Investigate failing testcase on macos 14 runners (arm64)

[veenstrajelmer]

What is the need for this task.
In issue #444 and the associated PR #636, the linux and macos testbenches were restored. However, in that process it was noticed the runner macos-latest results in one failing testcase: FAILED tests/dflowfm/test_net.py::test_create_1d_2d_1d2d

This is the plain HYDROLIB-core code, the test fails on assert network1_link1d2d.shape == (21, 2):

import numpy as np
import matplotlib.pyplot as plt
plt.close("all")
from meshkernel import GeometryList
from hydrolib.core.dflowfm.net.models import Branch, Network

def get_circle_gl(r, detail=100):

    t = np.r_[np.linspace(0, 2 * np.pi, detail), 0]
    polygon = GeometryList(np.cos(t) * r, np.sin(t) * r)
    return polygon

# Define line (spiral)
theta = np.arange(0.1, 20, 0.01)

y = np.sin(theta) * theta
x = np.cos(theta) * theta

dists = np.r_[0.0, np.cumsum(np.hypot(np.diff(x), np.diff(y)))]
dists = dists[np.arange(0, len(dists), 20)]

# Create branch
branch = Branch(geometry=np.stack([x, y], axis=1), branch_offsets=dists)

# Create Mesh1d
network = Network()
network.mesh1d_add_branch(branch, name="branch1")

# Add Mesh2d
network.mesh2d_create_rectilinear_within_extent(
    extent=(-22, -22, 22, 22), dx=2, dy=2
)
network.mesh2d_clip_mesh(geometrylist=get_circle_gl(22))

network.mesh2d_refine_mesh(polygon=get_circle_gl(11), level=1)
network.mesh2d_refine_mesh(polygon=get_circle_gl(3), level=1)

# Add links
network.link1d2d_from_1d_to_2d(branchids=["branch1"], polygon=get_circle_gl(19))

mesh2d_output = network._mesh2d.get_mesh2d()
assert len(mesh2d_output.face_x) == 152
mesh1d_output = network._mesh1d._get_mesh1d()
assert len(mesh1d_output.node_x) == 100

network1_link1d2d = network._link1d2d.link1d2d
network1_con_m1d = network._link1d2d.meshkernel.contacts_get().mesh1d_indices
network1_con_m2d = network._link1d2d.meshkernel.contacts_get().mesh2d_indices
assert network1_link1d2d.shape == (21, 2)
assert network1_con_m1d.size == 21
assert network1_con_m2d.size == 21

_, ax1 = plt.subplots()
network.plot(ax=ax1)

Plots this figure (on windows):

macos-latest results in macos-14 which has arm64 architecture. However, the testbank is green on macos-12 and macos-13, which seem to have x86_64 architecture (details in Deltares/MeshKernelPy#162). It seems that this specific part of the meshkernelpy code is not covered by meshkernel tests and it fails on arm64 and not on x86_64 architecture. It seems not a hydrolib-core issue, but this is still valuable information.

What is the task?
Check if this hypothesis is correct and add this testcase to the meshkernel testbench after removing hydrolib-core dependencies.

[veenstrajelmer]

This testcase can be reproduced also when all hydrolib-calls are excluded. Therefore, a meshkernelpy issue is created to add this test to the meshkernelpy testbank: Deltares/MeshKernelPy#164. After fixing that issue, hydrolib-core should also have a green testbank on macos-14/macos-latest and this os can be added in ci.yml