Blog Syndicate content

  • » Will code for food

    Tue, 12/22/2009 - 05:34

    Ha, what a bit of a bribe can do to me… Someday earlier today a user of the PHP plugin for KDevelop brought up the flickering issue in it again. Well as I told him: I myself find it very annoying and wanted to fix it since quite some time, but never got around to it… Usually that would be it and I’d go watch some more FamilyGuy until I’m in the mood to track this bugger down. But well, thankfully Phlogi wasn’t so easy to dispatch:

    <Phlogi> milian: ok… I’ll send you pizza and beer if you fix this!

    Hours of gdb sessions later, I finally committed a fix. So Phlogi, if you read this: You owe me ;-)

    To all others: If you tried the PHP plugin out and the flickering was too much for you: Give it a try again! I’m personally totally overwhelmed, the difference is huge! I often perceived the PHP plugin to be magnitudes slower than the C++ one. Well, looks like most of this was only due to the flickering. Now things are much smoother.

    Happy holidays!

  • » PHP and PHP-Docs plugins now up for review (meaning: SVN location moved!)

    Thu, 12/17/2009 - 22:29

    Hey all!

    Just a quick note: Niko and me moved PHP & PHP-Docs to kdereview, we hope to move both plugins to extragear/sdk/kdevelop-plugins. So, if I understood things correctly, after a two week period the plugins will get moved there (well, if we pass the review, but I think we can do that).

    So for anybody that uses the plugins from SVN, you’ll have to relocate. The new addresses are:

    • /trunk/kdereview/kdevelop-php
    • /trunk/kdereview/kdevelop-php-docs

    See you in two weeks :)

  • » Real FOSS appreciation

    Mon, 12/07/2009 - 03:25

    So, it’s been roughly a year since my first commit to kdelibs. According to Ohloh it’s been in November 2008. And boy have I learned much in this year. I learned C++ just to be able to contribute to KDE, since I thought it would be awesome to be able to “fix your own itch”. I have to say: It was the best decision I ever made.

    I really came to appreciate FOSS in a whole new light: Contributing to a big project like KDE gets you in contact with lots of nice people. And they will help you get things done. What’s better is that in the process you learn lots of things. And I mean lots. I can now use GDB, Valgrind, now my way around some parts of the KDE/Qt API, can investigate performance related questions… And since these are such huge topics, there’s always more to learn, much more!

    I doubt someone could learn that much by reading books or writing his own little application without the help of a community. The wealth of possibilities inside KDE will increase your horizon constantly. And there is tons of very good code to study! Want to know why something is not working? Look at the source. Still not helping? Ask your fellow developers. I really have to say it’s an awesome feeling to be part of this big community.

    Fixing an itch: Qalculate! backend for Cantor

    You know, I’m officially a Physics student, I just happen to be way more interested in programming (the practical part of it, not really that much the theoretical part).

    Since I started with Physics two years ago, I always required a good calculator, esp. for experimental physics. The best calculator I found was Qalculate!, especially it’s great support for units and constants made me solve tedious exercises in a fraction of the time it required my fellow students that could not use Qalculate (it only runs on Linux). Really, it’s an outstanding piece of software in my opinion.

    But to get to the point: Qalculate was the last KDE 3 app on my desktop, and I wanted to change that. Now I read about Cantor, esp. once fellow KDevelop hacker Apol wrote about his KAlgebra backend for Cantor, and I knew: This is the perfect fit for libqalculate!

    So last week I started and already have a somewhat working backend available:

    http://websvn.kde.org/trunk/playground/edu/cantor-backends/qalculate/

    This once again showed the utterly insane greateness of FOSS: I started to hack on the backend and looked at the existing backends for guidance. I also contacted arieder, the author of Cantor, directly and chatted with him, getting help. But I did not only /take/. I reviewed his API, gave suggestions, reported whishes and bugs and eventually hacked a little bit on the sources themselves. Imagine this in a corporate environment: I’d probably have had to report in some shabby tracker and wait ages for my feature request to be closed as wontfix…

    And while writing the backend I had this realization that also triggered this blog post: I actually did something useful in a few days. I could never have imagined being in that position one year ago. I always thought that C++ was a bit of black magic, especially writing things from scratch. But now… Now I’m somehow able to grasp code and come up with something working in a few days.

    The problem this brings, is of course that you can easily loose track and overload yourself with work… I could spent lots of hours in any part of KDE. My TODO alone bears for KatePart, KDevelop-PHP, KDevelop itself, Quanta, …. You name it! I think this will become a fun holiday ;-)

    PS: Just a quick note: I really think that Cantor will become a great and useful application for science students. It will be your central application for any calculations, either numeric (e.g. Qalculate) or symbolic (e.g. Maxima).

