Today I've found a way to cabal-install offline documentation – with source hyperlinks! – for a Haskell package.

(The reddit link was quite helpful by pushing me in right direction.)

To obtain documentation without source hyperlinks you just

• run `cabal install' with `--enable-documentation' option ...or
• enable `documentation' parameter in ~/.cabal/config

Then reinstall the package with `cabal install --reinstall' command. Documentation will appear in ~/.cabal/share/doc/PACKAGE-VERSION/html/ directory.

But if you want proper documentation – with `Source' links to the right of declarations – try these commands (replacing `iteratee' with your favourite package):

$ cd /tmp
$ tar -xzf ~/.cabal/packages/hackage.haskell.org/iteratee/0.2.4/iteratee-0.2.4.tar.gz
$ cd iteratee-0.2.4
$ cabal configure
$ cabal haddock --hyperlink-source

## NOTE:
##  * Make sure `documentation' is *disabled* in ~/.cabal/config.
##  * Don't use `--enable-documentation' in `cabal install' command.
## Otherwise links to source code will be omitted from documentation.

$ cabal install --reinstall

That's almost it. But source code pages will not be syntax-highlighted, because hscolour.css is missing. I've used a workaround:

$ cd ~/.cabal/share/doc/iteratee-0.2.4/html/src
$ ln -s /usr/share/doc/ghc6-doc/libraries/Cabal/src/hscolour.css

Now everything is sweet (i.e., documented and hyperlinked to sources).

This post is a reminder to myself, which may possibly help some other haskeller.

SEE ALSO

• Cabal ticket #534: cabal haddock should --hyperlink-source by default
• Cabal ticket #517: make cabal install command support haddock options
  (Opened 8 months ago. Patch available.)
• Debian bug #500924: Should have a system-wide hscolour.css

(See part 1 and part 2.)

Motorola SLVR7e L7e phone has FAT16 filesystem on its memory card. Using wikipedia article on FAT16 design we will locate .mp3 entries in the filesystem and try to correlate them with .pla file (the playlist).

FAT16 consists of boot sector (1 sector is 512 bytes), two File Allocation Tables (FATs), root directory, and data region. Boot sector specifies various parameters of the filesystem. The ones of our interest are:

• [number of] sectors per cluster,
• sectors per FAT,
• maximum number of root directory entries.

We will use this helper function to read values from the filesystem:

read_at() {
    case "$2" in
        1) t='C';;
        2) t='v';;
        *) echo 'Usage: read_at OFFSET SIZE' >&2; return 1;;
    esac
    dd if=/dev/sdb1 bs=1 skip=$1 count=$2 2>/dev/null | \
        perl -wspe '$_ = unpack("'$t'") . "\n"'
}

And these are the boot sector values we need:

$ read_at $((0xd)) 1   # sectors_per_cluster
32
$ read_at $((0x16)) 2  # sectors_per_FAT
239
$ read_at $((0x11)) 2  # max_root_entry
512

To be finished...

Итак, проблема

Тормозит отрисовка окон: scroll и при перетягивании мышью. (Изначально "проблем с Иксами" было больше, но я вам про них не расскажу, чтоб не сбить нарратив.)

Решение

Поставить драйвер NVIDIA.

Реализация

...Затрудняется тем, что драйвер из пакета `nvidia-kernel-source' не собирается для ядер старше 2.6.26. Я его пропатчил, как мог (чтобы собрался с двумя директориями linux-headers-*-686 и linux-headers-*-common, и свежим API), но Xserver отказался с запускаться с тем, что получилось.

Решение

На работе времени мало. Значит, надо принести туда исправный ноутбук из дому, а глючную Тошибу взять на выходные на "debug с пристрастием".

...Пятничным утром, благородно дав мне возможность ввести в строй замену (перебросить почту, всё такое), ноутбук-с-тормозящими-исками перестал реагировать на клавиатуру. За что был выключен кнопкой (в последствии — неоднократно).

Новые симптомы

Xserver перестаёт реагировать на клавиатуру. Мышка лазит, но толку с неё. "Подвисаем" по-разному: один раз не смог даже username вбить при входе в сессию, другой раз только успело xterm нарисовать; обычно — вскоре после начала работы.

