Sam Danielson

Google Apps is(are) the new MS Office complete with spreadsheet, slide presenter, database, and word processor. They all suck. Commonly cited reasons for failure are that the browser is not an OS, Javascript lacks support for concurrency, Javascript is slow, businesses are slow to adapt to new technology, community authorship is already solved by VCS and wikis, etc. Here’s the real reason they fail: fighting the last war. Word processors and spreadsheets are generic, where “generic” carries the same negative connotation as when applied to performance art.

Google shouldn’t be trying to make generic apps easier to use on the web. Email, IRC, IM, wikis, forums, and social bookmarking have an evolved interface balancing generality with complexity. They are also not stand alone programs that can have the network part removed like a word processor. Even if all the technical browser/script problems are solved, what will Google Apps offer other than a plethora of features bolted on to something resembling a wiki? It’s generic.

Features can increase generality, but at some level of generality the competing technology becomes Sqlite, Linux, or Java; not MS Word. I’ve witnessed the humor of a Salesforce tech-support conference call. My client was using a form that posted to Salesforce and was trying to add some work-flow features on the Salesforce end. What I heard astounded me, the client was getting a lesson in scripting from the tech. I left the call early, glad my part was limited to creating the form. A solution can go in two directions as it becomes more general; more abstract, or more complex. MS Word and Powerpoint embrace complexity. Programming languages lean toward abstraction. Either way, generic solutions are always harder to use.

The MS Office suite thrived because it made it easy to share. Embrace, extend, extinguish. One app to open them all! The complexity was necessary for generality, and generality was necessary for sharing. On the isolated PC there wasn’t room for a special version of Powerpoint for teaching with another for pitching sales. Word had to handle letters, resumes, and instruction manuals, in formats including HTML, text, and binary. It had to interface with Access and embed Excel documents, and each of those programs had to share well with others in the suite. It was a very capable one-size-fits-all solution that was very complex as a result.

With the web offering effectively unlimited storage on top of easy distribution and sharing, there is no longer a need for one-size-fits-all suites. New apps can depart from the generic view of the suite and approach problems directly. There should not be a Powerpoint of the Web . Instead, applications should specialize so that one for e.g. teaching aircraft systems or training sales reps will bundle a tailored likeness of Powerpoint. Slides in this new hypothetical software might pull facts from manuals or link to simulations, each of which is another specialized application, reiterating the importance of sharing. Thus, The Web

Google Apps solve the problem of sharing by offering a uniform API (so 1990), on the web (so 2000). In contrast the new application suite will have as many APIs as there are applications, and as many applications as there are problems to solve. Google Apps are generic and lame.

My aviation logbook project, Sqlpilot, recently become usable (for me). I was able to turn my old data into CSV with plenty of grep, sed, awk, sql, manual labor, trial and error, and luck. Sqlpilot is a C program linking to gtk and sqlite. Using a relational database feels like the right way to manage a logbook and I now have a whole new view of my flight log. I've been querying things like average leg time and distance for specific aircraft, number of days with over eight hours of flying, number of approaches flown grouped by month, total time in each aircraft and type, and other interesting trivial things that I would have never known had I not started this project.

Read the rest of this entry

Tara and I frequent a Thai Restaurant about a half mile from our apartment. Without too much thought I placed an order for something on the menu called Sator Shrimp since I like shrimp and spice doesn’t bother me. The waiter asked if I had ever had sator before and warned that the dish had a strong smell. That piqued my interest so I stuck with the order. The dish wreaked of odorized propane gas and the flavor of sator defies description. I didn’t know a plant could taste like that even though I ate things like pine needles and radish leaves as a kid. I finished the dish because I was extremely entertained that someone somewhere actually planted sator beans with the intent of selling them as food. Each bite was like the culinary equivalent of the Jerry Springer Show. After the beans were in my stomach and the sensory-masking effect of curry wore off I began to notice the smell again. It was my breath; onions are to sator what tictacs are to onions.

Googling after dinner I found that sator (parkia speciosa) is informally know as a Stink Bean

I can’t even believe a plant like sator is legal to grow in the USA considering growing relatively benign plants like Cannabis can win their owner a prison sentence. We need laws to protect our children from sator now :)

If there was ever a phrase which should be banned from all discourse it is “missing the point”. Missing the point means making an irrelevant conclusion. It’s similar to a non-sequitur with emphasis on the conclusion rather than the evidence. Instead of saying a conclusion is wrong because the evidence is irrelevant (non sequitur – does not follow), “missing the point” directly attacks the conclusion as irrelevant and therefore wrong.

