A few days ago someone said something that reminded me about my audio player, which I had abandoned for more than a year already. The reason was mostly that the two Phonon backends, VLC and gstreamer, for some reason or other couldn't play the files I had without any gaps between songs.
To be honest, the first bug end up being me not properly encoding the filenames.
If you first URL-encoded the filename and then built a Q/KURL with that, then
it's all fine. It took me more than 12 months and a few rereads of the thread
to realize it. Fixes apart, it seems that the bug still exists for other
instances of gstreamer errors, so we're not out of the woods. In any case, I
switched to the VLC backend and it seems that now is able to fire the
aboutToFinnish() signal properly, so for the moment I'm using that.
All that is fine, but that's not what I wanted to talk about in this post. Given
that this project largely precedes my interest on testing, it has no testing at
all. Most of the project is straightforward enough to almost no need any, but
there's a critic part that would not suffer at all if it had any, namely the
Collections handling, including passing files from one to another and
automatically updating new/removed
So after fixing the bug mentioned above I tried to figure out the current state
of affaires regarding
Collections, and boy, they're in bad shape. The code was
locally modified, never commited, deactivating any notifications of filesystem
changes (new or removed files), and other code I can't really understand the
Because if this last detail is that I decided to start testing three classes:
Collection, which handles a set a
Songs with a common root directory;
CollectionAggregator, which handles a set of collections and should coordinate
Song from one
Collection to another; and
CollectionIndexer, which scans
Collection's root dir to find
All went fine while I tested the first class, Collection. There was a tricky
part where I had to setup a
QApplication in order to make signals work. The
problems began when I started testing
CollectionIndexer. Tests started blocking
endlessly, signals stopped being either emited or firing the connected slots,
life was bad.
I tried to search the available documentation and mailing lists for a hint about the problem, but besides a quite complex example that didn't seem to properly converge to anything useful, I was mostly on my own.
This morning I got my eureka moment: I noticed that if I executed each test class
by itself, it worked, but both at the same time blocked and never finished. Then
I remembered something said in
For any GUI application using Qt, there is precisely one QApplication object, no matter whether the application has 0, 1, 2 or more windows at any given time.
That was it: I was creating the application, first in the
setUp() method, then
as a class attribute, but I had one test class per class to test, each in its
own file. Somehow this last fact lead me to think that somehow they were executed
in separate processes, which is not true. Luckily, even with this limitation, there's
none on the amount of times you can
quit() the same instance, so
that's what I did: I created only one instance and reused it everywhere. I was
already doing that for each test method, but again, somehow having several files
mislead me to think they were isolated from each other.
So now all my unit tests work without mysteriously blocking forever. Now I just
hope I can keep riding the success wave and bring
satyr into good shape. A new
release wouldn't hurt.
 No matter how much I try, I can't get any vaguer.
 Ok, maybe the
Playlist combo wouldn't hurt to have UTs either.
After several months thinking about it, and just two requests, I finally decided
satyr's code. I decided to use github because I already switched to
git, mainly for testing and understanding it. I think I
can live with
hg, althought branch management in
git seems to be given more
thought and a good implementation.
So, without further ado: satyr in github
Remember, it's still a test software, by no means polished or ready for human
consumption, and with very low development force. Still, I think it has some
nice features, like interchangeable skins and a well defined backend,
support, quick tag edition, reasonable collection managment, and, thanks to
Phonon, almost-gapless playback and things like «stop after playing the
current this file» (but not «after any given file» yet).
In Debian Sid it mostly works only with the
GStreamer backend; I haven't tried
xine one and I know
VLC does not emit a signal needed for queueing the
next song, so you have to press «next» after each song. AFAIK this is fixed
satyr-0.3.2 "I should install my own food" is out. The Changelog is not very impressive:
- We can save the list of queued Songs on exit and load it at startup.
- The queue position is not presented when editing the title of a Song that is queued.
- Setting the track number works now (was a bug).
