I have created a new blog called Decoded Node for future technical entries. Effective immediately, no new entries will be written in the Tech category here on Pawprints of the Mind.

This will seriously dent the posting rate here, but since only boring posts will remain, nobody will care.

I've been trying to improve my digital art skills lately, and it seems that the problem of the day is Gimp. I've finally figured out how to describe what's wrong with it.


I think the problem here is that GIMP followed the obvious implementation of their paint tools: every so often, another splat gets added to the canvas at that instant in time. Opacity is applied, the brush head moves without coloring, and 20%* of the way across the brush width, splat! another stamping is fully rendered. The quality problem arises because this series of splats interacts with itself, laying down ribs where the most overlay happens. It's not tracing the path of the brush and filling in color along the way.

Gimp's way is probably the fast way, but it's also getting in my way. I can't get around this texturing problem with the paintbrush, because putting down two strokes next to each other can never really be done cleanly—once the first stroke is rendered, if it's not fully opaque, I'll end up with either a gap or a darker spot if the regions overlap. The only thing to counteract this texturing seems to be to go over it with smudge or blur, which are too strong or not strong enough. At the edge of a selection, both tools create artifacts, because they use the color from pixels outside the selection, instead of extending the selected color. So, I get black or the neighboring colored area smudged into the selection instead.

I think what I want is a painting tool that tracks the region covered, and smoothly fills it in with color. Obviously this would take more CPU time, but realistically, it would have to get down to a quarter of the speed before it would be faster to clean up after the sloppy airbrushing with the smudge and blur tools.

* The default spacing is 20%, anyway. Smaller numbers yield finer ribs, until you get down to 1% which is pretty slow and moots the point of using a pressure-sensitive tablet.

Hot on the heels of yesterday's Chill theme comes the first (ever!) release candidate of Termitheme: 1.2-rc1.

Also, termitheme's repository is now public on github: sapphirecat / termitheme.

So get your own copy of termitheme-1.2-rc1.zip or clone it from GitHub today!

Today, termitheme 1.2-beta2 succeeds the previously unannounced first beta. Termitheme now lives at www.sapphirepaw.org/termitheme and the themes gallery has moved to www.sapphirepaw.org/termitheme/themes.php.

The web page has been expanded a bit, and the 1.2 series is distributed in a single zip file instead of separate 'built' and 'source' versions. [Edited to remove download link; this version has been superceded. Please check the termitheme project page for the latest version.]

1.2-beta2 brings the following new features compared to 1.1:

• Support for PuTTY on Windows.
• Credits can be included in theme files.
• Much improved support of non-ASCII characters in themes.

This is still a beta, because the last item has yet to be tested on Windows (or any non-UTF-8 Unix, for that matter), and the actual charset support is undocumented. There be dragons.

if [ "$SHLVL" -le 1 ]; then
  _ps1_code="0;7"
else
  _ps1_code="$(( ($SHLVL - 2) / 6 % 2));3$((6 - ($SHLVL - 2) % 6))"
fi
export PS1='\[\e[36m\]\@ \[\e[30;1m\]\h:\w \[\e['"$_ps1_code"'m\]\$\[\e[0m\] '
unset _ps1_code

This uses the value of $SHLVL to color the '$' in the prompt. A login shell will display with reverse video; deeper shells (e.g. when using :sh in vim) will get a color according to their depth. In this case, it goes 'backwards' from cyan to red, then bold cyan to bold red, before repeating the colors. Due to the special handling of level 1, reverse video is never repeated in the cycle. Also, since the SHLVL is supposed to be constant in a given shell, the prompt color is calculated once, and the result is inserted into PS1.

The deeper purpose of this code is to give me some sort of visual indication of whether I'm in the base level shell (reverse video) or a deeper one (color). The difference in colors allows me to notice if I unintentionally ran vim again from inside another shell, instead of exiting the shell to return to vim.

The preview is displayed in the BlackRock theme.

Quoth the Delicious Firefox addon TOS:

1. Licensed Uses and Restrictions.
b. YOU MAY NOT:
...

(iv) use the Delicious Software to operate nuclear facilities, life support, or other mission critical application where human life or property may be at stake. You understand that the Delicious Software is not designed for such purposes and that its failure in such cases could lead to death, personal injury, or severe property or environmental damage for which Delicious is not responsible.

