Issue 167

Fix the code completion model so that it computes the completion items in a background thread again.
Seems like this was broken when the code was moved from cpp to kdevplatform.

This makes the UI more responsive when automatic completion is enabled.

It might lead to slight regressions, because we use multi-threading where it accidentally wasn't used before (foreground lock was added in one place).

Don't acquire the foreground lock twice recursively, once is enough.

Don't try entering the "release+trylock" loop if we already hold the foreground lock. Else, we might get a deadlock if acquire it recursively.

Don't crash when setting an invalid map theme and the fallback theme does not work either.

forward-port r1215778: don't crash in equirectangular projection when the map doesn't cover the whole view

Guard against faulty routes.

Comply with INDI wire protocol for sending enableBLOB.
Delete temp files are fitsviewer shows them.
Thank Gerry for reporting this!

Use new kxkb dbus APIs

Add a small safety net here, though I'm not sure if it's necessary, if it's the problem at all

Switch off BSP indexing in the preview graphics view.
Workaround, presumably a Qt problem.

Fix wrong JPEG 2000 MIME type

Together with fixes in kdelibs it is now possible to save JPEG 2000 images in KolourPaint with adjustable quality.

Note that for JPEG 2000, the quality parameter really means a file size percentage, so you should start with values around 15, which already results in good quality.

If you do not get the "Quality" widget in the save dialog, delete the kolourpaintrc file, as KolourPaint caches the relevant MIME information in its config file.

Moved the call to takingThread() so it definitely comes after the value of the the variable 'state' is updated in transitionToRunning().

This avoids two instances of the same DynamicThreadPriv being scheduled at the same time.

Don't remove the Konsole widget from the layout when it has been destroyed already.
This fixes a crash.

Assure that no endless searching is done if a link points to a parent directory

FIXED-IN: 4.6.0

Check if list is empty.

Remember all iterated directories to prevent an endless recursion by links.
Tested with search all files below /

QString is not thread-safe; since the matches get shuttled around between threads, we need to protect access to it.
as writing is rare, this should hopefully not degrade performance too much. if it passes testing, then i will backport this.

Fully clear cache metadata in clearInternal().

KSharedDataCache uses the clearInternal() method to reset the cache, which is used if
corruption is detected, and when directed through KSharedDataCache::clear(). For whatever
reason I thought that simply setting the "first page" pointer for each possible index
entry would be sufficient, but that is not the case at all, as various methods which go
through the cache entries also check things such as the use count, last access time, etc.

Assuming no other errors (:-/) this should be the cause of bug 260309. (The
exact sequence would be a KSharedDataCache user runs ::clear(), later adds some
entries that invoke removeUsedPages(), and removeUsedPages() goes through *all*
index entries, including ones that appear to be in use since their use count is
>0, but their first page is still invalid). Easiest way to invoke this bug I've
found is to simply run "plasmoidviewer clock" at a konsole.

While reviewing code I also added a comment clarifying the usage of a page
variable as an index entry, and changed the index entry comparison functions to
look for any invalid page instead of checking for -1, as it takes me 5 minutes
every single time I review removeUsedPages to prove that -1 is the only
possible value <0. These changes will not be backported to KDE Platform 4.6.

Differently from the previous version, now power profiles can be valid even if they have no groups in them: change the code to reflect this change.

Found how to reproduce the qfswatch hang bug in the unittest.
Fixed by adding a qAddPostRoutine which destroys KDirWatch::sef() while qApp is still around.
If this proves stable in trunk, let's backport it to the 4.6 branch

Don't assure the visibility of the leaf, if another directory has been opened by the [+]-toggle.

Properly handle DBus error replies when checking the init state of services.

This fixes a bug where services are started before the deps are initialized and all sorts of problems appear.

Improve DolphinTreeViewTest's stability.

When running the test in the background while using the computer for
other things, it may happen that the test's widgets are not focused,
which led to an infinite loop before this commit. I've removed the
checks that caused this loop because the only thing that is important
for the test is that the view loses the keyboard focus at some point
(see bug 220898).

Fix wrong JPEG 2000 MIME type

This fixes saving JPEG 2000 images from KSnapshot or KolourPaint.

Fix issue that the "From Here" search restriction does not work on subfolders when Nepomuk/Strigi has been activated.