- Fixed a bug in setup.py.
but the last one is somewhat important (Thanks Chipaca). Also, 2 months ago, I made the satyr-0.3.1 "freudian slip" release, when user 'dglent' from http://kde-apps.org/ found a packaging bug. It was not only a bugfixing revision, it also included new features:
- nowPlaying(), exported via dbus.
- 'now playing' plugin for irssi.
- Changes in tags is copied to selected cells (in the same column). This allows 'Massive tag writing'.
- Fixed "NoneType has no attribute 'row'" bug, but this is just a workaround.
- Forgot to install complex.ui.
Now go get it!
Continuing with the development of
Satyr, which doesn't have any GUI. I
thought it would be faster to make a
DBus interface that a good GUI, not
to mention more interesting.
The code snippet of today is this:
class DBusServer (dbus.service.Object): def __init__ (self): bus= dbus.SessionBus () bus_name= dbus.service.BusName ('org.kde.satyr.dbus.test', bus=bus) dbus.service.Object.__init__ (self, bus_name, "/server") @dbus.service.method('org.kde.satyr.dbus.test', in_signature='', out_signature='') def bing (self): print "bing!" @dbus.service.method('org.kde.satyr.dbus.test', in_signature='', out_signature='') def quit (self): app.quit () dbus.mainloop.qt.DBusQtMainLoop (set_as_default=True) dbs= DBusServer () sys.exit (app.exec_ ())
This simply defines a class which registers itself with the session bus
under the name
org.kde.satyr.dbus.test, exporting itself under the path
/server and then defining a method that goes bing! :) and another one
that quits the app. Note the decorator for the methods.
You might notice the
call. This is needed because both Qt and DBus in asyncronous mode (which is
the one we're using and the only one that works under Qt or Gtk, AFAIK) both
have their own event loops, and that makes some kind of magic that let both
loops coexist without blocking the other. This must be called before
connecting to the bus; otherwise, you get this error:
<span class="createlink">RuntimeError</span>: To make asynchronous calls, receive signals or export objects, D-Bus connections must be attached to a main loop by passing mainloop=... to the constructor or calling dbus.set_default_main_loop(...)
So, let's test the beast. We run the script in one terminal and in the other:
mdione@mustang:~/src/projects/satyr/live$ qdbus | grep satyr org.kde.satyr.py-10154 org.kde.satyr.dbus.test mdione@mustang:~/src/projects/satyr/live$ qdbus org.kde.satyr.dbus.test / /server mdione@mustang:~/src/projects/satyr/live$ qdbus org.kde.satyr.dbus.test /server method void org.kde.satyr.dbus.test.bing() method QString org.freedesktop.DBus.Introspectable.Introspect() mdione@mustang:~/src/projects/satyr/live$ qdbus org.kde.satyr.dbus.test /server org.kde.satyr.dbus.test.bing mdione@mustang:~/src/projects/satyr/live$ qdbus org.kde.satyr.dbus.test /server org.kde.satyr.dbus.test.quit
And in the other console:
mdione@mustang:~/src/projects/satyr/live$ python dbus_test.py bing!
It goes bing!... and then finishes. The next step is to make my
class to export its methods via
DBus and that's it!. More info in the
Satyr handles paths. There are some problems with paths and (sigh)
encondings. Of those, here are two: there's no way to know in which encoding
the filenames in a filesystem are enconded (f.i., there's no way to ask the
filesystem), and even if that were possible, the filenames might not even be
enconded in that enconding. In these (still!) transitioning times, lots and
lots and shitloads of filesystems are used in UTF-8 environments, but some
filenames are still in old ISO-8859-1 or whatever the system was using
QString. I'm taking a path from the command line; this path is
the location of the (right now only)
Collection for the player. I'm
handling the command line using
KCmdLineOptions, which returns
As we all know,
QString, just like the
unicode type in
all the data internally as Unicode, which is The Right Thing™. If you
really need the internal data, say as bytes, you can always call the
constData() method and be happy with it. This would be the case for paths;
you need the bytes.
PyQt4. For some reason, which maybe I will ask in the pyqt
constData() is not available. What to do? Well, that's what this post is
about. What you're about to read is hacky as it can be, but then it works. I
might feel dirty, but I can live with it. As long as I mark it as a
utter/über hack and promise to revert it once that's possible...
