GeezerVison

The name of this file is problematic. It breaks the Qt3 backend (since 'qt' is the name of the PyQt library version 3 -- the backend tries to import the bindings but instead gets this file.) It's also confusing to read "from qt import ..." in the Qt4 backends and not have that refer to the main qt library.

Perhaps it should be called "qt4_compat.py" or something similar?

Good point. I just blindly copied the file from IPython but they don't support Qt3.

With ubuntu 11.4 (pyside 1.0.1 & qt4 4.7.2), I get a following exception.

TypeError: 'PySide.QtGui.QImage' called with wrong argument types:
PySide.QtGui.QImage(str, int, int, PySide.QtGui.QImage.Format)
Supported signatures:
PySide.QtGui.QImage()
PySide.QtGui.QImage(PySide.QtGui.QImage)
PySide.QtGui.QImage(PySide.QtCore.QSize, PySide.QtGui.QImage.Format)
PySide.QtGui.QImage(QString, str = None)
PySide.QtGui.QImage(int, int, PySide.QtGui.QImage.Format)
PySide.QtGui.QImage(buffer, int, int, PySide.QtGui.QImage.Format)
PySide.QtGui.QImage(buffer, int, int, int, PySide.QtGui.QImage.Format)

Converting stringBuffer to a buffer object explicitly seem to solve the problem (both pyqt and pyside works), but i'm not sure if this behavior will depend on the pyside version.

        qImage = QtGui.QImage(buffer(stringBuffer), int(self.renderer.width), 
                              int(self.renderer.height),

We may need a same change at line 117.
The agg renderer has a "buffer_rgba" method but QImage does not seem to support RGBA format.

-JJ

This is a PySide bug fixed in 1.0.2 http://bugs.pyside.org/show_bug.cgi?id=489

We should probably add a version check against PySide then, since we know 1.0.2 is the minimum workable version.

Good to know that this is a pyside bug. However, using explicit buffer call does any harm to matplotib w/ pyside v1.0.2?
I could be wrong but my guess is that if we call buffer explicitly, the code will work for both v1.0.2 and v1.0.1?

Yes. Wrapping with buffer makes it work in both versions. I was under the impression that mpl wasn't in the business of fixing bugs for specific versions of on specific platforms etc. In any case, there's another bug (819, see my opening comment) that wasn't fixed until 1.0.2 (try saving a figure). There is also an outstanding bug that prevents this backend working on mac (809) that won't be fixed until 1.0.3.

I'd suggest that 1.0.3 should probably be the minimum supported version to keep things simple. In which case this can stay as is. Unless there is a good reason to want to wrap with buffer() in any case?

In my experience, mpl tries to broadly support as many versions of a library on as many platforms as feasible. I don't know if it's stated explicitly anywhere, but at least in my work environment, it is very important to support rather old and possibly buggy versions of core dependencies (e.g. GUI toolkits) for quite some time. Look at the GTK backend, for example, for a number of version-specific hacks that have been required over time. Eventually the idea is to prune that stuff out -- my own rule of thumb is to support at least as far back as whatever was in the (current release - 2) of RHEL.

In this specific case, however, with PySide being so new and cutting edge to begin with, I think it's ok to require a minimum version and not add hacks for 1.0.x which were fairly ephemeral and not widely-used releases. But the version check is important because user's should understand it's breaking because of an old version of PySide and not some other reason.

I was inclined to fix the buffer thing since pyside v1.0.1 is packaged for ubuntu, and this simple fix at least makes the backend usable. Anyhow, I agree with Michael and the current approach is okay.

Ok, that's good then.

As a side note, I was under the impression that Ubuntu was more or less the development platform for PySide. Every version of PySide is packaged for it within a few hours of release isn't it?