Older blog entries for pesco (starting at number 22)

Sometimes it can be so easy...

...as chvt(1). I'll just run an extra X-server without WM on a separate virtual terminal and switch to it remotely via SSH. It won't give of fancy transition animations à la Beryl or Mac OS X, missing some of the futuristic-look-factor, but the functionality is what counts, right?

On a side note, this whole plug-me-in-and-I'll-autojoin-the-larger-setup business reminds me of reading about "FlashMob computing" the other day. It sounded pretty much like they were still missing such "auto-integrate" (and auto-"exegrate") functionality...

PS. I apologize for a lack of links in my posts, but I've been sending them from my cell phone (yay advopost) and can't look up the references, so bear with me for having to google out the links yourselves... ;)

12 Feb 2007 (updated 12 Feb 2007 at 23:26 UTC) »

My laptop as a Plug&Play device

I'm the owner of two laptops, one slightly older and rather bulky 17" PowerBook, and a recent Panasonic T5. I now use the PB as my (stationary) main display via remote X while all work is done on the T5. To this end I've already got a script to automate the "attaching" of the T5, i.e. launching a window manager on the remote display and joining the two displays wrt. keyboard and mouse input with x2x.

I still have to manually kill the old WM on the PB, which I keep running when the T5 isn't connected (the PB still runs my MP3 player and an "emergency" web browser), and start up the network connection on the T5. The latter would be easy to automate but what I'm really looking for now is a way to also automatically have the WM on the PowerBook squirelled away in some "fast user switching"-like fashion so the regular environment could set itself up. Then on some other event, the whole thing should revert to the disconnected setup, automagically unsquirreling the "solo" environment again. I wonder how well the current state of the art in Linuxland will allow me to do that.

I've got my project for the weekend...

4 Feb 2007 (updated 4 Feb 2007 at 15:53 UTC) »
Oops, a duplicate!
4 Feb 2007 (updated 2 Feb 2008 at 00:02 UTC) »

Oh my god, I finally found it...

...a decent MP3 player with stream support. Finally.

Hello, MOC[1]! XMMS, go eff yourself.

Of course I'm sorry for not using hmp3[2], but it doesn't do streams, appearently. At least not that I can tell from the website. Also missing Ogg support is a bit of a let-down.

"Music On Console" http://moc.daper.net
"hmp3 :: purely functional sound" http://www.cse.unsw.edu.au/~dons/hmp3.html

Back from the dead, with a nonlinear parser

Soo, everything went well different than planned. What was supposed to be a holidy clean-up rewrite of a fun weekend project has turned into a half-year side project running next to university.

To recap, I initially set out to implement a Markdown[1] parser in Haskell so I could post formatted text to my Advogato blog. An email-to-Advogato gateway was quickly whipped up[2]. The first prototype version of a Markdown parser was also finished within reasonable time[3]. Unfortunately, the code was a mess, so I set out for the rewrite[4]. Much progress was made but it kept screwing up in certain minor but annoying cases and the code still looked convoluted. Basically, Parsec just didn't want to bend in the right direction...

So I replaced Parsec. The module is called Text.ParserCombinators.Nonlinear[5] because it allows one to slurp in parts of the document in one part of the parser and reparse them again later. This allowed me to split up the document according to its block-level structure and re-assemble, for instance, the text pieces of quoted or indented lines (without the leading quote marks/indetation) and run the corresponding parser over the thus extracted subdocument. Such embedded parses can also work with a completely different token type than the enclosing parser, a capability which also came in handy.

I recently came across "Frisby"[6], a Haskell implementation of PEG grammars, which I had never heard of before. The description sounds cool. I wonder if my Markdown variant could be represented by one? My parser library is neither optimized for space nor speed, and PEGs sound compelling in that regard...

Anyway, the implementation based on my nonlinear parsers worked out really nice wrt. the code structure and doesn't show any of the kinks that plagued the Parsec version. Since I've deviated somewhat from Markdown syntax in the places I didn't like, I've dubbed the package k-tex. I've still got to update the documentation but if anyone is interested in looking at or even improving the code, you can find it at http://www.khjk.org/~sm/code/k-tex/.