It is quite unclear to me how one would use a bookmarking website addon for Firefox to operate much of anything, let alone anything like nuclear facilities or life support. Especially considering the restrictions in 1.b.i and 1.b.ii, which prevent any modifications or embedding of the software into hardware or firmware.

Considering the other provisions of this license—including their insistence on Santa Clara, California as the jurisdiction, regardless of any conflict of law provisions—I declined. I am now safe to operate nukes with my Firefox, I think. It's been a while since I read that EULA...

Looking at the Ambiance theme and the update manager tonight (as my administration is done by a separate user), I'm reminded of a few problems in both the latest Ubuntu release, and Ubuntu in general.

First is the decision to move the window controls in an LTS release. There are plenty of rants out there about the moving; my point tonight is that it was done, and replaced with nothing, in an LTS release. Assuming that the windicators or Super Magic Gesture Button arrive in a future Ubuntu release, it leaves everyone on the LTS with a half-baked start of a project that doesn't bring any value to the platform by itself. And if Canonical is thinking of supplying their next k3wl feature as an update for Lucid in 10.04.2+, that runs the risk of destabilizing Lucid—and making it less desirable for exactly the audience that the LTS releases are intended to address.

Aside from that, the new theme itself has some problems. The oranges in the palette for Ambiance and Radiance don't quite match the oranges in the icon theme, with the GTK theme being a bit more pink than the icons. It clashes just enough to make me wonder how Canonical could seriously be putting this scheme forward as its new, "professional" look. It also bugs me whenever I see a progress bar that hasn't gotten full enough for the fill to be at least circular; the overlapping end caps look entirely silly and occasionally pixellated. Again, this is supposed to be good enough to look at for years?

The update manager is lacking some fit and finish as well. Whenever I check for updates, it starts out on something like "1 of 17 files", then stops in the 20s somewhere, maybe a couple of times, and finally gets up to 39 files before it is truly finished. Each time it stops to download a big file, it's a file or two from the last one, and the bar almost fills up—only to drop down again when the next files are discovered. I've been watching this happen for the past 3 years that I've been on Ubuntu. Is it really that much effort to cache how many files/bytes there were last time for future estimates, or provide a complete manifest on the server?

Another thing that hasn't changed in the past 3 years in Ubuntu is the lack of multiarch support. You know, that Fedora 7 added to RPM about the time Feisty was released. If you want to run 32-bit code on a 64-bit system, you can't just use 32-bit package files; you need package files for the 64-bit architecture that have been carefully compiled to contain 32-bit code for the 64-bit environment. These are, of course, only produced for popular libraries, so if you have a package you want to install that depends on a library that's not available, you get to compile it yourself. If you can.

The missing multiarch support wouldn't be much of a problem, except that developers are generally producing 32-bit versions of packages before 64-bit ones. Flash, NaCl, V8, LuaJIT, Amazon MP3 Downloader, and zsnes, for instance.

Ubuntu rose to prominence based on its focus on the user, but lately—especially with the release of Lucid—it seems to be floundering. It seems that Canonical is working on experimental innovations at the expense of producing a really great product, and expecting everyone to fix the questionable decisions in Lucid by upgrading to Maverick. Thereby breaking their system and defeating the point of offering Lucid as an LTS release.

Slashdot linked to Jalopnik's story on the RSV on Friday. Among the comments is this gem from Kulprit442:

I remember seeing car ads from the past (just because I am an automotive pop culture nut) and it wasn't uncommon for the small cars of the 80's, 70's and even 60's claiming 30-40mpg...so why with all the technology in 30 years can we not make alot more vehicles get alot more than they are now. I still say conspiracy!!!!

The thing about those old cars is that they didn't have much horsepower. The RSV included the 1.7L engine from the mid-70's Honda Accord. The sedan of today offers as its smallest engine a 2.4L 177 hp version, with a 3.5L 271 hp V6 available on the more premium trims. In 1988*, the Accord only offered 98 hp in the DX/LX and 120 hp in the LXi, both with 2.0L engines.