# path is a QString qba= QByteArray () qba.append (path) path= str (qba) # now path is a list of bytes/string.
Even if this part of the bug is fixed, then
Phonon.MediaObject.play() fails when feeded that same path with this
ERROR: backend <span class="createlink">MediaObject</span> reached <span class="createlink">ErrorState</span> after 1 . It seems a <span class="createlink">KioMediaStream</span> will not help here, trying anyway.
or simply refusing to continue. Sure, in my case I should simply ignore the filename and inform the user what's going on, but sometimes you can't be so gentle.
 but then it doesn't make much sense now since Phil wants to get rid of QString (for several reasons, which might most possibly include this one).
satyr got the possibility to queue songs for playing very early. At that
moment there wasn't any GUI, so the only way to (de)queue songs was via
dbus. Once the GUI was there, we had to provide a way to queue songs. As
satyr aims to be fully usable only using the keyboard, the obvious way was to
setup some shortcut for the action.
Qt and then
KDE have a very nice API for defining 'actions' that can be
fired by the user. The ways to fire them include a shortcut, a menu entry or a
button in a toolbar. I decided to go with the
According to [the
classKAction.html#_details), to create an action is just a matter of creating
the action and to add it to an
actionCollection(). The problem is that nowhere
it says where this collection comes from. There's the
but creating one and adding the actions to it seems to be not enough.
If you instead follow the
see that it refers to a
KXmlGuiWindow, which I revised when I was desperately
looking for the misterious
actionCollection(). I don't know why, but the
documentation generated by
PyKDE does not include that method. All in all, the
tutorial code works, so I just ported my
class MainWindow (KXmlGuiWindow): def __init__ (self, parent=None): KXmlGuiWindow.__init__ (self, parent) uipath= __file__[:__file__.rfind ('.')]+'.ui' UIMainWindow, _= uic.loadUiType (uipath) self.ui= UIMainWindow () self.ui.setupUi (self) [...] self.ac= KActionCollection (self) actions.create (self, self.actionCollection ())
def create (parent, ac): '''here be common actions for satyr skins''' queueAction= KAction (parent) queueAction.setShortcut (Qt.CTRL + Qt.Key_Q) ac.addAction ("queue", queueAction) # FIXME? if we don't do this then the skin will be able to choose whether it # wants the action or not. with this it will still be able to do it, but via # implementing empty slots queueAction.triggered.connect (parent.queue)
Very simple, really. But then the action doesn't work! I tried different approaches, but none worked.
The tutorial I mentioned talks about some capabilities of the
one of them, the ability to have the inclusion of actions in menues and
toolbars dumped and loaded from XML file (hence, the XML part of the class
name), and that is handled by the
setupGUI() method. From its documentation:
«Configures the current windows and its actions in the typical
[...] Typically this function replaces
createGUI()». In my case the GUI is
already built by that
self.ui.setupUi() that you see up there, so I ignored
this method. But the thing is that if you don't call this method, the actions
will not be properly hooked, so they don't work; hence, my problem. But just
calling it make the actions work! I'll check later what's the magic behind this
method. For now just adding
self.setupGUI() at the end of the
So, that's it with actions. As a result I also get several things: a populated menu bar with Settings and Help options (but no File; that'll come later with standard actions, of which I'll talk about later, I think), with free report bug, about satyr and configure shortcuts options, among others. The later does work but its state is not saved. That also will come in the same post that standard actions.
PD: First post from my new internet connection. Will
satyr suffer from this?
Only time will tell...
 then you had to pick the index of the songs either by guessing or looking at the dump of the playlist.
There's a problem with
Phonon. If you try to play a file in a format that
it doesn't support, instead of failing or simply sending the
signal it just does nothing. I'll have to look more deeply into this and
figure out if it's a bug or feature or how to work-around it.
