[Qgis-developer] [Qgis-psc] Clarification of outcome of PSC meeting 16/1

[Matthias Kuhn]

Hi all,

Sorry for the long mail but I think it is an important topic.

Increasing the major version number is a rather big step and should be
considered carefully. So before going into too many technical details
(no worries, you find them below) I think it is important to have a
clear idea why we would want that and what we expect from this move. In
order to avoid another troublesome major version change (QGIS 4) for as
long as possible.
Qt 5 and Python 3 are IMO mandatory changes when QGIS 3 comes but they
should not be the cause for it but rather dependencies of it.

I thought that it would be a good idea to use the QEP process to handle
this. We could create a label QGIS3 there which we could apply to QEPs
that will require QGIS3. And another QEP specifically for QGIS 3, that
lists other things to do (cleanup of deprecated API, Qt 5, Python 3...).

Some ideas what could be labelled with QGIS3:

 * Composer -> Layout
 * Database aware datasources (Make QGIS aware of
datasources(providers/tables) on the same database. That would allow to
better handle database dependent properties like foreign keys (which
involve multiple tables).
 * Abstraction of the .qgs file structure. Now it depends on XML. A new
project format including portable layers (memory saver) may use
something different. Or stick to XML.
 * Geometry redesign (It's a rather big rework, I guess there will be
some implications on the API leading to different behavior. I don't know
it in detail and the roadmap)
 * [Your idea goes here]

And now to the technical notes:
I think that both Qt 5 and Python 3 are ready to be used. They are
available for almost any platform and are can be considered stable. If
we switch the major version number in QGIS to 3, they should be made a
highly recommended dependency and developers should treat them as the
main target.

At least for Qt5 I can say that it is compatible to Qt4 to a very high
degree. I think also for python it is possible to write code that works
with python 2.7 and 3.x (The pandas module which is quite complex is
compatible with both). That means, it should be possible to make plugins
work independent of these versions.

There is one thing however that causes me some headaches which is PyQt5.
This must not be set equal with Qt5. It is possible to combine Qt5 with
PyQt4 (Read that again!) [1]. With this combination all the above
compatibility notes (should) apply.
Upgrading to PyQt5 however comes with some caveats. PyQt5 removes all
the deprecated things from Qt. That may affect some plugins. Even worse,
it forces python plugins to change their imports. Instead of "import
PyQt4" one will have to write "import PyQt5". Basically rendering all
plugins incompatible.

So we could consider to switch to Qt5 and keep PyQt4 and have an easy
transition. However, I hardly doubt that every platform will offer this
combination. Most distributions will probably do the obvious and bundle
PyQt4 linked against Qt4 and PyQt5 linked against Qt5.

This leaves us with the options:

 1. Include PyQt4 in our sources (you may deduce my stance to this from
the discussion here [2])
 2. Switch to PyQt5 and force all plugins to update

Version 2 would render all plugins subsequently incompatible with QGIS 2
what would probably disappoint many. So I thought that python may be
flexible enough  to offer aliases for modules (is it, python experts?).
So would it be possible to alias both - PyQt4 and PyQt5 - with PyQt?
That way plugins could be updated with an easy search-and-replace to
just import PyQt (and check for PyQt.version if required) and still have
them work with QGIS 2 and QGIS 3.

Regards,
Matthias

[1] http://pyqt.sourceforge.net/Docs/PyQt5/pyqt4_differences.html
[2] https://github.com/qgis/QGIS/pull/1726


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://lists.osgeo.org/pipermail/qgis-developer/attachments/20150117/f563b89d/attachment.pgp>