Many Internet discussions have no point and therefore none to miss. Good discussions have many points and that’s why most comment systems are threaded – so different points can be clearly discussed! Don’t tell me I’m missing the point because I introduce a new spin on things. Instead, be a genius and fork the thread. If you disagree then say so, but how could you disagree if I was truly missing the point? If you rebut irrelevance with anything other than an explanation of why something is irrelevant then your clever rebuttal is non-sequitur since the chain of logic is broken. This is what downmods are for. Use them. “You’re missing the point”, is just a red flag that means irrelevant, obvious, or arrogant statements follow.

Which brings me to arrogance. By claiming that I am missing the point you have implied that there is some “one true” point that I have missed. How is that even possible? Does each cause have one and only one effect, and vice versa? It’s nearly the definition of simple-mindedness. Like thought police you seek to limit the scope of discussion to protect the logical purity of your narrow, self-idolizing views though mindlessly repeated idioms. It’s doubleplus ungood for real discussion.

Think I’m missing the point or full of shit? Consider that Google claims over three times the number of hits for missing the point then for the overused phrases full of shit and full of crap combined. Please stop polluting discussions with this trash.

As part of my effort to seek enlightenment through Haskell I’ve made some music. It’s actually sort of neat creating a waveform that when played through a DA resembles something other than white noise. This isn’t controlling a sequencer – it’s just composing functions to make make a sound representation. It’s a lot like naive ray tracing where all objects are considered for each pixel in that for each sample I sum the waveforms for all notes, many of which are zero since a note only plays for a short duration and is otherwise silent.

Notes are made by transforming a sound frequency Signal (Double -> Double) with pitch and amplitude Shifters (Signal -> Signal). For the note envelope I create a signal that represents amplitude and modulate the sound signal with the amp combinator, which is a Modulator (Signal -> Signal -> Signal).

One thing missing from this implementation is the ability to write (signal = signal1 + signal2). I now would write (signal t = signal1 t + signal2 t,) since a signal is a function over time. mauke in #haskell (irc.freenode.net) was helpful but I still don’t understand why some things work the way they do so I chose not to write a (+) instance for signals until I understand it better. The following will play a few measures of Mozart’s something something as I remember it. I forgot what it’s called but it’s in C and it’s popular.

Here’s an ogg vorbis of the output: mozart.ogg

And the illiterate Haskell.

-- Compile and listen as follows
-- ghc -o mozart mozart.hs && ./mozart | mplayer -v -rawaudio rate=16000:channels=1:samplesize=1 -demuxer rawaudio -

import System.IO
import Data.Char

type Signal = Double -> Double
type Shifter = Signal -> Signal
type Modulator = Signal -> Signal -> Signal

main :: IO ()
main = putStr . toBytes . take 200000 . sample 16000 $ sound

sound :: Signal
sound = ampc 0.35 . shifterSum song . pitchc (440 * hstep ** 3 * 2 * pi) $ wave

song :: [Shifter]
song = zipWith (\e p -> amp e . pitchc p) (lhEnvs ++ rhEnvs) (lhPitches ++ rhPitches)

wave :: Signal
wave = flange (\x -> sin (x * 2)) sin

lhPitches :: [Double]
lhPitches = map (* (1/2)) $ [du,so,me,so,du,so,me,so,re,so,fa,so,du,so,me,so] ++
                            [du,la,fa,la,du,so,me,so,re,so,fa,so,du,so,me,so]
lhBegins :: [Double]
lhBegins = take 32 $ toTempo [1..]

