Because feature.geometry().geometry() is confusing, and impossible
to search for in python code (e.g. is input.geometry() a QgsGeometry
or a QgsAbstractGeometry?)
But more importantantly: also add a const version
QgsGeometry::constGet(). The non-const
version is slow, since it now forces a detach to avoid corrupting
geometries (since QgsGeometry is shared, it's not safe to directly
access its primitive QgsAbstractGeometry and start messing with
it without first detaching). This is a big risk in the 2.x API
which could potentially corrupt feature geometries with unexpected
outcomes.
Update all uses to constGet where possible.
This is potentially interesting for race conditions, but whenever it is specified,
iterators will need to be carefully checked for validity and error handling needs
to be taken care of.
By moving logic to abstract geometry subclasses so that they
can trivially retrieve the vertex number, instead of relying
on the inefficient coordinateSequence method.
Has flow on speed boosts to many geometry operations like
calculation of closest points in a geometry, which will
benefit snapping related operations.
Previously the method in QgsGeometryUtils was relying on
QgsAbstractGeometry::coordinateSequence, which is an absolute
performance killer.
Instead move to optimised methods in the various abstract
geometry subclasses which rely only on trivial calculations.
and add new QgsGeometry::fromPolyline which uses QgsPoint
We want to encourage people not to use the QgsPointXY method, as it drops
Z/M values. So it's moved across to a different name to make way
for a new QgsGeometry::fromPolyline which uses a QgsPoint list
instead of QgsPointXY, thus keeping Z/M values intact.
Similarly, QgsPolyline now is a list of QgsPoint (keeping z/m values)
and the old 2d QgsPolyline type was renamed to QgsPolylineXY.
Making the QgsPoint method the "preferred" method and making linestrings
just as easy to create from z/m dimensioned points as 2d points is
important to push people to write code which does not discard
these important dimensions.
As a bonus, the QgsPoint methods are more efficient anyway, since
they don't require creation of a temporary list.
