[QGIS-Developer] Should we disable Processing modeler for the next 3.4?

[Nyall Dawson]

On Tue, 11 Sep 2018 at 03:11, Luigi Pirelli <luipir at gmail.com> wrote:
>

>  Should we disable Processing modeler for the next 3.4?

Hell no! (no wait, that's not strong enough...! #$%#$% no!!!) ;)

> I'm having really unreliable experience with processing modeler and grass commands in master. Other have the same experience?

I don't have any direct experience with models using grass, so would
be keen to hear other's feedback here. My experience is generally with
using native and Python provider algorithms alone, in which case
modeler works well.*

* (Using modeler in QGIS has always been a balancing act of
frustration... I find that I'm always getting 95% of my model
implemented, before hitting some missing feature in modeler which
would allow me to do that final 5%. Maybe over time this has increased
from 80% to 95%, but we're still not at 100% yet! When all parameter
widgets are ported to the new modeler widget api this will get my
models to maybe 96%, with another 3% bump when full expression
contexts are available to modeler parameters in the "precalculated
expression" mode. But that's all feature requests anyway, not
bugs/regressions!).

> some problems has been already reported: https://issues.qgis.org/issues/19700

I do expect some short-term fallout from
https://github.com/qgis/QGIS/pull/7761, but we should easily be able
to get this in shape for 3.4 final. And the benefit of #7761 is that
each parameter which is ported is gaining 100% unit test coverage,
something widget wrappers have never had in the past. Parameters
ported so far are string, boolean and CRS parameters (the rest will
need to wait for 3.6).

> IMHO there are two big problems
> 1) lack of test for algs in modeler... kind of integration tests whilst other part of processing are well covered.

That's a good point, but conceptually there shouldn't be any
difference to an algorithm running in modeler vs outside of it. I
think the issue with grass is more to do with "running more than one
GRASS algorithm in a single qgis session".

> 2) There are not so many users moving models to 3.x => many use cases are not covered yet.

Yes, many use cases are definitely not covered yet. But I don't think
these are regressions, just missing features/functionality  (let me
know if they are regressions).

Nyall