-- Applies 0.1 second duration to left hand begin times
-- resulting in a list of complete amplitude envelopes
-- for each note.
lhEnvs :: [Signal]
lhEnvs = map (pianoEnv 0.1) lhBegins

rhPitches :: [Double]
rhPitches = [du,me,so,ti/2,du,re,du,  la,so,du*2,so,fa,me]
rhBegins :: [Double]
rhBegins = toTempo [1,5,7,9,12,12.5,13, 17,21,23,25,28,29]
rhEnvs :: [Signal]
rhEnvs = map (pianoEnv 0.5) rhBegins

-- attack decay sustain release to resemble a piano
-- apply duration and begin for complete envelope signal
pianoEnv :: Double -> Double -> Signal
pianoEnv = env 0.005 0.05 0.4 0.7

-- eigth note tempo
tempo = 160
toTempo :: [Double] -> [Double]
toTempo ts = map (\t -> 60 * t / tempo) ts

-- Sums list of signal shifters and mutes the original signal
shifterSum :: [Shifter] -> Shifter
shifterSum xs = foldr (shifterAdd) (const . const 0) xs

shifterAdd :: Shifter -> Shifter -> Shifter
shifterAdd a b input t = a input t + b input t

-- 1        b
--         / \c_____d
--        /          \
-- 0 ___a/            \e___   Makes an amplitude envelope for a note
--
-- Will div by zero in some cases.
env :: Double -> Double -> Double -> Double -> Double -> Double -> Signal
env attack decay sustain release duration begin t
    | t < a || t >= e = 0
    | t < b           = (t - a) / attack
    | t < c           = 1 - (1 - sustain) * (t - b) / decay
    | t < d           = sustain
    | t < e           = sustain - sustain * (t - d) / release
    where a = begin
          b = a + attack
          c = b + decay
          d = c + duration
          e = d + release

sample :: Double -> Signal -> [Double]
sample rate signal = [signal (t / rate) | t <- [0..]]

-- yuck
toBytes :: [Double] -> [Char]
toBytes = fmap $ chr . clamp 0 255 . (\n -> truncate $ (n + 1) * 127)

clamp min max x
    | x < min = min
    | x > max = max
    | otherwise = x

pitchc :: Double -> Signal -> Signal
pitchc mult input t = input (t * mult)

ampc :: Double -> Signal -> Signal
ampc mult input t = mult * (input t)

amp :: Modulator
amp control input t = (control t) * (input t)

-- only works for control where d/dt = 0
pitch :: Modulator
pitch control input t = input (t * control t)

flange :: Modulator
flange ff w t = w (t + ff t)

hstep = 2 ** (1/12)
du = hstep ** 0
re = hstep ** 2
me = hstep ** 4
fa = hstep ** 5
so = hstep ** 7
la = hstep ** 9
ti = hstep ** 11

Tara, my wife, has been maintaining her personal blog primarily with php for templating and no database. Recently she had the desire to add RSS feeds but since her pages and articles consited of only php files we would need to either scrape the feed from the pages or start using some real blog software. I have been using mephisto so I set that up for her. After we both had done considerable work on templating and transfering content she decided to just retemplate her php setup and worry about RSS later. I think it was a sound decision. After all, editing in the browser sucks. She’s been using vim but now prefers the aquamacs style of emacs since I introduced her to that. What can I say? A blog should be nothing more than a directory full of files that gets scanned into an index for display via html or rss. The metadata in liunx is a bit sparse since there is no creation date on files – just accessed and modified time. Publishing, tagging, and categorization could be handled through symlinks. Search could be a done via grep and cgi but I’m a little worried about escaping the search string correctly. Comments can be an entirely seperate system for all I care. If only linux wasn’t so tantilizingly close to a multiuser CMS/Blog. It’s close – but no cigar.

On January 7, 2008 I had my appendix removed more than six weeks after initial symptoms.

I began feeling fatigued during a trip on Nov. 22nd. With symptoms of fever, extreme fatigue, and abdominal pain I called in sick and went home on the 24th. On the 26th I went to a primary care provider and was diagnosed with hematuria caused by a possible kidney infection or stone. I left with a prescription for seven days of steroids and ten days of antibiotics. Ten days later I still felt ill so I returned to the clinic where I again tested positive for hematuria and got a referral to a urologist. The urologist referred me to a CT scan where finally, on the 10th of December, I found the cause of my sickness to be an enlarged appendix. The urologist referred me back to primary care where I received a referral to a surgeon. The surgeon reviewed the CT scan and decided to remove the appendix but advised that since my condition was improving it would be best to allow some healing to occur before cutting. He prescribed some antibiotics and said it would be okay to work but agreed it would be best to consult an aviation medical examiner. The aviation doctor also said that I could work until surgery but cautioned me that I should not work with pain since it was a distraction. Since I was pain free for three days I returned to work on the 17th of December but the altitude changes and turbulence caused my abdominal pain to return. As should be expected the aviation doctor was right, pain is really distracting. I called in sick again and decided to wait until I had recovered from surgery to return to work. The surgeon did a great job and I have had less pain during recovery than I did during much of December. I don’t need pain medication but I have been using one dose at night to sleep through the minor discomfort of three incisions in my stomach. Finally, almost two months after beginning sick leave I will be fit to return to work. I have used the majority of my sick bank and filed for FMLA. What should have caused me to miss three weeks of work turned in to seven and a half. The hospital staff, surgeon, urologist, and CT scan folks were excellent and I have great respect for them and their profession.

