[IPython-dev] Qt api selection re. ipython and matplotlib

[MinRK]

On Mon, Jul 18, 2011 at 11:27, Darren Dale <dsdale24 at gmail.com> wrote:
> On Mon, Jul 18, 2011 at 2:16 PM, MinRK <benjaminrk at gmail.com> wrote:
>> On Mon, Jul 18, 2011 at 10:26, Darren Dale <dsdale24 at gmail.com> wrote:
>>> On Wed, Jul 6, 2011 at 4:23 PM, Eric Firing <efiring at hawaii.edu> wrote:
>>>> On 07/06/2011 05:34 AM, MinRK wrote:
>>>>> I updated my IPython PR with my understanding of the recent discussion.
>>>>>
>>>>> Now, QT_API will be asked *first*, and if QT_API=pyqt, IPython will
>>>>> set the v2 APIs.
>>>>>
>>>>> QT_API=pyqt : use pyqt with v2
>>>>> QT_API=pyside : use pyside
>>>>> QT_API unset : ask matplotlib for pyqt4 or pyside, will not set v2 APIs
>>>>>
>>>>> The default, and most likely case remains PyQt4 without updating sip
>>>>> apis, falling back on PySide.
>>>>
>>>> I have updated my mpl PR in a compatible way; the only difference is
>>>> that I have not included a fall-back to pyside.  Either PyQt4 or PySide
>>>> is specified as an rcParam, and if whatever is specified (either
>>>> explicitly there or via the QT_API env var) is not there, an ImportError
>>>> will result.  Anyone who has only PySide will have to know about it and
>>>> use the rcParam or QT_API to request it.
>>>
>>> Unfortunately, I think we may have to reopen this discussion. As the
>>> master branch currently stands, the embedding_in_qt4.py example is
>>> broken when I have the QT_API environment variable defined. The reason
>>> is simple: the example imports from PyQt4 first, which defaults to the
>>> v1 api, then the example imports the qt4 backend, which fails in its
>>> attempt to set the api to v2. One workaround is to perform the mpl qt4
>>> backend import before the PyQt4 imports, but such a requirement could
>>> be disruptive to lots of existing code. A better alternative might be
>>> to wrap the setapi calls in a try/except:
>>>
>>>    import sip
>>>    if QT_API == QT_API_PYQTv2:
>>>        try:
>>>            sip.setapi('QString', 2)
>>>            sip.setapi('QVariant', 2)
>>>        except ValueError:
>>>            pass
>>
>> QT_API=pyqt *does* imply PyQt4+v2api, since it is an ETS variable, and
>> that's how they use it.
>
> I haven't argued otherwise.
>
>> Also note that this variable is only likely
>> to be set in cases where users have both PySide and PyQt4, and the
>> default is behavior is *not* what the user wants.  That said, a
>> warning rather than raising is perhaps preferable, to handle users who
>> have ETS applications, for which they use QT_API, but also other Qt
>> programs that want to behave differently.
>
> This was my point.

Sorry, I guess it wasn't clear that I was agreeing with you, I only
wanted to note that this is going to be a pretty rare case.  I think
your solution, simply replacing 'pass' with a warning is the way to
go.