HTTP code in GWeather

One of the things that pisses me off about gweather is that it occasionally hangs and stops updating. It is a bit easier to tell when this has occurred these days, since it is quite obvious something’s wrong if gweather thinks it is night time when it clearly isn’t.

The current code uses gnome-vfs, which isn’t the best choice for this sort of thing. The code is the usual mess you get when turning an algoithm inside out to work through callbacks in C:

  1. One function opens the URL with gnome_vfs_async_open().
  2. The callback that gets triggered on completion of the open calls gnome_vfs_async_read().
  3. The callback that gets triggered on the end of the read checks the status. If it is at the end of the stream, then process the data and close the stream. Otherwise, perform another read (which will loop back to this step).

This logic is repeated 5 times for the different weather data sources. To clean this up, I started looking at libsoup which doesn’t try to be a full file system abstraction, but provides a better API for the kind of things gweather does.

I put together a simple HttpResource class that wraps the relevant parts of libsoup for apps like gweather. It can be used like so:

  1. Create an HttpResource instance for the given URI.
  2. Connect a handler to the resource’s updated signal.
  3. Call the _set_update_interval() method to say how often the resource should be checked.
  4. Call the _refresh() method to kick off periodic freshness checks.
  5. When new data arrives, the updated signal is emitted.

Since the code is designed for periodic updates, I added some simple caching behaviour. If the server reports that the resource hasn’t been modified, we don’t need to emit the updated signal.

There are a few things that still need doing:

  • Some code to keep a SoupSession instance up to date with the proxy configuration settings in GConf.
  • Correct handling of the Expires: response header. If we are checking for updates every 30 minutes, but the server says the current weather report is current for the next hour, then we shouldn’t check again til then.
  • Support gzip and/or deflate content transfer encoding to reduce bandwidth.

This code should be pretty trivial to integrate into gweather when it is done, and should simplify the logic. I guess it would be useful for other applets too, such as gtik. The current code is available in my Bazaar archive:

baz get

Overriding Class Methods in Python

One of the features added back in Python 2.2 was class methods. These differ from traditional methods in the following ways:

  1. They can be called on both the class itself and instances of the class.
  2. Rather than binding to an instance, they bind to the class. This means that the first argument passed to the method is a class object rather than an instance.

For most intents and purposes, class methods are written the same way as normal instance methods. One place that things differ is overriding a class method in a subclass. The following simple example demonstrates the problem:

class SubClass(ParentClass):
    def create(cls, arg):
        ret = ParentClass.create(cls, arg)
        return ret

This code is broken because the ParentClass.create() call is calling the version of create() method in the context of ParentClass, rather than calling an unbound method like it would with a normal instance method. The most likely outcome will be a TypeError due to the method receiving too many arguments.

So how do you chain up to the parent class implementation? You use the super() object, which was also added in Python 2.2 as an alternative way to chain to the parent implementation of a method. The above code rewritten as follows:

class SubClass(ParentClass):
    def create(cls, arg):
        ret = super(SubClass, cls).create(arg)
        return ret

If you haven’t ever used the super() object, this is what it is doing in the above example:

  1. SubClass is looked up in the list cls.__mro__ (a linearised list of ancestor classes in the order used for method resolution).
  2. The class dict for each ancestor class coming after SubClass in cls.__mro__ is checked to see if it contains “create“.
  3. The super() object returns a version of “create” in the context of cls using the __get__(cls) “descriptor get” method.
  4. When this bound method gets called, cls will be passed in instead of the parent class.

Previously I’d ignored super() for the most part, since I could use the old chaining syntax. This shows a place where the old-style syntax can’t be applied.

pkg-config patches

I uploaded a few patches to the pkg-config bugzilla recently, which will hopefully make their way into the next release.

The first is related to bug 3097, which has to do with the broken dependent library elimination code added to 0.17.