As for so-called primary care, this is not the health care I grew up with and I don’t really know where to point my finger. The primary care PAC and her doctor focused on kidney issues due to hematuria and assumed the abdominal pain was associated. Blood and urine were taken for additional testing but the blood results took days to complete. Blood tests confirmed I had an infection of some sort. There was no X-ray, CT, ultrasound, MRI, or any imaging in the office. Getting imaging done takes a referral and an appointment several days away. A CT scan would have removed the possibility of a kidney infection or stone. I can only assume that if there was imaging equipment in the office I would have been referred to a surgeon two weeks earlier than I was, maybe even had my appendix removed that same day. Since it takes three to five days to schedule a CT scan it would seem reasonable to begin taking drugs for the most likely problem until the scan could be completed. Even then, I would most likely need a specialist referral and that can take even longer to schedule than a scan. Why can’t I go to the doctor and get some real tests done so I can get a diagnosis? Do I blame the system or just assume I chose bad primary care? This isn’t the first clinic I have found in the area. I have been to a few and shopped around. They are all in professional buildings and are staffed by approximately ten doctors and nurses. They have a microscope and a stack of magazines. Any real tests require a referral. I personally think primary care has become nothing more than a sales outlet for pharmaceuticals.

While at the hospital I told the nurse about my referral saga and asked if she knew of a primary care clinic that had at least an X-ray machine. She advised that she was not allowed to give recommendations. Fair enough, but she also said the care I received seemed standard. Usually primary care, the place I go with symptoms, is a professional building with very limited equipment. What are these places? All they can do is prescribe drugs and when I’m sick I may need more than drugs. The hospital is where the real care is but unfortunately I don’t qualify for that clinic since I actually have insurance and I’m not a senior citizen. In the end I paid more in co-pays than an emergency room would have cost. My profession as an airline pilot is also incompatible with referrals and waiting since I am essentially grounded until I know what’s wrong. I can’t just go to work while I wait for test results for documented symptoms. I am required to have a medical exam every six months and it is clear that going to work sick is illegal, stupid, and a willful violation of the law. Despite the cost I am considering skipping the bullshit and going to a real hospital with real equipment next time I have symptoms.

This leads to billing. I have great insurance. It’s a 90/10 plan with no deductible and a reasonable out-of-pocket maximum. I chose this plan over 100% coverage because even with my surgery the total cost in premiums and coinsurance is about equal to the cost of 100% coverage. These are the benefits of working for a good company and being part of a strong union like ALPA. I lost no pay since I had time in my sick bank and FMLA will keep me in good standing with my employer. This is to say, I am not forgetting to count my blessings.

I do, however, have some issues with the billing which make me worry for the day when I have no insurance. A typical test for blood or urine might be billed at six times what the insurance company actually pays. A $380 lab test gets covered at $60 by BCBS. When I asked how much my CT scan would cost I was told $2000. They billed BCBS for $850 and BCBS paid around $500. My bills all share this trait with lab work having by far the greatest discrepancy between what is billed vs. paid. Before the surgery I tried to find out what it would cost but no one could tell me. I signed that I would pay the bill if the insurance company denied my claim without a clue what it would cost. I know it depends on the resources and time used. If I had to stay additional nights in the hospital or have more surgery and blood transfusions it would cost more. Still, why can’t I get an itemized estimate ahead of time? An accurate estimate would be nice. Don’t tell me it’s going to cost $30,000 and then bill the insurance for $6500 while getting paid $4000. This is exactly what would have happened so I didn’t really mind not having an itemized bill.