Best regards, Sven Moritz

PS. Yep, the Advogato gateway[7] already uses k-tex, and if this post appears on my blog[8], it's working. ;)


"Structural plain-text, next iteration"

Structural plain-text, next iteration

So, the first working version of my "structural plain-text" parser is just working and I'm already dissatisfied with it. Well, the reasons are simple:

  1. It doesn't work correctly in all cases.

  2. The code is a mess, even though it was already the third or so iteration internally ("plan to throw...", you know).

So here we go for the next iteration. I hope I'll be able to make the parser code substantially cleaner and more modular and implement a few cool-to-have additional features along the way. But first I decided to carefully write out a description of the intended grammar. I just finished the first draft which is still missing details on the planned new features. It's probably still full of inaccuracies because I've changed some details from last time (in hopes of simplifying both the parser and the syntax explanation) and I haven't implemented the actual parser, yet. Still, if you're interested, have a look[1].

PS. Oh yeah, after all, I've included "strong emphasis" in this version, because there are uses for it, specifically in technical documentation (which I intend to write). It carries a Fat Warning (TM) in the docs, so I'm happy and people can do whatever they want. ;-)

Structured text in Haskell redux, or "Posting to Advogato ASCII style"

A few weeks ago I posted my initial thoughts about how to represent a structured text document in Haskell. As often with initial thoughts, they were flawed. ;-) In particular, I wrote:

I've tried to design the above types in such a way as to minimize the possibility of forming non-sensical or ambiguous documents. That's why there is such deep nesting of different types instead of just one big algebraic data type with constructors for concatenation, paragraph and section breaks, etc.

Without going into too much detail, I stumbled over the problem in the representation of block quotes: The corresponding constructor expected a full Doc and I couldn't break sufficiently small fragments out of those for quoting. For example, had I wanted to quote one item out of a numbered list, it would have gotten the number "1" again in the quote, because it would have been the first item within that document (fragment)...

So, I thought, should I make one big data type for the whole document after all? But still, I didn't want to include too much possibilities for such senseless combinations as bulleted lists and blockquotes within section headings. After some pondering I've decided on using a two-level structure, with one data type representing documents (or document fragments) at the block level (paragraphs, sections, etc.) and a second one for in-line structures only. In particular, a document is taken to be a list of "block-level tokens" ('Btok's):

type Doc = [Btok]
Analogously, (logical) lines are represented by line-level tokens:
type Line = [Ltok]
Apart from plain strings, these consist of things like emphasis, code spans, etc.:
data Ltok =  St  String           -- character string
          |  Em  Line             -- emphasis
          |  Co  String           -- inline code
Note that under their intended interpretation, these constructors form homomorphisms into the monoid of Ltok-lists (wrapping their result in a single-element list): e.g. [St a] ++ [St b] is expected to be equal to [St (a++b)], [Em a] ++ [Em b] == [Em (a++b)], etc. This enables one to split a line in two at an arbitrary point without "losing any information". :-)

I noticed that if something like the above could hold for the block-level Doc type as well, my block quote problem would be solved pretty elegantly. And indeed, I found the following representation:

data Btok =  TEXT  [Line]    -- text block (logical lines)
          |  PARA  Line      -- new paragraph (w. opt. title)
          |  SECT  Int Line  -- new section (of given level)
          |  QUOT  Doc       -- blockquote
          |  BULL  [Doc]     -- bulleted list
Notice that the PARA and SECT constructors do not wrap the corresponding paragraph or section. Their meaning is analogous to that of a new-line character in plain-text: they start a new structure which spans everything up to the next break. This is exactly what makes QUOT work: We can break the document apart in the middle of a section or paragraph and rightfully consider the resulting parts fragments in the common sense that they can be glued back together to form the original. Or to put it differently, a quoted fragment carries all of its "own" structural information (where the meaning of "own" also results from the data type's structure).
"Wait, there's more fun to be had!"

Remember that the whole point of my foray into structured documents originated in my little project of posting to my Advogato diary via email. So, say I had a document to be posted in memory as a value of type Doc as above. I'd want to convert it to HTML (or Advogato's subset thereof). While the representation of section breaks as stand-alone SECTs within a stream of block tokens closely fits with the <h1-6> tags to appear in the HTML, paragraphs are supposed to be wrapped in <p> tags. One obvious (but annoying) solution to this discrepancy would be to build a second data type, probably a little more like my original idea again, and convert to HTML via the detour of this intermediate type.