  • » Kate Love: HighlightInterface, Autobrace

    Sun, 11/22/2009 - 20:15

    Well, I have to admit: I didn’t spent much time developing the PHP plugin for KDevelop these past weeks. Instead I hacked on Kate:

    HighlightInterface

    I added another Kate interface, this time to access some of the highlighting information:

    • what’s the Attribute for a given default style right now? Default styles are those known from syntax files, e.g. dsKeyword, dsFunction,…
    • what are used Attributes in a given line and what range do they occupy?
    • what modes do we embed? E.g. PHP embeds HTML, JavaScript, CSS, …
    • what mode is used at a given Cursor position?

    This made it possible to port the “Export to HTML” action to a real plugin. If you come up with other output formats I might add them, I wondered about LaTeX support… might do this at some point.

    This should also make it possible to use KatePart in other applications and than export the highlighting to a different format, e.g. a Flake shape for Koffice. Afaik this is actually planned by The_User - lets see if it works out!

    The other stuff gives huge potential in various places, but I fear it won’t make it in KDE 4.4. But think of it:

    • simple code completion based on keyword databases, dependent on the mode at the position where completion was requested
    • same as above for snippets (actually this will make it to 4.4).
    • insert your ideas here :)
    Auto-Brace plugin

    Jakob Petsovits created this gem of a plugin some time ago, yet it lived in playground was probably only used by few. I imported it to kdelibs, hence it will be shipped with KDE 4.4. It supersedes the limited “auto-brackets” feature of Kate and only adds braces when a newline gets added. I find this fits my personal coding habits much better than blindly copying brackets when they get added.

    And I don’t just copied to kdelibs, I also added a few features:

    • automatically add a semicolon after the closing brace when we start a new struct/class in C++ mode
    • check for auto-brackets feature and disable it automatically

    Also did this:

    • don’t add brace when current line contains namespace and a following line starts with class or struct (C++ mode only)
    • don’t add brace when current line contains class, interface or struct and the following line contains private, public, protected. C++ code is also checked for signals, Q_SIGNALS", other modes are checked forfunction`.

    This should fix the bug for code like (note the indendation levels):

    1. namespace foo { // insert line here
    2. class bar;
    3. }
    4. class asdf { // insert line here
    5. private:
    6. ...
    7. };
  • » KDevelop & KWrite/Kate hacksprint 2010 in Berlin

    Mon, 11/16/2009 - 16:38

    Hey everybody!

    If you are a KDevelop and/or Kate/Kwrite developer and do not read the mailing lists: There’s a hack sprint coming up in Berlin in 2010. I think there’ve been enough sprints in Berlin already so that you know it’s a great city for such an event. Though this time it won’t be at KDAB or Nokia offices, but at the Physics Faculty of the FU-Berlin. Since I (currently) work there as an IT admin, it was my first choice and worked out. I hope it will be a good location for the meeting. If you want to attend, vote on doodle:

    http://www.doodle.com/vkyh9up9794zr4s8

    But you probably also should register either on the KDevelop or KWrite mailing lists so I have some kind of way to contact you.

    PS: in unrelated news I’ll do an internship at KDAB next year! yay

  • » Improving KDevelop-PG-Qt

    Mon, 10/26/2009 - 19:13

    Good news everyone :)

    Documentation

    After years of pretty much no documentation (except looking at the sources…), The_User aka. Jonathan Schmidt-Dominé started documenting the parser generator that is used for most KDevelop language plugins (java, python, php, …). You can find it here: http://techbase.kde.org/Development/KDevelop-PG-Qt_Introduction

    It has to be improved and more examples have to be put in there, but it’s already a huge improvement over the situation before…

    Performance

    In related news I did some profiling on the parsing of the quite big file that includes all internal PHP declarations (i.e. all functions, classes, definitions,…). It drops in at a whopping 1.9M, with ~80k lines. Well, turns out that this showed a pretty easy to fix bottleneck in KDevelop-PG’s LocationTable, which used to use a linear lookup algorithm. Profiling showed that nearly 75% was spent in that function. But I used the past tense for a reason:

    I replaced it with an algorithm that combines a relative search (i.e. relative to the last lookup) with a binary search fallback. That’s comparatively blazingly fast. I added some benchmarks to KDevelop-PG-Qt that proofs that (benchmark below run with release mode build):

