[QGIS-Developer] OGC API Features provider

[Even Rouault]

> I would probably advise against this approach, from the architectural point
> of view, I think that it would be better to address the new OGC JSON-based
> API with an abstract base class for providers that consume that kind of API
> and make WFS3 the first concrete implementation of that base.

While I understand this idea, I see several difficulties:
- the candidate providers are the feature one, a coverage and a tile / map 
one, but vector and raster providers have little in common
- at the time of writing, to the best of my knowledge, only the OAPI-F has 
reached a sufficient degree of maturity in its spec and initial 
implementations. In particular while I think there is an intention to have a 
OAPI-Common at some stage, I don't think this has emerged yet (the current 
github repo for it is a copy&paste of OAPI-F), so making an abstraction with 
just a single implementation is not going to work well at that stage.
- if looking at the existing providers in QGIS, the arcgisrest ones seem to be 
the closest to what you mention above. From what I see, the code between the 
AMS and AFS is quite separate, likely for the reasons I mentionned in my first 
point. They do have some utility functions qgsarcgisrestutils.h in common, so 
that would be more the kind of communality I can imagine if a OAPI-Tile/Map 
provider is later added. Perhaps things like parsing service metadata.

Regarding the UI, I'm not completely sure of the best option. I can understand 
the opinion I got that reusing the WFS UI could be better because people are 
familiar with it, and that would limit the number of provider entries (people 
just have to figure out if the URL they are provided with is a WFS, OAPI-F or 
ArcGIS REST Feature one...). It can also makes sense if the OAPI-F stuff is 
added to the existing WFS provider.
The other point of view, having a dedicated UI & provider, has also its 
advantages in term of code clarity (but provided the point below can be 
solved) 

> Re-using the WFS-sqlite feature cache looks a good idea, you might want to
> refactor it out to core to make it re-usable from a family of providers.

I haven't completely given up on trying that idea. Just scares me :-)

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com