Floating point errors in `.align_to()`

Describe the bug The below code:

import sectionproperties.pre.library.bridge_sections as bridge_sections
import sectionproperties.pre.library.primitive_sections as standard_sections
from sectionproperties.pre.pre import Material
from sectionproperties.analysis.section import Section
from sectionproperties.pre.geometry import CompoundGeometry

Dslab, w, t_f = 180, 2100, 75

precast = Material(
    name="65 MPa",
    elastic_modulus=37.4e3,
    poissons_ratio=0.2,
    yield_strength=65,
    density=2.4e-6,
    color="grey",
)
insitu = Material(
    name="40 MPa",
    elastic_modulus=32.8e3,
    poissons_ratio=0.2,
    yield_strength=40,
    density=2.4e-6,
    color="lightgrey",
)

super_t = bridge_sections.super_t_girder_section(girder_type=5, w=w, material=precast)
slab = standard_sections.rectangular_section(d=Dslab, b=w, material=insitu).align_center(super_t).align_to(super_t, "top")

geom = CompoundGeometry([super_t, slab])
geom.plot_geometry()
geom

Produces this…

But it SHOULD produce this…(Note the red svg denoting “invalid” geometry and the hole indicated on the plot)

Additional context This seems to be a problem with shapely. I opened an issue with shapely after realizing that one of their examples in their docs does not reproduce as given (another example of the same problem): https://shapely.readthedocs.io/en/stable/manual.html#polygons

exter = [(0, 0), (0, 2), (2, 2), (2, 0), (0, 0)]
inter = [(1, 0), (0, 1), (0.5, 1.5), (1.5, 0.5), (1, 0)]
(Polygon(exter) - Polygon(inter)) # Creates valid MultiPolygon even though two points in common

Produces this…

Something else that’s funny (odd, not ‘ha ha’)

If I take the geom object that was created in the bridge section code (at top) and break it out with .points as such:

geom = CompoundGeometry(MultiPolygon([Polygon(slab.points), Polygon(super_t.points)]))

# Instead of...

geom = CompoundGeometry([slab, super_t])

Then I also get this…

I don’t fully understand but I think it is shapely related… Will post more on this as I learn more.

[connorferster]

Gah! Simple solution: round the resulting coordinates, not the offset. <eyeroll>

In Geometry.compile_geometry(), which generates the .points, .facets, etc. the coordinates are already being rounded to self.tol which is set to 12 decimal places. This is why the geometry can be created without holes but when using the generated points in a new geometry with .from_points(), we get what we want.

Totally forgot about that tolerance thing in the creation of .points. Going to re-order Geometry.__init__() so that the shapely object is exactly the same as the generated points because, right now, there is a a slight difference between the geometry you see and the points that are generated from it.

[connorferster]

@robbievanleeuwen

So, it seems that just rounding the offset some decimal places does not work: it seems like it can create a small overlap between two geometries which causes triangle to spiral out of control.

I think that numpy.clip could hold a solution but it will be a bit tricky to implement. Started playing with it this evening. Will take some thinking.