The patch adds a Requires.private field to .pc files that contains a list of required packages like Requires currently does, which has the following properties:

  • When verifying that a particular package name is available with “pkg-config --exists“, dependencies in both Requires and Requires.private are checked.
  • When running “pkg-config --cflags“, flags from dependencies in Requires are included.
  • When running “pkg-config --libs“, flags from dependencies in Requires are included.
  • When running “pkg-config --static --libs“, flags from dependencies in both Requires and Requires.private are included.

The purpose of this is to list dependencies that are not exposed in the API of the library in question while not making users of the library link directly to those dependencies. This means that private dependencies can be upgraded to new incompatible versions without breaking applications that only depend on them indirectly.

This is intended for cases like Cairo, which links to libpng, but doesn’t expose any of the libpng API itself. It is not intended for dependencies like gtk+ depending on pango. Of course, this header will cause the .pc file to be incompatible with pkg-config versions prior to 0.16, because those versions don’t tolerate unknown fields.

The other changes are related to the associated autoconf macros:

  • Add a PKG_CHECK_EXISTS() macro. This would be similar to PKG_CHECK_MODULES(), except that no variables would be set or substitutes — it would simply run the ACTION-IF-FOUND or ACTION-IF-NOT-FOUND arguments. It is basically a less heavy weight macro for cases where you just want to see if a set of modules is available (bug 3530).
  • Get rid of the caching behaviour in PKG_CHECK_MODULES(). Since 0.16, this macro has cached the result of the check based on the variable prefix passed as the first argument. Since pkg-config is quite fast and configure doesn’t store its cache between runs by default, this doesn’t result in any noticable speed improvement and causes build problems for configure scripts that call PKG_CHECK_MODULES multiple times with the same variable name prefix but different package lists (e.g. Eye of Gnome). It seems simplest to just remove the caching, resulting in a simpler and more reliable macro (bug 3550, patch not yet uploaded).

With these changes, hopefully 0.18 will fix up the last few small incompatibilities in the recent releases.

Clipboard Handling

Phillip: your idea about direct client to client clipboard transfers is doable with the current X11 clipboard model:

  1. Clipboard owner advertises that it can convert selection to some special target type such as “client-to-client-transfer” or similar.
  2. If the pasting client supports client to client transfer, it can check the list of supported targets for the “client-to-client-transfer” target type and request conversion to that target.
  3. The clipboard owner returns a string containing details of how to request the data (e.g. hostname/port, or some other scheme that only works for the local host).
  4. Pasting application contacts the owner out of band and receives the data.

Yes, this requires modifications to applications in order to work correctly, but so would switching to a new clipboard architecture.

With respect to your no-transfer cut/paste of a movie example, that’s more of a component architecture problem than a clipboard issue. In the context of Bonobo, it can be done provided that the clipboard owner can provide the data as a Bonobo Embeddable, and the pasting application can embed Bonobo Embeddables in its documents:

  1. Clipboard owner advertises that it can convert the selection to the target “BONOBO_EMBEDDABLE” (or some other agreed upon targer name).
  2. Pasting application requests that the selection be converted to “BONOBO_EMBEDDABLE”, and receives an IOR for the component. Pasting application owns a reference on the component due to the clipboard transfer.
  3. Pasting application queryInterface()‘s the component to the Bonobo::ControlFactory interface, and calls the createControl() method to create a control to embed in the document.
  4. When it comes time to save the data, the component can be converted to one of the Bonobo::Persist interfaces, and written out.

Of course, there are reasons why people don’t do this (apart from not liking Bonobo), including:

  • With the classic X selection model, you don’t need to special case local or remote transfer cases.
  • Works in cases where the two applications can only communicate via the X connection (e.g. in the presence of transparent X proxies such as ssh).
  • It delegates all the permissions/authentication issues to the X server.

Anonymous voting

I put up a proposal for implementing anonymous voting for the foundation elections on the wiki. This is based in part on David’s earlier proposal, but simplifies some things based on the discussion on the list and fleshes out the implementation a bit more.

It doesn’t really add to the security of the elections process (doing so would require a stronger form of authentication than “can read a particular email account”), but does anonymise the election results and lets us do things like tell the voter that their completed ballot was malformed on submission.