For now I just decided to use
python-magic to decide if the file is
playable or not. This module uses
libmagic, the same library that the
file utility uses. But I hit a problem with it. One of its ussages is to
magic.file() with the filename, but if the filename contains a
non-ascii character you get something like this:
Traceback (most recent call last): File "satyr.py", line 173, in next self.play () File "satyr.py", line 128, in play mimetype_enc= self.magic.file (self.filename) <span class="createlink">UnicodeEncodeError</span>: 'ascii' codec can't encode character u'\xed' in position 37: ordinal not in range(128)
I found a workaround. There's another usage, which is to call
magic.buffer() with a piece of data to recognize. So I just simply
open() the file,
read() 4KiB and pass the data to that function.
f= file (self.filename) data= f.read (4096) mimetype_enc= self.magic.buffer (data)
Now it works properly. I will file a bug report to the
file package once I
resort out my mail setup (I cannot send bugs with
So I mostly finished implementing a
QAbstractTableModel subclass with
read/write support, and some more stuff. Here's a resume of what changed from
the implementation of [a read-only
columnCount(), the first important difference occurs
data() itself. Now the
QModelIndex.column() is important, because it
tells us which column the view wants to populate. In the case of the
DisplayRole I ask the
Song for the corresponding atribute/key; in the
case of the
SizeHintRole I use a static array of strings to provide default
values. I will explain why later:
def data (self, modelIndex, role): if modelIndex.isValid () and modelIndex.row ()<self.aggr.count: song= self.aggr.songForIndex (modelIndex.row ()) if role==Qt.DisplayRole: attr= self.attrNames [modelIndex.column ()] data= QVariant (song[attr]) elif role==Qt.SizeHintRole: size= self.fontMetrics.size (Qt.TextSingleLine, self.columnWidths[modelIndex.column ()]) data= QVariant (size) else: data= QVariant () else: data= QVariant () return data
The second difference is related to the view itself. A
QTableView presents not
only the contents but also what they call 'headers'. These are the column
headers ans row numbers that normally appear on top and left of tables. By
default numbers are shown, which in may case is ok for rows, but not for
columns. So we implement
headerData(), which is somehow similar to
section parameter refers to the column when direction is
def headerData (self, section, direction, role=Qt.DisplayRole): if direction==Qt.Horizontal and role==Qt.DisplayRole: data= QVariant (self.headers[section]) else: data= QAbstractTableModel.headerData (self, section, direction, role) return data
Once I have a basic read-only table, I start nitpicking on its appearance. First, all the columns have the same width, which is not acceptable. I want narrower 'Year', 'Track' and 'Length' columns, wider 'Artist', 'Album' and 'Title' columns and a huge 'Filepath' column. So, we set the column widths from the ui:
class MainWindow (KMainWindow): [...] def connectUi (self, player): [...] # FIXME: kinda hacky self.fontMetrics= QFontMetrics (KGlobalSettings.generalFont ()) for i, w in enumerate (self.model.columnWidths): self.ui.songsList.setColumnWidth (i, self.fontMetrics.width (w)) class QPlayListModel (QAbstractTableModel): def __init__ (self, aggr=None, songs=None, parent=None): QAbstractTableModel.__init__ (self, parent) [...] # FIXME: hackish self.columnWidths= ("M"*15, "M"*4, "M"*20, "M"*3, "M"*25, "M"*5, "M"*100)
Now the rows, the rows! They're too thick! I can only see tweintysomething songs
in my 1440x900 screen with a small plasma panel on top, and I already got rid of
the empty menubar and the status bar.
headerData() is also called with a
SizeHintRole, so let's use it (once more, the third time already, using the
QFontMetrics). Note that it also sets the width; and while I'm on it,
I also align the row numbers to the right:
def headerData (self, section, direction, role=Qt.DisplayRole): if direction==Qt.Horizontal and role==Qt.DisplayRole: data= QVariant (self.headers[section]) elif direction==Qt.Vertical: if role==Qt.SizeHintRole: # again, hacky. 5 for enough witdh for 5 digits size= self.fontMetrics.size (Qt.TextSingleLine, "X"*5) data= QVariant (size) elif role==Qt.TextAlignmentRole: data= QVariant (Qt.AlignRight|Qt.AlignVCenter) else: data= QAbstractTableModel.headerData (self, section, direction, role) else: data= QAbstractTableModel.headerData (self, section, direction, role) return data
Now I can see 36 rows maximized and almost 39 when in full screen. Maybe I can make the columns headers thinner... but once more I digress.