I have it better than most people in America. Many folks pay larger premiums and have per-person deductibles which for a family can lead to terrible cost. Others have no insurance. Do they pay the full 600% markup on lab work? Primary care is a WTF. Billing is a WTF. At least I am alive. Thanks also to my wife for helping me with the innumerable pains and annoyances that happen during illness.

Write yourself a Scheme in 48 Hours

This is the most excited I’ve been since I learned how to make a character bounce around the screen on my C64! Learning a new programming style is hard: big programs are difficult to understand and small programs have too little to tweak. How do I learn from a ten line tutorial or a one-hundred line program? The ten line tutorial doesn’t do anything interesting and forces me to write new code to grow. OTOH, one-hundred lines of Haskell usually do something neat but I can’t begin to untangle how. For everyone there is a threshold of complexity, the understanding of which enables creating more complex things. Once this threshold is achieved a person can bootstrap their learning as an autodidact. Below it there is neither enough knowledge to guide their study or test their knowledge for correctness.

My capacity for understanding is often far lower than the bootstrapping threshold and my learning stalls as I fumble through tutorials gaining one small bit of knowledge without correlation until one day things begin to click into place. I then wonder how I learned in eight hours what had stumped me for eight months and feel profound appreciation for the wizards that understood it first. It’s like I was blind and they could see, and now through some miracle I also see. It’s real sight too, not just a description in words but the real sunrise seen for the first time by the formerly blind. In this case the misfitting pieces were the static algebraic types, monads, type classes, currying, and lazy evaluation of Haskell. It’s quite a bit to swallow just as the more familiar loops, conditional statements, functions, and variables once were. Many of these things are interdependent and cannot be understood separately. Jonathan Tang begins small and works from the ground up to a larger program, explaining each change so every line’s purpose is thoroughly understood. The culmination is a very small but capable scheme interpreter that is larger than what I would have understood without my building it. It is a perfect demonstration of the building-block technique of instruction I learned and used as a flight instructor.

Mr. Tang put the pieces together concisely. Wow.

Today as my file dialog opened to save a freshly downloaded pdf it defaulted to my $HOME/Documents directory. Why? I have never created a Documents directory and I certainly don't use it. I didn't even know it existed! Where did it come from? Where did Music, Pictures, and Videos come from? I know I deleted them from the default skeleton for Ubuntu and I'm sure the Arch installer didn't put them there. Who decided I needed these things and that I should use them? They're capitalized, don't begin with a '.', and are empty. They are a stupid suggestion and an eyesore. Should I really put all my videos in the same subtree? I don't mind .xinitrc, mail, .bash_history, Desktop, News, etc... These files have a purpose. Documents, Music, Pictures, and Videos are totally useless and empty. Thank you for demonstrating that directories exist. Thank you for causing my file dialog to conveniently default to Documents. If this were Windows I'd have to add a gracious Thank You!!! to all software for assuming that I save everything in My Documents, and for that matter, that everything I save is a Document! I don't even know who I'm thanking. Who's idea was this anyway? Maybe it's my fault for assuming that words have meaning or for not keeping my underwear-drawer perfectly labeled and sorted. Whoever you are please stop creating useless directories in $HOME for the sole purpose of imposing your world view. If I ever get the chance to write to your $HOME I'll create a directory called Idiot's Files and move everything else into that.

Xmonad is an awesome window manager and these are my complements as a fanatical user. The thing I love about Xmonad is that it just seems to get everything right. Wmii, my first favorite tiling WM, really got me interested and I found no fault with it until I used Xmonad.

Internally Xmonad manages focus with a zipper data structure. Though there are more naive approaches used in early versions it’s clear from maintainer Don Stewart’s blog that a lot of thought went in to my sanity as a user. The sorted order provided by the zipper allows for completely automatic management of my windows. For me, the order reduces my workspace to essentially one dimension. In Wmii I used vim key-bindings to move focus up/down/left/right. In Xmonad I use only two keys for this. In every other window manager Alt+Tab is a total pain due to the random order of the windows or the fact that they need to be manually sorted. Xmonad always inserts and deletes windows in a way which preserves order as if I were inserting or deleting lines from a text file. Better, Xmonad can be extended with, among other things, algorithms to arrange windows based on this order. There are tabbed, grid, and circular layouts to name a few. One interesting extension even arranges windows in a Fibonacci spiral.