Over time, manufacturers have offered more power, and buyers have accepted it. Today's 110 hp engines find their way into subcompacts ranging from the Toyota Yaris (106 hp) to the Honda Fit (117 hp), or end up as part of a hybrid system, as in the 2009 Toyota Prius. (Ye olden subcompacts often made do with about 66 hp, but nobody truly liked them.) I don't really blame manufacturers nor buyers for this, though. Power is a lot more fun than economy, no conspiracy needed.

Although weight could soak up the performance difference between modern and older cars, the trend appears to have been for power to increase faster than weight. That 1988 Accord is only 2482 lb. for the DX sedan, for 25.4 lb/hp. The 2010 LX sedan weighs 3230 lb, for 18.2 lb/hp.


* This being as far back as MSN autos provides data for it, and I'm not interested in doing heavy research on Memorial Day weekend.

Watching the New York Times, CNN, and others (recently GQ in the ads on Ars Technica) embrace the iPad so wholeheartedly, I wondered why they were doing it. What was so awesome about the device that publishers would embrace it in a way that they had never approached a tablet any time in the previous decade? I think the answers are all “design” and “control.”

The iPad is not a commodity. There will be only one manufacturer, one overall design, and one software environment. Although six models are being offered, they differ only in storage and network connection, neither of which have any bearing on the resources available to run actual apps. Effectively, any iPad is the one and only iPad from an app’s point of view.

The processor is always an Apple A4 at a single clock speed; the memory is always 256 MB; the screen is always 1024×768×32 and always driven by the A4’s integrated PowerVR chip; and your app is single-tasked, so the crapware a typical PC has will never slow it down and darken your good name. The only real variable is whether the screen is portrait or landscape, and even that only alters your dimensions by 33%. Unlike unrestricted PC hardware, which could be running anything from my venerable 5:4 monitor to a (comically short) 16:9 widescreen—to a 700×500 window stuffed in the corner of a much larger desktop.

All this, combined with the fact that iPad apps are not mired in the matrix of browser capabilities and technologies of the Web, gives designers a tighter set of constraints to work with, which allows them to produce designs that much better suited for the target device. It’s so much better to design in a framework of either-or than “anything from 800 to 2560 pixels wide, and whatever you do, IE will mess it up.”

iPad publishers also gain a measure of control over their content when it’s not on the Web, vulnerable to deep linking and copypasta aggregators like Google News. For additional control, I would bet Apple did the same for publisher’s apps as they did for iBooks, letting the publisher set the price rather than taking Amazon’s hardline “$9.99 or no Kindle sales for you!” stance. Finally, on competing platforms, there is a common DRM scheme in use platform-wide, which makes the payoff associated with cracking it much higher. A successful crack of a common system opens the entire platform. Whereas an app for each publication may only compromise a single app/publisher when a weakness is discovered.

So overall, the iPad is a much more attractive proposition for traditional publishers than the current crop of competitors. This will most likely remain true over time, as Android/WebOS knockoffs will try to out-spec each other in a race for the nerdiest audience, totally neglecting the other benefits that the iPad offers to developers. Both in traditional media, and in new development.

I’ve been putting together a podcast for running, using Audacity to record myself and mix in some tracks from my wife’s “Energy” playlist. Unfortunately, this has not been the smoothest possible experience. There’s no line in the time ruler to show where the mouse is in the track, nor in any other track. So trying to edit in one track based on the events in another becomes pure guesswork.

For its second trick, the UI for the “Auto Duck” effect suggests that it’s exactly what I want to fade the volume of a track down for a bit (to quiet the music while I’m talking), but the effect does not seem to change the track at all. [Actually, I read the manual, and Auto Duck uses the level of the track below the selected track to do its ducking. It has nothing to do with the current selection, as I had been expecting it to work.] So I’m stuck using the envelope tool instead to perform the effect by hand.

Which wouldn’t be so bad, except that it’s rather light on feedback. Just like there’s no vertical lines for timing, there’s no indication of what volume level you’re moving the envelope to, nor (as far as I can see) a way to set the handle to a specific value. Either by using a reference handle, or by entering a specific dB value.