Отладка

Есть два полезных документа: ServerDebugging на Xorg Wiki и Xserver Debugging Hints. Они убедили меня поставить пакет с отладочной информацией, `xserver-xorg-core-dbg'.

После этого оставалось:

• дождаться зависания иксов,
• зайти с другого компьютера по ssh,
• gdb /usr/lib/debug/usr/bin/Xorg `pgrep X`,
• увидеть бесконечный цикл внутри модуля `nv' (пакет `xserver-xorg-video-nv').

Победа

Исправляем в /etc/X11/xorg.conf драйвер с "nv" на "vesa". Рестартуем xdm (всё в том же ssh). И прекрасно работаем – стабильно и достаточно быстро – с vesa драйвером.

Резюме

Посылаю драйвера nv и nvidia в виртуальную задницу. (Сидеть им там не пересидеть, в 3D игры я не играю.)

Из полезного: настроил русский язык в текстовой консоли. (Всего-то и надо `console-cyrillic' поставить.)

Пост получился ослепительно нудным, но я предупреждал.  Удачи!

$ nc -w5 -vz haskell.hackage.org
haskell.hackage.org: forward host lookup failed: Host name lookup failure

Let's tune to #haskell IRC channel...

С лицом, измученным и серым,
На белой смятой простыне,
Как жертва бешеной холеры,
Лежит коленками к стене.
Протяжно стонет, как при родах.
Трясется градусник в руках.
Вся вековая боль народов
Застыла в суженных зрачках.
По волевому подбородку
Струится за слезой слеза.
Он шепчет, жалобно и робко:
"Как ты с детьми теперь – одна?.."
В квартире стихли разговоры
Ночник горит едва-едва.
Темно. Опущены все шторы.
У мужа – тридцать семь и два...

Это про меня.

(Списал отсюда. Автор неизвестен.)

Problem

Arrow keys don't work in dosbox-0.72 .

Details

`dosbox -startmapper' shows that `left' key has become `right alt', `up' — `print screen'; `down' and `right' keys are not recognized.

Solution

Unset `usescancodes' in your ~/.dosboxrc.

$ cat ~/.dosboxrc
[sdl]
usescancodes=false

NOTE: This bug may be fixed in 0.73 (there is evdev mentioned in the changelog).

See also

Debian bug #549611

В C++ не хватает степеней "публичности" для элементов классов. В [больном на голову] коде часто встречается такое:

class Foo : public Bar {
public:
  enum One { one = 1 };
  enum Two { two = 2 };

public:
  Foo(int n);
  virtual ~Foo(); // the empty body is defined elsewere

public:
  int val;

protected:
  enum Three { three = /* a lot */ one + two };
};

C+++ должен понимать, что перечисления One и Two — публичные, конструктор и деструктор — очень публичные, а переменная(-член) — совсем публичная(-чный).

``How to Write More Clearly, Think More Clearly, and Learn Complex Material More Easily.''
Wonderful presentation by Michael A. Covington.

I like it up to the last slide (especially the last slide).

[ via Chung-chieh Shan's blog entry ]

(See part 1.)

As long as .pla file is binary, we will view it in hex. Here's a hex dump of a playlist containing first three tracks of Led Zeppelin IV:

$ hd lz4.pla
00000000  00 03 00 01 00 00 00 00  01 40 00 04 00 00 3d 60  |.........@....=`|
00000010  01 40 00 0e 00 00 3d 60  01 40 00 02 00 00 3d 5f  |.@....=`.@....=_|
00000020

This playlist is named `lz4.pla'. We will refer to it below.

Header

I've created several playlists "manually" and noticed that the second byte is always equal to the number of tracks in playlist (`03' in lz4.pla). Bytes with offsets 0 and 2–7 are the same for all playlists I've seen. Thus we can guess the format of .pla file header:

00 <nn> 00 01 00 00 00 00,

where <nn> is number of tracks in playlist.

Entry

Playlist header is followed by entries — 8-byte-long track specifications. There are as many entries as there are tracks in a playlist (lz4.pla file has three: address ranges 0x8–0xf, 0x10–0x17, and 0x18–0x1f).