There’s plenty of fud floating around about the inextensibility of purely functional programs but Xmonad proves otherwise, at least for Haskell’s case. I feel like a dog watching TV when viewing the source but I do know this; the extensions manage a variety of state and the code is concise. Oh, and the whole thing runs in a memory image a little larger than an instance of bash.

Special thanks to Spencer Janssen for authoring the project. Xmonad is doing everything just right so my Archlinux machine is polished like a Macintosh, only more useful.

Yes! Arch is great. For more detail read on.

Arch's Appeal

The main difference between Arch and Unbuntu is that Arch is designed to be molded whereas Ubuntu is a complete desktop/server distro. Like Ubuntu, most things were configured and worked upon install. Unlike Ubuntu, Arch does not have a default desktop environment although XFCE4 could be considered well supported. Arch will appeal to users who wish to start with a minimal system and a package manager, adding bit by bit until the system is tweaked to their desire.

Ubuntu's Appeal

Ubuntu is for those who just don't want to mess with it - giving users the choice of a server, or a Gnome / KDE / XFCE desktop. There is also a minimal (alternate) Ubuntu installer for those who want more control. I wont say Ubuntu 'wins' here or anywhere because it's apples and oranges. Yes - Arch vs. Ubuntu is a stupid comparison but if you're like me and you can't help but type X vs. Y into Google before doing anything then maybe identifying some differences isn't so bad. In short, Arch maintains bleeding-edge, occasionally broken, packages with no release cycle whereas Ubuntu needs upgrading (breaking) every six months. Ubuntu is good software with target markets. Arch is good software.

Package Management

Arch uses a rolling release cycle where major releases consist of a snapshot. For a personal computer, avoiding Ubuntu's six month upgrade cycle is a definite benefit. On the other hand it could cause breakage occasionally, which is why I still prefer Ubuntu (LTS) for my server. Arch packages tend to be more bleeding-edge from what I can tell browsing the repository. For example, the packaged version of GHC in Ubuntu 7.10 is 6.6.1 while Arch will install 6.8.1.

Pacman, the Arch package manager, is similar to apt. One thing I really like is that headers are bundled with library binaries. This makes sense since headers have a tight coupling to the code they describe. They can also be used as documentation when man pages are weak. With Ubuntu I had the option of not installing headers but that proved to be sort of a PITA when compiling or building a new development system. Ruby gems also uses gcc occasionally and it's nice to avoid the search for *-dev packages. pacman -S imagemagick && gem install rmagick just worked.

Post Install

Everything is installed, configured, and working as per the normal Ubuntu setup. Arch requires a few extra steps to get a desktop environment installed. Here's a few examples.

Xorg

Unfortunately I had to configure X. I say, 'unfortunately', because I hate configuring X. Best to just install Ubuntu on another partition and copy the xorg.conf. Currently I'm using the open-source ATI drivers since I couldn't get the proprietary ones to install properly. I'll be working on that later.

Sound

Running alsaconf asked a few questions and my sound card worked. Easy.

Suspend and Resume

Suspend and resume with swap and ram work great using the pm-utils package. I had to configure the resume scripts to run alsactl restore to get sound working after a restore but other than that it was configured perfectly upon install.

CPU Frequency Scaling

Ubuntu Forums had a great post on this topic. Since it's dealing with the kernel it works on Arch too. http://ubuntuforums.org/showthread.php?t=248867

In Total

I'll be using Arch for the foreseeable future, maybe even contributing if I'm ever capable of fixing something. The documentation on their wiki is outstanding and the community has momentum. In general, Ubuntu is great for the automated installation and configuration of everything to include the kitchen sink. Arch is great for a lightweight and fast system with a rolling release cycle and bleeding-edge software.

These steps worked on Ubuntu Feisty on a 32 bit system and an Ipod Classic.

The packaged version of ffmpeg on Ubuntu Feisty wasn't doing the job, giving me this error while trying to convert an mpeg 2 to an mpeg 4: Unsupported codec for output stream #0.1 Newer versions of Ubuntu may fix this so try your luck with the following command.

sudo aptitude update
sudo aptitude install ffmpeg
ffmpeg -i inputfile.mpg -b 300 -qmin 3 -qmax 5 -g 300 -ab 96000 -s 320:240 -aspect 4:3 test.mp4

If this doesn't work ffmpeg will need to be compiled locally with the following instructions.