Now to the interesting part: making the model read/write. First we must
implement a method called
flags() which says a lot of things about cells; in
particular, if they're editable or not. Clearly 'Length' and 'Filepath' are not
def flags (self, modelIndex): ans= QAbstractTableModel.flags (self, modelIndex) if modelIndex.column ()<5: # length or filepath are not editable ans|= Qt.ItemIsEditable return ans
setData() itself. It has a similar interface as
including the new value. It should return if it could successfully change the
data value, which in this case means if
Tagpy/Taglib could modify the tag.
Note that the role while changing the value is
def setData (self, modelIndex, variant, role=Qt.EditRole): # not length or filepath and editing if modelIndex.column ()<5 and role==Qt.EditRole: song= self.aggr.songForIndex (modelIndex.row ()) attr= self.attrNames[modelIndex.column ()] try: song[attr]= unicode (variant.toString ()) except : ans= False else: self.dataChanged.emit (modelIndex, modelIndex) ans= True else: ans= QAbstractTableModel.setData (self, modelIndex, variant, role) return ans
It's really that simple. Now we have a
QTableView and a
which are read/write. This means that
satyr can modify and save tags now,
which is one of my most wanted features.
One last detail: for the way
satyr has its
QTableView configured, one can
enter in edit mode only by typing something in a cell, but then there are two
columns which cannot be modified. If you type something while the cursor is in
one of these columns,
QTableView will call
match() for matches to the typed
letters. The default implementation basically searches through all the cells,
which means that
data() is called for all the
Songs, which finally implies
that all the tags are read from disk, which is expensive for making the simple
mistake of typing in a read-only column. As
satyr already has a search
feature, I fixed this 'misfeature' with this code:
def match (self, start, role, value, hits=1, flags=None): # when you press a key on an uneditable cell, QTableView tries to search # calling this function for matching. we already have a way for searching # and it loads the metadata of all the songs anyways # so we disable it by constantly returning an empty list return 
 I have my reasons to provide both interfaces.
 Plans for saving the state of the
satyr main window will include the
column widths, but that's another story.
 It's usefull for writing tags of Songs with no tags... and because I can!
 ... which might return in the future to show scanning process, but again I digress.
 Yes, another digression: at some point it will be possible to move the file to a filepath based on its tags; in other words, to arrange your music collection based on tags. Should not be too far in the future.
 Last digression: I will modify this behaviour as soon as I finish with this post.
One of the features I planned for
satyr almost since the begining was the
possibility to have 'skins'. In this context, a skin would not only implement
the look and feel, but also could implement features the weren't available in
the shipped classes. I also planned to implement this feature after I had most of
the others one already done. But then I was bored this weekend with nothing to do
and I decided to set off to at least investigate how to do it. Of course, what
happened was that I implemented it almost completely.
Up to now,
satyr's user interface was implemented in two files:
which was compiled with
default.py, and some code in
satyr.py itself. This
of course would not scale, and I always had the idea of moving the behaviour
satyr.py to a file called
default.py and load the ui directly
default.ui file without compiling, getting rid of the need for a
compilation at the same time. This also meant that then a skin would
consist of a
.py file and possibly a
.ui file. There are three problems to
solve for this: getting the local-to-the-user's skin directory, loading the skin
and loading the correspondant
The first part is simple from the
PyKDE4 point of view:
# get the app's dir; don't forget the trailing '/'! appDir= KStandardDirs.locateLocal ('data', 'satyr/')
I'll first explain the other two parts, loading the skin and the
before returning more deeply to the consequences of this solution.