Apply better fix for the use of KDirWatch::self() in the destructor of a singleton:
make it possible, by not destroying the KDirWatch early, only its qfswatcher.

Patch by alter schwede, thanks!

remove the item pointer from all collections when it is deleted as it is also added to all of them when first added

Assure that the enabled plugins for previews are set for the Information Panel and the tooltips.
This fixes the issue that a disabled thumbnail-plugin might be chosen.

For 4.7 it should be considered to move this code into KIO::filePreview() as default already (will provide a patch on the reviewboard).

Make sure we honor width: set on a button with display: inline.

This is done by mutating it into inline-block (WebCore does that,
but a lot more other stuff, too, which doesn't make sense here given
our native-widgets approach).

Fixes the star (so that's what the weird fancy character was) and loupe
icons on google search results getting cutoff.

Fix problems with negative indeces in bracket ops in cases we can statically determine arguments to be of int32 type. Fixes reply problem on userbase.kde.org.

I think I'll probably just have to bite the bullet and add in a uint32 type to FrostByte's type system, but getting conversion rules right
for that will be tricky... and it'll be extra-annoying that all the bitops like to produce int32.

Unbreak KJavaAppletViewer. QStringList::begin is not an output iterator.

Fix a bug in khtml::StyleBoxData::operator == that made us incorrectly hashcons different styles, messing up paste.kde.org

This fixes severe performance problems with complicated proxy scripts.

1) Whenever we get a string that may be either a hostname or IP, try
checking it for being an IP with QHostAddress first, rather than handing it to
QHostInfo::fromName directly, as that method does reverse lookups on IPs.
Takes proxy overhead on a certain complex but well-written script from
0.5 seconds (per-request, yikes) to about 50ms or so.

2) Use the KIO DNS cache when possible in cases we have a hostname, as there
is obviously quite a bit of locality in hostnames used. This takes the
overhead down to the 2-5ms range.

Reviewed by Thiago.

Various fixes for the new twitter website:
1) The alleged DOM 0 DOMNode::item doesn't exist in Gecko and WebCore at least,
and it confuses the script loader (as we have NodeList-like .item, but not an
index [[Get]]). Remove it, but add one to HTMLSelectElement as that's legit
HTML5.

2) When document.domain is set to the present domain, we still want to call
setDomainFromDOM(), as it would set the 'script-set domain' bit moving the
origin into that namespace --- otherwise it will be inaccessible to other
things that opted in.

3) XMLHttpRequest should complete relative URLs in its own context, not the
dynamic one.

- Update the status correctly when the Network Status changed
and not only when the status of the connmanager created by
solid changed.

- Check the online status by querying the adapters ( exposing
Networking::checkStatus to avoid code duplication )
instead of relying on the connmanager created by solid.

- Ignore the status of USB Serial connections.

Stay consistent with the non-indexed searching and also return folders as search-result when searching in indexed folders.

There is no point in counting results when a limit is set.
Virtuoso will ignore it anyway and the query might take very long.

* Fixed the handling of quotes and keywords such as "AND", "OR", and "NOT" in LiteralTerm.
Now correct bif:contains or regex filters are created for values.

* Made the query parser merge LiteralTerms into a single one to improve query performance.
While merging two LiteralTerms into one does not yield the exact same query (when merged
both literal tokens need to appear in the same property value while with separate LiteralTerms
the tokens can appear in different properties) it should cover close to all typical use cases
while increasing the performance significantly and getting rid of the nasty "Virtuoso goes
crazy when I use KRunner" bug.

Fix mouse selection for embedded KCompletionBox

If a KLineEdit is embedded in QML, the completion box should bypass the graphics proxy widget, otherwise it won't receive mouse events.

always reset the cursor, and do so before the drag

- Remove the use of private qt api. This did not work as intended
so the status was never updated. When there was no instance of
networkingcontrolmgr. (The minimalnetworkingclient used to test
before does that)

- Instanciate WinCE Networking Control Manager because it handles
the Window Messages notifing about a Network status change.

- Create the Networking control manager as a child window
of the correct qt main window. So Messages are recive correctly
even in multithreaded applications with multiple windows.