1. Uninstall ffmpeg

sudo aptitude remove --purge ffmpeg

2. Get the latest ffmpeg source.

An svn checkout is also available at http://ffmpeg.mplayerhq.hu/download.html

wget http://ffmpeg.mplayerhq.hu/ffmpeg-export-snapshot.tar.bz2
tar -xvjf ffmpeg-export-snapshot.tar.bz2
cd ffmpeg-export-???

3. Install compile dependencies.

sudo aptitude install libfaac-dev liba52-dev libdc1394-dev libgsm1-dev libogg-dev libvorbis-dev libxvidcore-dev

4. Configure the build, make, and install.

./configure --enable-libxvid --enable-libfaac --enable-gpl --enable-pp --enable-pthreads --enable-libvorbis --enable-libogg --enable-liba52 --enable-libgsm --enable-libdc1394 --disable-debug --enable-shared
make
sudo make install

5. Edit shared library configuration.

sudo cp /etc/ld.so.conf /etc/so.conf.bak
echo "echo /usr/local/lib >> /etc/ld.so.conf" | sudo sh

6. Update library simlinks.

sudo ldconfig -v

7. Make sure bin paths are configured correctly.

which ffmpeg

This should output /usr/local/bin/ffmpeg. If not, restart the shell and check the PATH variable. Ensure /usr/local/bin comes before /usr/bin and /bin.

8. Test it out somewhere.

ffmpeg -i inputfile.mpg -b 300 -qmin 3 -qmax 5 -g 300 -ab 96000 -s 320:240 -aspect 4:3 test.mp4
file test.mp4

That's it. The remaining is icing on the cake.

9. Automate it.

Make was designed for automating software builds. The advantage in using it to automate file conversion is that it will recognize when no work needs to be done for an output file. Create a directory called ipod which is a sibling of the mythv recordings directory. Place the following make-rule inside a Makefile in that directory.

File: recordings/../ipod/Makefile

%.mp4 : ../recordings/%.mpg
        ffmpeg -i $< -b 300 -qmin 3 -qmax 5 -g 300 -ab 96000 -s 320:240 -aspect 4:3 $@

Create a script to tell Make which output files to create. In this case it's the basename of each MythTV mpeg with an mp4 extension. Make will use the above rule to figure out how to create an mp4 file with a given name. Make will do nothing if the file already exists and is newer than the input file. Name the script convert and chmod it to 755.

File: recordings/../ipod/convert

#!/bin/sh

PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin 
export PATH

make `ls ../recordings/*.mpg | sed 's/\.\.\/recordings\///' | sed 's/.mpg$/.mp4/'`

The ugly recording filenames consist of two parts. 1005_20071004210000.mpg, for exampe. The 1005 is the chanid column in the 'channels' table and the rest is a date and time. Luckily the information about a recording is in the 'recorded' table so that can be used to create pretty names for the new mp4's. Here's a my solution. TIMTOWDI, right?

File: recordings/../ipod/make-pretty

#!/bin/sh

PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin 
export PATH

mkdir -p pretty

for filename in `ls *.mp4`; do
        basename=`echo $filename | sed 's/.mp4$/.mpg/'`
        prettyname=`mysql -umythtv -p???????? -e "select concat(title, ' ', progstart) as name from recorded where basename = '$basename' limit 1;" --database mythconverg -NE \
        | grep name: | sed 's/name: //' | sed 's/00$//' | sed 's/://g' | sed 's/$/.mp4/'`
        ln -sf "../$filename" "pretty/$prettyname"
done

Add the following line to /etc/crontab or create a new file called /etc/cron.d/ipod and place it there instead. Be sure to sudo chown -R the ipod directory to the same user specified in the crontab. Every morning the ipod/pretty directory will be filled with shows to watch on the morning commute.