Luckily I remembered that Oleg[1] wrote something about folds being superior to cursors[2] as interfaces to data collections, so I implemented a function of the following humiliating type signature. ;)

-- Fold over a Doc's block-level structure.
  ::  (Biblio -> Line -> line)     -- logical line
      -> ([line] -> tok)           -- text block
      -> (String -> tok)           -- code block
      -> (doc -> tok)              -- blockquote
      -> ([doc] -> tok)            -- bulleted list
      -> (Int -> [doc] -> tok)     -- numbered list
      -> ([(String,line)] -> tok)  -- bibliography
      -> (  line                   -- para. title
            -> [tok]               -- para. content
            -> para  )             -- paragraph
      -> (  Int                    -- sect. level
            -> line                -- sect. title
            -> doc                 -- sect. content
            -> sect  )             -- section
      -> (  [para]                 -- initial paragraphs
            -> [sect]              -- following sections
            -> doc  )              -- document
      -> Doc -> doc
As you can see, it takes ten functions as parameters, each telling it what to do with a certain part of the document, which it then transforms accordingly. Given this function, an HTML export for Docs is straight-forward to define.

NB. Notice how the different type variables are threaded through the type signature. It feels almost like LISP, in the sense that the functions can return any type they want, as long as everything fits. But unlike in a dynamically-typed language, the typechecker will tell you up-front if you make a mistake. :-P

"Show me the code!"

Ah right, most important! The full definitions of the Doc type, the folddoctree function, and some further utilities can be seen at the following address:


Enjoy! :-)


Oleg Kiselyov, Haskell (and Scheme (and probably more)) guru. See http://okmij.org/.
Oleg Kiselyov: "Towards the best collection traversal interface". http://okmij.org/ftp/Computation/Continuations.html#enumerator-stream
11 Jun 2006 (updated 11 Jun 2006 at 22:30 UTC) »
Fun with email -- or how I spent my day

I'd like to share the results of my latest "freshmeat" spree. For context, I keep all my email on a University server, so I can access it from anywhere (No, there are no webmailers!). I have access through IMAP which I used with Thunderbird, until yesterday -- when I finally got fed up with it. Thunderbird regularly mixes up its idea of which mail folders there are on the IMAP server, and where. So there I was, looking for a new MUA. Lot's of dissatisfaction ensued. Until I found two wonderful programs:

  1. OfflineIMAP
  2. Mutt-NG

The first does something very simple (but in a smart way): It synchronizes IMAP mailboxes with a local repository. Both ways. So I set it up with all my IMAP settings (through a nice, and "unuserfriendly" config file -- sic!) and let it sync every five minutes. New mails get pulled into local Maildir folders where any old MUA can conveniently handle them. And whatever changes I make to them are propagated back to the server. And no, it doesn't lose any mails; so the author assures. The author, btw., turns out to be a fellow Haskell programmer -- John Goerzen, thank you: You just made my day!

Thanks to OfflineIMAP I can finally use my old-time favourite MUA again: Mutt. But wait, as I discovered, there is even Mutt-NG now, which finally adds the one thing I had been missing all along: A sidebar that shows an overview of (new/old/etc.) messages in my incoming mail folders! Oh, and for extra pleasure, it also speaks SMTP, so I don't even have to set up a local delivery agent.

And here's some more general Mutt praise: With just a few pretty simple config file (sic!) settings, mutt pretty much takes care of moving read mails away into yearly archive folders all by itself:

save-hook ~A =sa-spam

This means that pressing 's' to move a message to another folder presents sa-spam as default, which happens to be the folder my spamassasin learns from.

set record="=Archive/`date +%Y`/Sent"

This is where mutt saves Fcc's, i.e. copies of sent messages. Notice the automatic year selection.

set mbox="=Archive/`date +%Y`/Inbox"
mbox-hook =Somelist   =Archive/`date +%Y`/Somelist
mbox-hook =Otherlist  =Archive/`date +%Y`/Otherlist

Finally, these are the real fun: Remember how default Mutt setups usually ask on quit whether they should save read messages to "mbox" in your homedir? A completely stupid question, nobody wants that. But with the above, this suddenly changes to something sensible: The correct archive folder for the current year, sorting different input folders into a corresponding archive subfolder. Wonderful.

Maybe you cannot really appreciate how good this is. But that's usually the case with improved usability, you don't notice until afterwards how much better things feel with just that little tweak. Compare my previous post about bitlbee. Life just started being fun again! Time to throw out all those black clothes and start listening to eurodance.


PS. Friendly apologies for praising config files go to lkcl who recently bitched about "this crap obscure text file editing shit". ;-P

10 Jun 2006 (updated 10 Jun 2006 at 16:00 UTC) »
"About the Swedish analysis of Nazi Germany's crypto teleprinters"

I've just returned from a two-semester exchange at the Charles University[1] of Prague. There I had the chance to take a course about classical cryptography, for which I wrote an essay that might be interesting to others:

Everybody knows about the Enigma machine[2]. It was a cryptographic device developed by the Germans, used extensively by Hitler's troops, and famously broken by the allies at Bletchley Park. It was an "offline" device in that the plaintext was keyed in letter by letter, indicating the corresponding ciphertext character with each keystroke. Now, what many are not aware of is that the Germans also used a series of online crypto devices during WWII. One of these was the Siemens T52[3], a.k.a. the "Geheimschreiber". This machine was an actual teleprinter with built-in automatic encryption and decryption, i.e. it encrypted and transmitted the characters as they were typed (or fed via punched tape), and vice-versa. It's cryptographic algorithm is considered significantly more sophisticated than that of the Enigma.