- Cleanup debug output a bit.

use the evil underscore here too to map to the ugly twitter api

When editing the name of file items opening a context-menu for the text-editor may not result in closing the rename-operation.
Thanks to Zé for the patch!

try to reize columns based on applets hints.
center if the contents are smaller than the view

this fixes the behaviour of newspaper in plasma-desktop quite a lot, is probably worth backporting

Refactoring of SendFile Wizard, now it only has 1 page.

FIXED-IN: 1.1

Reset the timer each time transferProgress is emmited.

The session is actually used when the transfer is going on, so we
have to reset the timer in order to avoid invalid sessionTimeout.

FIXED-IN: 1.0.2

If org.openobex is unregistered, restart the server.

Seems that obex-data-server is crashing a lot for some people
(including me) so we have to handle it properly.

FIXED-IN: 1.0.2

Deleting a batch of incidences should be an atomic operation.

Otherwise, When deleting 1000 events you'll be asked 1000 times if you want to send invitations.

Separate global from folder specific messagelist config

The user can configure a global messagelist display format
in the settings dialog now and overwrite it with a folder specific
display format from the context actions menu.

Update the Transport combobox on identity change

This ensures that a special transport, configured for an identity,
is used in the message composer.

Fixes second part of 258350 -> closing

Add 'Next 7 days' mode to timeline view

Do not set the KFileDialog save default to a dir

This is more of a problem in KFileDialog when there are
no Qt filedialogs available. The setting for the start
FileName is not recognized as a folder and set to a
directory.

This only happens if there is no proposedFileName available.

Make sure the applications are unregistered when the module is unloaded.

MERGE: 4.6

Update the password of all Transport instances when
changing the password in the configuration dialog.

Fixes first part of bug 258350

KCal::Event::dtEnd() is inclusive, not exclusive.

This is an important detail for all-day events.

Fixed some more places that assumed that KCal::Event::dtEnd() was exclusive instead of inclusive.

Disable 'Save' buttons after click

Since the ContactEditor and ContactGroupEditor might need some
time to finish saving, we should prevent the user from clicking
save multiple times, otherwise the contact (group) is created multiple
times as well.

fix that widgets get lost
It seems that widgets get a setVisiable(false) from somewhere in qt, to fix this issue we revert its hiding.

This should be fixed in qt.
But i didnt found where it is called.

Set annotations correctly for new collections

This fixes the bug that collections, created by the Kolab
resource, lose their annotations and therefor do not show
up in kaddressbook/korganizer/knotes

Fixed the Burn composite option.

Should now work exactly like the Color burn mode in Photoshop
but only for fully opaque colors. Transparency is not considered jet.

Memleak fixes
Show a database select page after the connection page
(Blank for file based source atm)

Separate field for income and expenditure when importing CSV files

Conversion multiple records to "transfer" type cause all have the same ammount

fix text encoding bug
iso/iec 8859-12 doesn't exist, therefore no codec can be created.
thanks for the report!

Episodes without guid should also be checked.

Fix potential crash with the Similar Artists applet.

Thanks to Tuomas Nurmi for the patch.

Remove dead IDs from the queue

Remove track IDs that are about to be removed from the underlying
playlist model. Fix crash when activating the queue editor when an
already removed track was still in the queue.

Do not store track rating as tags in files.

Patch by Alan Ezusti.

Fix minor inconsistency in the Collection view.

Patch by Bhargav Mangipudi.

Fixed issue with Organize collection, which prevent to removeing newly copied tracks.

fixed bug in filesystem abstraction that broke file uploading

Fix tab preview flickering on mouse hover.

Patch by Furkan Uzumcu, cleaned up and reviewed by me

Fix tab bar not matching TabBarWidget state in some circumstances

fix uploading files to folders with special characters in the name

Fix mounting of shares with special characters such as hidden ones.

Fix a bit sending several outgoing webcam requests + a typo fix that denies in some cases sending webcam data.

From the ReviewBoard description:

a) seems like yahoo protocol doesn't send back in the webcam
invitation accepted which person did that. and there is a variable
to remember who we sent the last invite to. Problem would be if
several invites are going to be sent. So, I made the invitation name
to be an invitation *list*. Inspired by code in libpurple.

b) there is a typo when code tries to guess who the 'accepted
invitation' is coming from. I don't have a good way to replicate,
but mostly I suspect it happens when you're sending several
invitations (see point a) ) or when some online<->offline switches
are happening. Concrete side effect is that code tries to connect
to server "".