File: /etc/cron.d/ipod

05 2            * * *   mythtv          cd ???????? && nice -n19 ./convert && ./make-pretty

Toying with the Nokia 770 has been teaching me more than I ever wanted to know about C, gcc, and the other things we do to write software for handhelds. I had known about the existence of garbage collectors for C and C++ but had always assumed they worked similar to GObject. A GObject programmer must call g_object_ref() and g_object_unref() to manipulate the reference count of each object on the heap. When the count reaches zero the space is reclaimed. A Boehm GC, rather, can function as a drop-in replacement for malloc() and free(). To my amazement the free() procedure isn’t required at all and can be replaced with a noop. The word “automagically” doesn’t even annoy me in this case since the Boehm GC is quite a trick. It can determine when an object on the heap is no longer required by comparing places where pointers usually live with it’s internal book. For example, a function that calls malloc usually sets a pointer on the stack to the allocated memory. When the stack shrinks the reference is gone and the dead memory can be automatically freed.

The main benefit, for me anyway, is that I can write consing functions in C. Any linked list implementation has a foreach method. Foreach is easy because it doesn’t have to keep the return value of its lambda argument. Map, filter, and almost every other good thing that comes from functional programming requires consing up a new list. Allocating the right amount of memory for each return value is doable. The difficult part is the bookeeping required to later free all the unreferenced lists. A functional program might map, filter, then reduce in a single line as it takes the GC for granted. Manual allocation is a better fit for building fewer, less disposable structures. I’m not sure I’ll use map that much with C but it would be a good way to test-drive the Boehm GC.

There has to be a catch. I’m still wide eyed that the Boehm GC even works. It might be a sign that Mozilla and X, the two biggest hogs on my system, use it.

This is awesome: aduni.org

There are twelve courses complete with video lectures, handouts, problems, and exams. So far I’ve had time to view the first six lectures in Theory of Computation covering regular grammars, finite state machines, context-free grammars, pushdown machines, Chomsky Normal Form, and a whole bunch of things I’ve forgotten. I can’t yet wield lex or yacc like a pro but understanding some of the rigor behind scanning and parsing really opens the door to some neat things. Other courses cover discrete math, probability, networks, algorithms, memory, pipelining, object orientation, web programming, artificial intelligence, and of course the “Structure and Interpretation of Computer Programs”.

Shai Simonson is an excellent lecturer. In a way it’s like watching Jeopardy but instead of trying to guess the correct answer-question, I try to guess what Shai will write next on the blackboard. I’m sure the other courses are just as interesting but I’ve only skimmed their contents as of yet. The materials are licensed under Creative Commons.

It is often stated that it is a good idea to exclude code or functionality from templates. I have seen more than a few articles on the web entirely about how to approach this goal and the solutions are usually somewhat ad-hoc or arbitrarily restrictive. Some projects even seek to build a template language on top of the base language. We are typically left with the impression that there are no rules, or even good guidelines, by which to judge the amount of functionality performed by the template. Code seems like a necessary evil. We need it, but too much is bad and how much is a black art. Here is a simpler definition of a template: templates are pure functions.

A pure function obeys the following rules.

1. The function always evaluates the same result value given the same argument value(s). The function result value cannot depend on any hidden information or state that may change as program execution proceeds, nor can it depend on any external input from I/O devices. This is known as referential transparency.
2. Evaluation of the result does not cause any semantically observable side effect or output, such as mutation of mutable objects or output to I/O devices.

The singular purpose of a template is to transform it’s input into something viewable. If output does not depend entirely on input then the template is doing something it shouldn’t. This is boring and obvious to experienced functional programmers but, hey, we’re not all experienced functional programmers. Despite the theoretical simplicity there’s always a catch in the real world.

Automatic loading of composed objects muddies the pure function perspective. In this contrived example there is a user model composed of an address model with a street attribute e.g. user.address.street. The common practice, at least in rails, is to build a template which accepts the user model as input and chains to attributes of associated models. Does the output of a template using object chaining depend only on input? That depends on whether or not the requisite models were loaded before evaluating the template. If template evaluation caused a database hit then the output would partially depend on what was in the database at that time, breaking the pureness of the template. Automatic object loading is a hidden complexity that speeds development. Associations should be preloaded during production with a database join to prevent superfluous queries, a practice consistent with templates as pure functions.

There is no need to create a non Turing-complete subset of a language to build a proper template, nor is there a need to eliminate code from a template. The mantra that code doesn’t belong in templates is harmful and wrong. Templates may contain any code so long as the template remains a pure function.