I put all the skins in a
skins subdirectory. To make
it a proper python module I added an empty
__init__.py file. Now, I could
import skins.<skinName> and possibly instantiate some class in it, but
of course one cannot write that. I could resort to
eval ('import skins.'+skinName), but we know that
eval() has the most
long-standing typo in the history of computer languages, and it's actually
What we can do is resort to
__import__() instead. This little function does
approximately what we want. I say approx because it has some surprises in the
sleeves of its sleeveless code. I suggest you to go read carefully its
documentation. Meanwhile, the magic itself:
mod= __import__ ('skins.'+skinName, globals(), locals(), 'MainWindow') mw= mod.MainWindow ()
.ui file in the skin's code is rather simple: just get the skin
module's filepath, replace
.ui, and load it with
PyQt4.uic.loadUiType(). This function returns a generated class for the
topmost widget and its Qt base class. This generated class has a
method that is the one that actually builds de UI.
So, we just instantiate the main window's class and call its
from import uic # !!! __file__ might end with .py[co]! uipath= __file__[:__file__.rfind ('.')]+'.ui' # I don't care about the base class (UIMainWindow, buh)= uic.loadUiType (uipath) self.ui= UIMainWindow () self.ui.setupUi (self)
Note the comment about the
__file__ attribute of a module.
Now, and back to the first part, finding the local-to-the-user's skin directory is the easiest part. From there, things get a little bit more complicated:
skinssubdirectory might not exist.
- If you create it, you gotta make sure to also throw in a
- Once you've done it, you also need to add the local-to-the-user's app
directory to the path. It's easy, just prepend it to
sys.path, so it's used before any system-wide directory.
- The last problem that remains is exactly that: once the
__init__.pyis there and the user's dir is prepended to
sys.path, the user's local skin directory is always used when importing anything from the
skinsmodule, so if a skin is not there it is not loadable. All skins distributed with
satyrwill be inaccesible!
So I'm in a kind of dead alley here. I have a couple of ideas on how to work-around this, but they're at best hacky, and I don't want to implement them until I'm sure that it's inevitable.
 Not a very happy name, if you ask me.
 Very similar to what you get if you compile the
So far both the default and complex skins in
satyr were (ab)using the selection
in their respectives views to highlight the current song and, in the case of the
complex skin, that meant that you couldn't select cells. This could be (and will
be) handy for editing several tags at the same time or for grouping songs in an
Album. I decided that this should not longer be this way.
It was really simple: I simple reenabled selection, the extended version (Shift
and Control work as you grew to spect). Now the way to highlight the current
song is to simply change the background and foreground colors for the cells of
that row. How? Simple, extending the
QPlayListModel.data() method to provide
these properties and using
QApplication.palette() to get colors that would
stand out from the normal application colors (even the selection) and would
still make the text readable:
def data (self, modelIndex, role): if modelIndex.isValid () and modelIndex.row ()<self.aggr.count: [...] # check the row so we highlight the whole row and not just the cell elif role==Qt.BackgroundRole and modelIndex.row ()==self.parent_.modelIndex.row (): # highlight the current song # must return a QBrush data= QVariant (QApplication.palette ().dark ()) elif role==Qt.ForegroundRole and modelIndex.row ()==self.parent_.modelIndex.row (): # highlight the current song # must return a QBrush data= QVariant (QApplication.palette ().brightText ()) [...]
And that's almost it... because there's always a but. Whenever the current song
changes, there's a little lag between this happens (you can hear the new song
being played) and the highlighting updates. This is because I update that
self.parent_.modelIndex you see up there, but the
QPLM.data() method is not
called until the next repaint, which doesn't happen inmediately. So, we must
force that repaint. How? Again, it's simple: we already emited the
dataChanged() signal in the method
setData() when the data could be
successfully changed. Now we do the same just so the view updates the
def showSong (self, index): [...] # FIXME? yes, this could be moved to the model (too many self.appModel's) start= self.appModel.index (oldModelIndex.row (), 0) end= self.appModel.index (oldModelIndex.row (), columns) self.appModel.dataChanged.emit (start, end) start= self.appModel.index (self.modelIndex.row (), 0) end= self.appModel.index (self.modelIndex.row (), columns) self.appModel.dataChanged.emit (start, end) [...]