So it’s all guesswork and trial-and-error, instead of having software that helps you do what needs to be done. Professional audio on Linux has a lot more problems than what sound server a distro chooses. Not that I’ve ever seen Sound Forge in action, but hey. What good is an informed rant on the Internet?

(OK, my transcoding of m4a files should be done now—the m4a’s were crashing Audacity on import…)

[And as far as Auto Duck goes, it might be nice to just have a Duck effect that ducks the current selection. It could also be convenient to have Auto Duck work more like an adjustment layer in Photoshop, so that the audio below could be repositioned in time and the ducked section would move to follow, instead of the ducking being permanently recorded into the waveform of the upper track. But that's probably a ton of work to develop.]

1. A vanilla lame -V2 inserts 0.025s of silence at the beginning of its MP3s.
2. Encoding adds a lot of peak samples to a track. For instance, there are none in Alone Tonight ripped from CD to FLAC, but half the track is covered in red (at the “view entire project” zoom level) for pretty much any lossily encoded version.
3. Assuming I'm reading all of the meters correctly, the noise is held to -12dB in the 192k MP3, which matches the actual dynamic range of the envelope. So it’s no wonder that it's widely considered to be transparent.

I am less surprised that the differences between the rip and the encoding are concentrated in the higher frequencies, where there is the most to gain by discarding data. Particularly since LAME runs a lowpass filter on the input, so anything above the cutoff band will appear as a difference.

They say that familiarity with artifacts is a major predictor of how well one will do on an ABX test. In that case, I think I'd rather stay unfamiliar and enjoy my 128k MP3s. (I tested myself with abx-comparator and got 3 of 10 trials correct…)

And yes, I am bad at this hiatus thing.

The older I get, the more important it becomes to keep my hands and fingers in good shape. Upon reflection, generally being at the computer so much is mutually incompatible with this goal. Although I have ergonomic keyboards at work and home, it still feels like I'm typing too much.

Something has to give, and so I've decided to stop blogging, and severely limit time spent on projects like termitheme. (FWIW, I'm also going to try to quit the news again. I am so tired of the iPad and Section 3.3.1 right now. It just never ends, and neither Reddit nor HN offer much in the way of bringing the really interesting stuff to the top instead of the 25 most popular Apple rageblogs articles. Of course, the mainstream's fascination with Tiger Woods is not much more compelling, hence the desire to quit all news.)

I picked up some handwriting recognition software, but it doesn't seem to be a real alternative just yet. (CellWriter, for the curious. I've written and edited this entire entry with it, in half an hour.)

The new Ubuntu branding for Lucid is "Light", which is not too different than "Aura" which accompanied Windows Vista. In fact, I was looking at the default wallpapers available on my Vista machine at work, and noticed this one:


It particularly caught my eye because I've seen Canonical's choice for their default wallpaper in 10.04:


There are only so many ways to craft soft, glowy light themes, I suppose.

At work, I write PHP either directly in vim on the Linux test server (when debugging), or using gVim on Windows to edit files in my private tree. Today, I made a couple improvements to the setup, one to each environment.

First: using the keywordprg on Windows gVim to look up PHP functions in the manual. I have Python 2.5 installed (batteries included!), so I whipped up this script to work as keywordprg:

import webbrowser
import sys

def main (argv=None, keyword=None):
    if len(argv) < 2 and keyword is None:
        print >>sys.stderr, "You need to send a keyword to look up."
        sys.exit(2)
    elif keyword is None:
        keyword = argv[1]

    webbrowser.open("http://us3.php.net/" + keyword, 2)

if __name__ == '__main__':
    main(sys.argv)

I saved it as C:\Users\sapphirecat\bin\phpsearch.py and then edited C:\Users\sapphirecat\_vimrc to set keywordprg to the Python script. Now the K command in normal mode brings me to the PHP manual page for whatever function the cursor was on.

Second: preventing vim on the server from highlighting HTML and SQL inside PHP strings. This was happening on the server in spite of having let php_sql_query=0 and friends inside ~/.vimrc. It turns out that the default PHP syntax file checks for the existence of the variable, but not its value, so 0 is considered the same as 1. Adding lockvar php_sql_query php_htmlInStrings php_show_preg actually prevents all three of those from becoming defined, and now my PHP strings are really just strings.