[PROJ] Serious confusion regarding towgs or not...!

[Nyall Dawson]

On Mon, 16 Dec 2019 at 21:24, Even Rouault <even.rouault at spatialys.com> wrote:
>
> On lundi 16 décembre 2019 16:36:52 CET Nyall Dawson wrote:
> > Hi list,
> >
> > Been struggling through fixing some QGIS proj 6 related issues today,
> > and I've got to admit - I'm very confused about the role of towgs84 in
> > the proj 6 world.
> >
> > My current crop of questions:
> >
> > 1. Is towgs84 effectively deprecated? Should I be avoiding it as much
> > as possible? Is it inherently evil?
>
> It is not "evil", but the world of coordinate transformations is full of traps
> and headaches.
> It associates a transformation (among several potential ones) to WGS84 to a
> CRS. Once it is set, PROJ will honour it and will not try potentially better
> transformations that may use grids. So "+proj=longlat +ellps=GRS80
> +towgs84=,0,0,0" is a potential way of encoding boths GDA94 and GDA2020, but
> if you use it, then you'll get a null transformation between both.
>
> > 2. When I identify various WKT definitions using projinfo, I get a mix
> > between direct matches and matches to "BoundCRS of ...".
>
> I should warn that identification is a divination exercice / best effort /
> fuzzy logic, not something meant to be fully bullet proof.
>
> > I'm unsure
> > when this BoundCRS is important or not. Should I be treating a CRS
> > identified as a BoundCRS of ... as identical to the identified CRS,
> > and is it safe to just use the identified auth/code combo for that
> > CRS? Or is that losing important information and I should NEVER treat
> > a BoundCRS identified match as being an identified result in any
> > circumstance?
>
> I don't think there's an absolute answer. Depends on the context and the
> origin of that WKT. If it is a legacy WKT from GDAL < 3 & PROJ < 6 world, then
> you might just keep the Source CRS of the BoundCRS as the most significant
> part, and consider that the transformation to WGS84 captured in the
> ABRIDGEDTRANSFORMATION was not necessarily intended to be the one used.
>
> I see a number of tickets in PROJ tracker are files about several issues
> related to that. I'm not sure we'll ever manage to find the answer that
> satisfies both practicality and "correctness".

Hm. I'm now considering just dropping all code from QGIS which relates
to handling towgs and BoundCRS, and leaving the responsibility
entirely with GDAL and PROJ to insert or ignore TOWGS wherever it's
desirable/wanted.

>From a QGIS perspective I think the biggest downside of doing this
would be that many CRSes would be shown as "custom" user crses,
instead of via an auth/code identifier. But the alternative is that we
end up with a lot of logic relating to boundcrs downstream, and that
makes me nervous. So I'd be willing to live with the potential
mis-identification of known CRSes as custom CRS, and then deal with
each of those as upstream proj/gdal issues on a case-by-case basis.

But I'm keen to hear your thoughts on this approach...

Nyall