    1. ********* Start testing of KDevPG::Benchmarks *********
    2. Config: Using QTest library 4.5.2, Qt 4.5.2
    3. PASS : KDevPG::Benchmarks::initTestCase()
    4. RESULT : KDevPG::Benchmarks::positionAt():"initial, linear":
    5. 1,184.0 msec per iteration (total: 11840, iterations: 10)
    6. RESULT : KDevPG::Benchmarks::positionAt():"initial, random":
    7. 1,008.5 msec per iteration (total: 10085, iterations: 10)
    8. RESULT : KDevPG::Benchmarks::positionAt():"relative, linear":
    9. 1.3 msec per iteration (total: 14, iterations: 10)
    10. RESULT : KDevPG::Benchmarks::positionAt():"relative, random":
    11. 1,185.0 msec per iteration (total: 11850, iterations: 10)
    12. RESULT : KDevPG::Benchmarks::positionAt():"binary, linear":
    13. 31.1 msec per iteration (total: 312, iterations: 10)
    14. RESULT : KDevPG::Benchmarks::positionAt():"binary, random":
    15. 39.2 msec per iteration (total: 392, iterations: 10)
    16. RESULT : KDevPG::Benchmarks::positionAt():"binary & relative, linear":
    17. 0.8 msec per iteration (total: 8, iterations: 10)
    18. RESULT : KDevPG::Benchmarks::positionAt():"binary & relative, random":
    19. 40.2 msec per iteration (total: 402, iterations: 10)
    20. PASS : KDevPG::Benchmarks::positionAt()
    21. PASS : KDevPG::Benchmarks::cleanupTestCase()
    22. Totals: 3 passed, 0 failed, 0 skipped
    23. ********* Finished testing of KDevPG::Benchmarks *********

    In our “realworld” phpfunctions.php example the DUChain building process got twice as fast (in debug mode though, but still).

    We should probably get lost of the LocationTable altogether and use an existing container (QList, QVector, QLinkedList,… or any STL variant of them). But this would mean more profiling, and well, lets see when we get to it… But just got a reply on the KDevelop list showing interest on this, so maybe someone does that eventually!

    Profiling rocks, and KCacheGrind is such a great application. I love such visualization. Well done! And the QTestLib benchmark utilities are also very solid, nice!

    PS: KDevelop beta6 is in the pipelines and we’ll release a beta1 of the PHP plugin together with it. Packagers and users rejoice :)

  • » first experience with Archlinux

    Sun, 10/18/2009 - 02:24

    So, I kinda messed up my desktop right after the upgrade to karmic, because I was too greedy for performance and converted my root file system to ext4. Well, that worked like a charm on my laptop, but it broke my desktop. This is in no way karmic’s fault, it’s my own misbehavior. Thankfully I could rescue most of my data.

    Since I’d had to reinstall anyways, I decided to finally try out Archlinux. I find the rolling release mantra very intriguing. Together with a “simpler” packaging, namely no splitting between -dev and -dbg packages like debian/ubuntu does, this is destined to be a good environment for a developer. I always hated it to track down missing -dev packages when compiling software. And don’t get me started on outdated software in repos… I just compiled kdelibs and the only missing build dependency was hspell, that I don’t need anyways. Under Jaunty I had to compile stuff from kdesupport to fulfill updated dependencies. And the list of not-found optional dependencies was huge, since I did not spent time to install all those -dev packages by hand…

    My first impression of Archlinux is very good so far. I also finally migrated to 64bit wich works like a charm, no issues with flash or anything. Since I never used a 64bit Ubuntu/Debian I’m not sure, whether the perceived performance increase is due to the switch to 64bit or whether Archlinux optimized packages are responsible. Probably both. Nevertheless I can safely say that my system feels snappier than before.

    Of course, the installation and initial setup is not as straight forward / easy as with Debian/Ubuntu: Yet it’s no big deal for anyone with some Linux experience. And, once everything is setup, you are running KDE again, so no real difference. Thanks to the Chakra team for kdemod, it works like a charm!

    I might have spent a bit more time during the installation / initial configuration, but I think this would have happened also if I’d installed any other distro I’ve never used before, like OpenSuse or Fedora.

    Oh and since I can install sudo I can keep my old habits. Neat.

    The only thing I miss so far is aptitude with it’s straight forward command structure. Yaourt/Pacman is fast and nice, esp. with pacman-color, but the commands don’t feel as straight forward to me… Personal preference I’d say.

    To conclude: Archlinux is very nice, I can wholeheartedly recommend using it so far. Probably nothing for a novice Linux user, yet perfect for advanced users. Very good as a development environment. Fast. Up to date. I like it :)

    Now I can finally continue hacking on Kate/Kdelibs again :) I’m currently in the process of refactoring Kate’s implementation of the TemplateInterface. Even in it’s current state it already implements features like mirrored snippets and the like. But once I’ve finished with the cleanup I will try to implement some more of the features that are found in e.g. yasnippet for Emacs. I really wonder why nobody else did that already…

    Once this is finished, you can expect that I will deeply integrate that feature in various places in KDevelop, especially for code completion, snippet plugin etc. pp. Stay tuned!

