing the speed provided that the sleep is neither removed altogether or increased to 1s or greater. Beyond that, the output should be visually indistinguishable.

I was convinced that I couldn't get it shorter when I was at 80! Still, it really looks like it's close to optimum now - the 80 character version had lots of obvious redundancies, I just couldn't figure out how to remove them. This one doesn't.

caolan: That's cool, but I disagree that it can't be done smaller...

sballard@rainbow:~$ ./test; echo $?
42
sballard@rainbow:~$ wc -c ./test
     18 ./test

Of course, I cheated...

sballard@rainbow:~$ cat ./test
#!/bin/sh
exit 42

Badger: Someone on one of the Moz newsgroups posted a fast-update perl script - it would use the Bonsai webserver to identify the files that had changed, and then use CVS to download those only, bypassing the usual CVS negotiation of these issues. The discussion suggested that this really helped, although it seems to me that CVS really ought to be more efficient at this kind of negotiation if a *webserver* can provide the required information so much quicker. Try looking through n.p.m.builds (and ask if you don't find it).

lkcl: See EVE - a massively multiplayer space/strategy 3D game with the servers written in Stackless Python. Incredible screenshots, etc. No mention of whether it runs on Linux that I could find, which seems strange when they're using an open source language. Now if only { I had a decent graphics card || Xfree supported acceleration on the crap one I do have } ...