Patch by Cristi P <cristi.posoiu AT gmail>, thanks a lot!

REVIEW: 6333

Fix libktorrent not gettting updated when the download of a magnet link finishes and the FileSelectDlg is shown

Fix the display of the active tab in main_content.

The size of the page could be wrong because the content of a tab
can update after it is loaded. Since the current animation allow it,
the main_content can vary with the size of the child.

The animation was not run when changing the display since it is
considered as programmatic change by the CSS animations. We force
a run of the event loop to make sure the animatons runs.

If for some reason packages in main do not have the "Supported" control field, catch this any return rather than blindly assuming the data to be there and crashing.

Also add a safeguard against malformed /etc/lsb-release files to prevent crashes if /etc/lsb-release happens to be malformed


FIXED-IN:1.1.1

Somehow running a python script directly without calling python /path/to/script causes zombie processes on exit.
The Kubuntu Notification Helper also runs a different python application directly, so I shall fix it there too.

Don't show the app launcher dialog for transactions that are only changing the addons of an already-installed application

Fix a bug where painting caused near 100% CPU usage and mis-sized cards.

This one was a bit of a doozy. The route of the problem is KImageCache's
in memory pixmap cache was occasionally returning the wrong pixmap for
a given key. Why that was happening I still don't know. So a card would
go to paint itself and notice that its pixmap was the wrong size, so it
would ask for a new one. The KAbstractCardDeck would take the request,
notice that the requested image/size was in the cache and return the
cached version, but unfornuately this was actually the wrong pixmap and
had the wrong size. The card happily used this pixmap and painted itself
but the next time it went to paint it noticed that its pixmap was the
wrong size and the loop repeated, hence the high processor usage.

I've disabled the in memory pixmap cache as the benefit wasn't very
significant and I couldn't figure out what the issue actually was. There
were also some problems with the KCard and KAbstractCardDeck code that
made the problem worse than it should have been. Those have been fixed
as well.

Add an optional filter action to StandartOutputView

We still need to connect it to the methode which will filter the items

Add a bookmark manager to the Marble Qt application. So far it has view capabilities and limited, in-memory only edit capabilities.

Start a directory to collect some sample data. Those 2 files are interesting because:
- they can be used to test the "Change data file" tool
- their formatting is compatible with fixed-width (13 characters) and white-separated, with some whitespaces at the end to stress-test the datasource
- they provide some metadata (units + some context info) which we'll soon be able to make use of

We have to discuss on the list how we want to go about sample data. I think it makes no sense to include them all in the distributed files, otherwise their size will increase a lot. And as I think we need to compress them, we may run into platform issues. I have gzipped those, but extracting them under Windows may be a problem for some...

I am not even sure it is a good idea to include all sample files in svn, maybe some links are better. The ideal solution might be to have a collection of URLs and provide under Help->Tutorials->Get Data Files a small dialog where the user can select the files he wants and the destination folder. They would then be downloaded directly from the Internet. Or even more simple: we could put up a dedicated tarball on sf.net with all interesting samples files in .zip and .tar.gz format.

Thoughts?

Initial support for rebuilding thumbnails only in current album

Into context pop-up menu, "Assign Tag" action will show recently assigned tags list instead the whole tags list which can
bloat entire screen if tags collection is huge.

A new option named "More Tags" can be used to switch on tags tree view from sidebar and give focus to tags list as well.

piwigoexport : New checkbox for uploading high-quality version

* The progress dialog box now displays the current step

- The "ServiceProvider" source can now also be used to get information about a specific accessor by it's ID (eg. "ServiceProvider de_db"), instead just for information about the default accessor for a given country code (eg. "ServiceProvider de").

- When reading an accessor from one of the default service provider symlinks the real filename is now used, which also means that the real service provider ID is used. That way the datacache-file storing information about accessor scripts doesn't include default accessors twice, eg. "de_db" and "de_default" (symlink for de_db).

- Corrected the "defaultAccessor" for "international" from "others_flightstats" to "international_flightstats" in the "Locations" data source.

- Accessors can now use session keys that get parsed from some URL and stored for departure/arrival/stop suggestion requests. Session key usage is defined in the XMLs and they are stored in the accessor until a timeout has passed (5 mins).

- Accessors can now also use POST as method to get a result document.

