Arguments for Rectangle and RectangleMesh are inconsistent

[Patrick Farrell]

RectangleMesh takes in

RectangleMesh(double x0, double y0, double x1, double y1, ..)

Rectangle takes in

Rectangle(double x0, double x1, double y0, double y1)

Why?

[Anders Logg (Chalmers)]

I don't know why, but think the first one is more natural (specifying two corners that span the triangle).

[Patrick Farrell]

After reading it again, I think that is what the Rectangle constructor takes, just labelled very strangely:

    /// Create rectangle defined by two opposite corners
    /// x = (x0, x1) and y = (y0, y1).
    ///
    /// *Arguments*
    ///     x0 (double)
    ///         x0-coordinate of first corner.
    ///     x1 (double)
    ///         x1-coordinate of first corner.
    ///     y0 (double)
    ///         y0-coordinate of second corner.
    ///     y1 (double)
    ///         y1-coordinate of second corner.
    Rectangle(double x0, double x1, double y0, double y1);                                                                                                                                                          

When I saw (x0, x1, y0, y1), I thought {x,y} referred to the coordinate axis and {0,1} referred to the vertex. But the Rectangle constructor means it the other way around -- {x,y} refers to the vertex, and {0,1} to the coordinate axis!

[Anders Logg (Chalmers)]

True! This needs to be fixed. I think having (x0, y0) as the coordinates of vertex 0 is more intuitive. But I'm not sure this needs to be fixed in DOLFIN. This code is anyway being moved into mshr: https://bitbucket.org/benjamik/mshr

[Prof Garth Wells]

I suggested in the past that using tuples would be clearer:

mesh = RectangleMesh((x0, y0), (x1, y1))

[Anders Logg (Chalmers)]

That would be messy in C++.

[Prof Garth Wells]

@logg Doing

RectangleMesh mesh(Point(0.0, 0.0), Point(2.0, 1.0));

looks pretty tidy to me.

[Anders Logg (Chalmers)]

Maybe, but I still like RectangleMesh(x0, y0, x1, y1) better.

[Prof Garth Wells]

It's not a scalable solution. The actual interface is RectangleMesh(x0, y0, x1, y1, n0, n1), and in 3D it's BoxMesh(x0, y0, z0, x1, y1, z1, n0, n1, n2). There is plenty of scope for user error.

[Anders Logg (Chalmers)]

If so we should be consistent and insist on Point everywhere in the interface.

[Anders Logg (Chalmers)]

Update: the new version of Rectangle in mshr now has this interface:

Rectangle::Rectangle(dolfin::Point a, dolfin::Point b)

We should do the same in DOLFIN for RectangleMesh.

[Prof Garth Wells]

Yes, I like using dolfin::Point when it's a point.

[Anders Logg (Chalmers)]

I'll make a point of fixing this today.

[Anders Logg (Chalmers)]

Should be fixed now. Both RectangleMesh and BoxMesh now require Point, but IntervalMesh does not (would be consistent but a bit silly).

[Anders Logg (Chalmers)]

Fix issue #299: Use Point in RectangleMesh and BoxMesh interface

→ <<cset 5f692e51033e>>

[Martin Sandve Alnæs]

Removing milestone: 1.6 (automated comment)