[Qgis-developer] QGIS 3 plugins - require implementation as processing algs?

[Nyall Dawson]

Hi all,

Just something which has been on my mind and I thought I'd see if others agree:

Given that for QGIS 3.0 we will effectively start the available plugin
repo with a "clean slate", I'm wondering if it's a good time to put in
a restriction along the lines:

"If it COULD be made as a processing algorithm then it MUST be made as
a processing algorithm".

This is leading from Victor's excellent talk at Girona
(http://diobma.udg.edu/handle/10256.1/4300) But basically, there's
many reasons why we'd want this, including:

- consistent, robust and featureful UI for algorithms. Instead of
every plugin implementing its own UI with a different feel and feature
set, all processing algorithms have a consistent UI, which is well
tested and stable. This also makes things easier for plugin devs in
that they no longer have to waste time with UI, and they gain all the
extra features which are available in the processing UI (eg extent
selection with options from canvas, layers, etc) at no cost to
themselves.

- plugins become instantly more useful because they can be integrated
into models. This is better for users since these plugin algorithms
become more powerful. It's also better for plugin devs since their
plugins will get more use.

- easier path for valuable, widely used algorithms to become part of
core processing - if plugins are developed using the processing
algorithm framework then it becomes much easier for us to pull them
into the core qgis algorithms.

- better interface for QGIS. Instead of power users getting bogged
down with dozens of extra toolbar icons and menus for every plugin
they've installed they would instead be nicely and consistently
grouped into the processing toolbox.

(But seriously - watch Victor's presentation. He explains this all much better!)

Given this, should we require that for any plugin to be eligible to be
included in the official repo it MUST be implemented as a processing
algorithm IF it can be be fully implemented as an algorithm?

I personally can only see benefits to this approach, and timing it
with 3.0 (when people have to rewrite their plugins anyway) makes
sense.

Thoughts?

Nyall