- Better departure date guessing and the possibility for scripts to return "dates need adjustment" to indicate that dates couldn't get parsed (eg. when requesting departures two days in the future, these two days need to be added to the dates).

- Readded at_oebb with a script for parsing (instead of only regexps) and updated it's URLs. It can now also parse route information, journey news, platform and delays.
- Better stop suggestion support for at_oebb: faster by requiring less data to be transfered, added stop weights.

Make wallpaper capable of setting particular Image
http://svn.reviewboard.kde.org/r/6375/

TweetJob redesign to make easier to add new Twitter api support
FEATURE: retweet operation support

Adding support for overlapping changes in paragraphs example:- text insertion followed by paragraph deletion

Introduced a comparison with an epsilon value in the setSaturation function.

The blend modes "Saturation" and "Hue" are actually broken.
Since the modes "Luminosity" and "Color" are working i expect the error somewhere in the setSaturation function.
I programmed it after the ISO 32000-1 spec and cannot find where i made a mistake.

New feature: Multihand
Initial hack on the freehand tool.

So far there is hardcoded symetry mode with the axis point in center with six brushes.

Added more blending functions.

Arcus Tangent, Gamma Light, Gamma Dark, Geometric Mean, Vivid Light and Pin Light

Added feature to keep the last 10 file states as backups. In the future we can make this count customizable

Initial addition of capability to create duplicate change-ids in KoChangeTracker.
i.e create a new change-id from an existing change-id which acts similar to a C++ reference variable

Added KoCompositeOpGeneric class to easily create new composite modes by just defining a composition function.

The KoCompositeOpGeneric class is a template that takes a function pointer
to a composition function as a second template parameter.

So for the most composition modes it is not needed anymore
to derive a new class from KoCompositionOp (or from one of its derivates)

plan: scheduling optimization

Add Calendar::firstAvailableBefore() to methods that needs special timezone treatment.
Fix a couple of bugs in calendar.

Add timezone definition to unit tests as calendars cannot work without them anylonger.

now we can actually edit the escape direction of a connection point

Implement support for importing a table

Tested by importing several tables from akonadi using a custom socket.
Both complex and v large tables worked fine.

added actions for changing connection point escape direction
added action groups as member of the tool so we can iterate over actions

Stable version for support of multiple and overlapping changes ( loading and saving ).
It's still a work-in-progress. More support expected to be added over the next couple of weeks

Added even more composite modes.

Including HSL composite modes (Color, Saturation, Luminosity, Hue).
The HSL CompositeOps will only be added to RGB color spaces.

Add send_to_zeitgeist function, to be implemented later for sending media playback activity to zeitgeist

Refactoring of MetaTagLib.

Split it up on separate classes, one for each Tag format.

Added support of ASF tags (.wma).

Added cover arts support for M4A (.mp4) and ASF (.wma) files.

New class for tag guessing. If track has no Title and artist, MetaTagLib will try
to guess It from file name automaticaly.

Moved TagGuesser dialog and MusicBrainzFinder on new class rails.

Added ability to guess tags from whole track path. (TagGuesse dialog)
(RR: 100191)

Support for ASF embedded pictures

Patch by Anton Sergunov

Initial version of media player/server

Provides a basic implementation of ampache's xml api and the start of a html5 based media player

Initial PostgreSQL support

the text viewer now does syntax highlighting for code

Initial version of remote ownCloud access through filesystem abstraction

Added more MIME Types in `inc/lib_filestorage.php`

Here the types added:
* Flac / Ogg [1]
* GZip / Tar / Tar+GZip
* PDF / SVG / Tiff
* JavaScript

Other little changes:
* I changed the ordered to be alphabetical (by extension)
* Using single quotes instead of double quotes for strings
* Indentation

[1]: http://wiki.xiph.org/MIME_Types_and_File_Extensions
[2]: http://www.asciitable.it/mimetypes.asp

Basic HTML5 audio player plugin

* Much improved context menu support:
* Removed as much of the unrelated menu options for certain content, e.g. images.
* Added new RMB options for HTML video and audio tags.

initial version for user managment tab in the settings page

Enable a match between two engines.

Implement draw offers and result setting in XBoard

Add the code to setup and start, but not display, a match between two programs

Now really parse and display castling moves.