When Germany leased several telegraph lines from Sweden they were immediately tapped but the T52 was soon employed on them. My essay[4] describes how the Swedish cryptanalyst Arne Beurling broke the encryption without ever having seen one of the machines. It's mainly a condensed form of the description from "Codebreakers"[5], with less history and more mathematics.



"Univerzita Karlova v Praze"
Wikipedia: "Enigma machine"
Wikipedia: "Siemens and Halske T52"
Sven Moritz Hallberg: "About the Swedishh analysis of Nazi Germany's crypto teleprinters" (2006) http://www.khjk.org/~sm/distfiles/sm-2006-gschreiber.pdf
Bengt Beckman: "Codebreakers: Arne Beurling and the Swedish Crypto Program during World War II", Oxford University Press (2003)

Update: Typesetting a reference list in Advogato

For those of you who, like me, are not completely fluent in CSS (and maybe a little stuck with HTML wisdom of the 90's), here's how you produce hanging indents as in the reference list above:

  1. Somehow determine the width of your left column. For the reference list, I took the maximum number of characters in it plus one, times the width of the 'x' character (ex): "[n]:" are four characters, so I use 5ex.
  2. Put the left column text into a <div> and give it the attribute 'style="float: left"', so that the following text will flow down to the right of it.
  3. Put the right column text into another <div> and give it the attribute 'style="margin-left: 5ex"' (replacing 5ex with whatever left-column width you want). This results in the hanging indent.
For reference, here's the code to the first reference entry above:
<div style="float: left">[1]:</div>
<div style="margin-left: 5ex">
  "Univerzita Karlova v Praze"<br>
  <a href="http://www.cuni.cz">http://www.cuni.cz/</a>

For a reference list, it also looks good to indent the whole thing a little by putting it in another <div style="padding: 1ex">.


1 Jun 2006 (updated 1 Jun 2006 at 10:23 UTC) »
Representing marked-up text in Haskell

I wrote previously about my plan to use Markdown as the input format for advopost. I decided against re-using the existing Markdown-to-HTML converter, because I would have to strip the resulting output down to the Advogato subset of HTML in postprocessing; feels too clutchy. So I'm going to implement a parser for (a variant of) Markdown that reads the input into a structured Haskell data type 'Doc'. Here is my current design for that type:

   module Doc where
   data Doc     =  Doc         String      -- title
                               [Para]      -- body
                               [Doc]       -- subsections 
   data Para    =  Paragraph   String      -- paragraph title
                               [Block]     -- paragraph body
   data Block   =  Blockquote  Doc
                |  Bulleted    [[Para]]    -- unordered list
                |  Numbered    [[Para]]    -- numbered list
                |  Codeblock   [[Inline]]  -- list of lines, ignore Codespans
                |  Line        [Inline]
   data Inline  =  Str         String      -- ignore linebreaks
                |  Codespan    [Inline]
                |  Emph        [Inline]
                |  Link        [Inline]    -- link text
                               String      -- link target
                               String      -- link title
                |  Image       [Inline]    -- fallback alternative for this image
                               String      -- image location
                               String      -- image title

I want both the input format and the Haskell data structure to be independent of the output format being HTML. Therefore I'm not going to support inline-HTML in the input. I also want structural markup (as opposed to presentational), so I left out horizontal rules and forced linebreaks. Lastly, I've never heard of using a "strong emphasis" (as opposed to normal emphasis) in typesetting, so I dropped that as well.

I've tried to design the above types in such a way as to minimize the possibility of forming non-sensical or ambiguous documents. That's why there is such deep nesting of different types instead of just one big algebraic data type with constructors for concatenation, paragraph and section breaks, etc.. Comments welcome.

I hope that the 'Doc' type will be useful in further coding. For example, it would be really cool to have a fancy combinator library for 'Doc's along with a pretty-printer to turn them back into plaintext: Then we could use them for general pretty output from Haskell programs. While there are several existing pretty-printing libraries, to my knowledge none of them use structural markup and they are all targeted at console output only.

13 older entries...

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!