  • » KDevelop PHP digest - August to Oktober 2009

    Fri, 10/02/2009 - 15:08

    Hi there again! I’ve been silent again on my blog, but didn’t rest on development. In the one and a half months since the last digest, I started writing a PHP application This finally made me eat my own dog food :). It resulted in lots of polishing and quite a few bug fixes for the PHP plugin in KDevelop. Here’s a list of what I think are the notably changes since the last digest:

    (Note: to view screenshots, go to the bottom of this article.)

    • refactoring of parts of the Code Completion code, should already result in faster code under certain situations
    • properly mark constants as “Kind: Constant” in the declaration tooltips
    • offer argument hints for ctors during code completion in class init statements
    • greatly improve the generate inline documentation of built-in PHP functions, classes, properties etc. pp.
      • add documentation of public properties
      • support aliased functions (thanks to Victor Grischenko for his patch)
      • show more/all documentation, and not only the first paragraph
      • fix type-lookup
      • don’t get confused when a documentation file documents both, a method and a function (greatly improves e.g. MySQLi documentation)
    • don’t offer “jump to declaration” for built-in PHP declarations
    • add support for list(...) statement
    • cleanup code-completion list, esp. show the return type of functions in the prefix field, and not something a la “function ReturnType ($arg1, $arg2, …)”
    • improve the code-completion for include/require statements
    • add language constructs to code completion (e.g. class, while, foreach, print, …)
    • show declaration tooltip for magic constants, showing their current value
    • make functions, methods and classes case-insensitive, just like PHP handles them
    • some performance improvements, especially in code completion and parsing of the generated file containing php-internal declarations
    • lots of bug fixes, don’t want to iterate them all ;-)

    I created a Flickr account and will push screenshots to it for every upcoming digest I do. To see some of the features I talk about above, visit: http://www.flickr.com/photos/milianw/sets/72157622375662249/detail/

    EDIT: A note to those that want to try this out , but cannot / don’t want to compile it: I heard that there is / will be an AUR package for Arch users available. And I’d really appreciate it if others could create packages for their favorite distribution as well. You’ll need bleeding-edge kdevplatform + kdevelop though!

  • » Spotlight: linux-minidisc

    Thu, 09/03/2009 - 18:36

    Hi there!

    Today I want to abuse the fact, that my blog is aggregated on some planets, to bring a project of a friend of mine into the spotlight:

    The Linux-Minidisc project

    It’s a project to bring Read/Write access on mini discs to Linux. It consists of a CLI and a Qt Gui. All code is licensed under the GPL and can be accessed via Git.

    To get a nice introduction about the project, read this excerpt from a recent Linux Journal edition: http://users.physik.fu-berlin.de/~glaubitz/minidisclj.pdf

    The project itself has a wiki under the following address: https://wiki.physik.fu-berlin.de/linux-minidisc/doku.php

    As every other FOSS project, Adrian and his fellow developers need more man power. Especially someone who can spruce up the Qt GUI is needed. Help them!

  • » Easily access quassellogs from the CLI

    Thu, 08/27/2009 - 16:56

    Quassel is really a cool program. I like how I can use it from everywhere and access the same set of data. Now using IMAP and Quassel I’d really look forward for similar shared access to other IMs, but that’s not the topic of this blog post.

    What I want to introduce is a new addition to my set of shell helpers, called quassellog:

    1. $ quassellog -u milian -b "#kdevelop" | tail -n 1
    2. [2009-08-27 13:09:11] milian > hi all
    3. $ quassellog -b "#kdevelop" | tail -n 1
    4. [2009-08-27 16:43:35] Fersis!n=Fersis@190.245.176.231 > yeah i did
    5. $ quassellog
    6. quassellog [-u USER] [-b BUFFER] [PATTERN]
    7.  
    8. -u USER show only messages from users, who have USER at
    9. the start of their sender name.
    10.  
    11. -b BUFFER show only messages in this buffer
    12. valid buffers are:
    13. ##linux #khtml &IMP ...SNIP...
    14.  
    15. PATTERN a simple pattern, use * for wildcard matching
    16.  
    17. NOTE: order of options is not exchangable, i.e. first -u, then -b then pattern...

    You can get the code from github. To make it work from all your servers I recommend the following alias:

    1. alias quassellog='QUASSEL_HOST=theHostQuasselCoreRunsOn quassellog'
    2. # e.g. for me:
    3. alias quassellog='QUASSEL_HOST=jongleur quassellog'

    That way the script runs automatically on the right server (provided it’s also accessible there under the same path). Have fun!