[Qgis-developer] Composer... current status and the way forward?

[Nyall Dawson]

Hi all,

I'm seeking feedback about the best way to move forward with QGIS' map
composer. I'm currently running up against some large issues with the
current design and API of composer which are holding back important
features and fixes. Some of these issues include:

- there's too much composer logic tied up in app and gui. This makes
it very difficult for plugins to manipulate and export compositions
without duplicating large blocks of code
- there's too much item-specific logic and handling scattered through
QgsComposition, QgsComposerView and QgsComposer. This makes it
impossible to have features like plugin generated item types, and
makes maintenance difficult.
- everything is coded to expect measurements and sizes in mm. I can't
(nicely) add support for other units without breaking api or resorting
to a lot of hacks
- same for mixed page sizes and orientations within a single
composition, this requires an api break to implement cleanly
- I need to totally break composer api in order to fix the instability
in undo/redo commands (see http://hub.qgis.org/issues/11371)
- QgsComposition should not require a QgsMapSettings/QgsMapRenderer.
This should instead be set individually for map items. Doing so would
pave the way for features such as reprojection support for individual
map items.
- the composer is full is deprecated methods and legacy api

I've slowly come to the conclusion that the way forward is to move to
a bunch of new classes, much like what was done with symbologyV2. If
https://github.com/qgis/QGIS-Enhancement-Proposals/pull/9 passes then
these would be named QgsLayout, QgsLayoutDesigner, etc. If not, well,
I'll have to resort to QgsCompositionV2, etc.

The potential problem with this approach is how to handle the GUI and
existing projects. As far as I can see, there's a few options:
1. Expose both the existing composer and the new layout designer to
users. Composers aren't automatically upgraded to layouts. This
approach means that existing PyQgis code and plugins will still
function for existing projects, but at the expense of a confusing
experience for users.
2. Add all the new layout classes and keep the existing composer
classes. Composer would NOT be exposed in the GUI and compositions are
upgraded to layouts when projects are opened. This approach means that
standalone python code would still operate, but plugins or code which
are designed to be run from within QGIS would no longer function.
3. Move totally to the new layout classes and remove all composer
classes (unlikely)

I'm leaning toward option 2, but what are you thoughts? What's the
best approach to move forward? Obviously I'll submit all this as a QEP
when the plans are finalised, but for now I'm just after advice on the
preferred approach.

Nyall