When we open playlist in Motorola, we see the names of tracks included. But there is no chance for 8 bytes of .pla entry to contain the name of track: the third track is displayed as `03 - The Battle of Evermore' making a string of 27 characters.

Thus, we can assume that .pla entry encodes position of .mp3 file within the underlying filesystem.

Continued...

I've got Motorola SLVR L7e mobile phone. The device is capable of playing MP3 audio files, and I use this feature occasionally.

Once a musical album is uploaded to phone's microSD, it is better to create playlist, otherwise playback will stop as soon as the first track finishes.

And while uploading albums is very simple (you just copy directories like you do with ordinary USB memory stick), creating playlists on L7e is pain.

Continued...

В эту субботу LtU-Kiev собирается, вы в курсе?

Что такое LtU-Kiev и как оно началось?

The main thing uniting us is our interest in programming languages...

Try this command in your xterm (don't worry, it's harmless):

(trap 'xrandr -o normal' 0; xrandr -o inverted; sleep 20)

(-:

[Inspired by `Changing X11 resolution on the fly' article.]

(Re: Daily WTF to Ultimate Refactoring?)

Чувствую себя лейтенантом Рипли, которая забежала с зажжённым огнемётом в склизкую колонию "чужих" с благородной целью (сделать на этом месте солнечную лужайку с бабочками), но постепенно начала осознавать масштабы.  Подлая "матка" загадила яйцами невообразимые площадя...

It ain't easy to profile 2.6K SLOC program made of 3 functions.

PS: I've narrowed the problem: 99.1% time is spent in one of these functions; just 1000 lines of code.

Using Debian's required package, you can learn a lot about French cheeses:

  $ aptitude changelog login | awk -F\" '/The ".+" release/ {print $2}'
  Tome des Bauges
  Chevrotin
  Banonet
  Banon
  Chambérat
  Le Puant Macéré
  Rocamadour
  Morbier
  Bleu des Causses
  Bleu d'Auvergne
  ...

gcov - coverage testing tool

( sample output )

GCC only.

Update: gcov with `-b' option measures branch coverage.

See also:

• SQLite: A Lesson In Low-Defect Software blog post
• the presentation it mentions
• and the corresponding reddit discussion

Well, the promised explanations.

The screenshot shows work-log.el debugging session. The "a-a-i-i-d" patchwork you see there is a work log model (more on this below). A work log is a sequence of dates, active entries, and inactive entries (the latter include comments); see example here. In short: resolved tasks == inactive, pending ones == active, that's it.

Now. `work-log-hide-inactive' function (`C-c C-h') lets user concentrate on open tasks, making inactive "noise" invisible.

The first implementation of the hiding function did hide resolved tasks but also left some dates with no entries at all hanging void. Like this:

2009-07-03

  http://lib/it/processMMS: update scenarios' table

2009-07-02

2009-07-01

  import `recsep' to CVS

Note the "hanging" 2009-07-02.

Okay. Let's omit the boring details (head scratching, pencil sketching, and a couple of "aha!" moments) and goto right to the way I debugged the damn function...

Here the simplified model of a work log suffice: a "d-a-i" sequence -- sequence of ['d', 'a', 'i', '-'] characters, where 'd' represents date, 'a' -- active entry, and 'i' -- inactive. Hyphen ('-') is either empty line or the continuation of previous entry (a task entry in real log can consist of several lines).

All we need to do is change a bit the hiding function (substitute regexps, replace `forward-line' with `forward-char', and show the "hidden" regions in inverted color instead of actually hiding them) and run it on a bunch of dai-sequences.

Testing input is easy to generate:

import Data.List (permutations)

main = putStrLn $ concatMap (\(x,y) -> x++y) (xs `zip` repeat "-")
    where xs = concat $ permutations ["a", "a", "i", "i", "d"]

Given "i-d-a-d-i-d-a-i-" we get:

i-d-a-d-i-d-a-i-

Do you see a hanging date? The hidden (black) region in the middle should cover preceding date ('d-') but it does not.

Well, you got the idea. Once the testing routine was established, it was easy to fix the bug.

i-d-a-d-i-d-a-i-

Q.E.D.

* * *

Yes, and while you're here, I'd like to recommend this post:
Benefits of automated functional testing (was: Why unit testing is a waste of time)

Have fun!