- there could be multiple instances of AsciiSource
- also support multiple delimiters
- optimize when only one delimiter is used

Some ascii reading optimization:

Here some measurments: gyrodata c&p to 8 million rows

Delimiters (comment, column) : time in readColumns
-----------------------------------------------------------
No comment, custom: one space : 1.69102 seconds
No comment, whitespace : 2.11194 seconds
default comment, custom: one space: 4.01525 seconds (default comment with 4 characters)
default comment, whitespace : 4.34039 seconds
default comment, custom ,;| : 6.49383 seconds (space at the end of 4 characters)

Create settings widgets for some BQM tools only when they are needed.
Currently, at startup, all settings widget are created and kept in memory,
this should not be necessary.

A tool can choose not to create the widget in its constructor, but instead
implement a createSettingsWidget() method which must be called from
the framework before the tool is accessed. I hope to have found
all relevant paths where this must be done.

Not all tools are ported.

Highly improved initial update of legacy data.

Make korganizer start 0.7 seconds faster, and next/previous week/month more fluid.

HolidayRegion::holiday(date) is expensive, so, instead of calling it 42 times, call
HolidayRegion::holiday( start, start.addDays(41) ) only once.

Resurrect unsermake to see if we can make it work again here

unsermake should generate a faster set of Makefiles and it might also
obviate the need for automake at all.

Improved blending when alpha channel is locked.

The resulting color for compositing with locked alpha channel was not
100% right. I added special code for the case that the alpha channel is locked.
It should be slightly faster, too.

plan: scheduling speed optimization

Refactor project to use local time zone internally.
When a resource has a different timezone, it is handled specifically.
Improve caching of resource work time.

This should reduce scheduling time by ~25% for uncached project and ~90% for cached project.

Keep downloaded suggestions in memory.
This prevents from downloading it again, letting the completion
list be more reactive.

Only register the "Play" action for audio types that the browser is able to play.
Otherwise just leave the default action to "Download".

Improve performance of ktorrent in situations where there are many torrents:
- Log viewer no longer chokes on many log messages
- Load torrents earlier to prevent unnecessary gui updates
- Make GUI updates more efficient
- Don't sort when appending torrents to QM, seeing that orderQueue does that anyway
- Do not load all discovered torrents in one go in scanfolder plugin

only show files in the filebrowser that we can actually read.

in the (rare) case that there are unreadable files show an error since it's probably a case of incorrect premissions set by someone.

Only listen for signals from the worker if we requested a worker action ourselves by keeping track of when we have requested the worker and only connecting these signals when the worker has a) appeared and b) we requested it. This is a somewhat major behavioral change, so I will not backport it to the 1.1 branch for fear of causing regressions.

Due to a lack of foresight when I originally designed the worker/backend mechanism, I failed to design a way for the backend to be notified about things it requested. Instead the backend gets sent signals indescriminately over DBus that any other application using a QApt::Backend could trigger.

For QApt2 I plan on having the worker functions return a unique DBus object responsible for sending/recieving signals for a particular transaction.

Improve detection of short-noted moves

Don't invoke finder methods directly, but rather load the data beforehand and connect the thread's start signal to the finder's start slot so things can be truly asynchronous and non-blocking

- Dropped some service providers: de_vrn (seems like there are now only departure urls for specific lines), at_oebb (needs a script instead of only regexps for easier adoption to new website layouts, but but no urls directly to the timetables).
- Added debug output and error logging to accessor scripts to make it easier to find bugs, eg. if websites change. Errors are logged to a file, eg. "~/.kde4/share/apps/plasma_engine_publictransport/accessors.log".

- Updated fr_gares, dk_rejseplanen, us_septa, de_vvs to new website layouts.
- Date values are now correct for the de_db accessor, also if departures for a date other than today are requested.
- The de_db accessor now also recognizes inavailable delays for requests of departures that are more than two hours in the future. This allows to less often update data, because data with available delays gets updated more often.
- Updated the ch_sbb accessor to work for journeys again. The problem was changed HTML code from the SBB service provider.
- Added new functions to the ch_sbb accessor: Parse routes of journeys complete with departure/arrival times, types of the used vehicles, line string and platforms). Parse capacity information and add it to the journey news.
- Fixed ch_sbb to show all departures, not only every second.. This was a one character fix ;)
- Fixed cz_idnes, it didn't show new departures after some time.

Merged kio-magnet in KTorrent now, removed from playground.