[QGIS-Developer] Porting Create Grid algorithm to C++

[Nyall Dawson]

On Mon, 30 Sep 2019 at 03:06, Clemens Raffler <clemens.raffler at gmail.com> wrote:
>
> Dear QGIS Developers,
>
> lately I have been using the 'Create Grid' algorithm a lot on huge areas
> and with a rather fine grid resolution. I was not surprised that
> creating a very fine grid took a very long time as the algorithm is
> currently implemented in python.
>
> In order to speed up my work I decided to experiment a little with C++
> (coming from a python background) development in QGIS and ported the
> algorithm to C++ based on other native algorithms. The code is working
> very well, the grid creation process has been sped up significantly. I
> would now like to contribute the code to the QGIS project so that other
> users can benefit from the speedup.

Sounds fantastic, and a very valuable (and desired) contribution! Nice work!

> - Is there any reason why certain algorithms (eg. the Create Grid alg)
> should not be available in C++?

No. In fact, the opposite! The long-term goal is to port all these
algorithms away from Python and to c++. Python is GREAT for plugin
based algorithms and for quickly developing new algorithms, but it's
just not competitive in speed vs the c++ algorithms. We've also had a
LOT of issues in the past where Python algorithms break due to lack of
compile time checks when QGIS API (or other Processing Python code) is
revised. In contrast, I'm not aware of a single regression of this
nature in any of the ported c++ algorithms.

(My advice for readers out there: don't write big, complex
applications in Python. full stop.)

> - As I am not familiar with the processes how native processing
> algorithms get committed to the main repository (pull requests,
> prepare-commit.sh, sip, etc.), would anybody like to help/mentor/guide
> me through this process and review the code?

**raises hand**

> - Are there any special requirements on testing infrastructure for C++
> algs which should be implemented too (I have not worked with tests so far)?

Looks like there's already good test coverage with the existing Python
algorithm. These existing tests will carry over to the c++ port, so
it's likely that nothing extra is required here.

In general, there's some "desirable changes" which we are encouraging
when porting an existing algorithm to c++, but none of them apply in
this particular case. Specifically:

- use the QgsProcessingFeatureBasedAlgorithm subclass wherever possible.
- add support for dynamic parameters where desirable.


> So far, my experience with porting python algs to C++ has been very good
> with Create Grid and I would like to work on other ports if everything
> goes well.

That's great to hear!  I'd suggest having a look at a really simple
feature based algorithm like DensifyGeometries as a great candidate
for learning. There's almost no logic in the Python code for this one,
so it would be a great one to learn the overall process of porting to
c++ without the complexities of having to port a lot of
python-specific logic.

Nyall