scipy 1.16.2__cp312-cp312-win_arm64.whl

scipy/interpolate/tests/test_bsplines.py

@@ -0,0 +1,3754 @@
+import os
+import operator
+import itertools
+import math
+import threading
+import copy
+
+import numpy as np
+from numpy.testing import suppress_warnings
+from scipy._lib._array_api import xp_assert_equal, xp_assert_close
+from pytest import raises as assert_raises
+import pytest
+
+from scipy.interpolate import (
+        BSpline, BPoly, PPoly, make_interp_spline, make_lsq_spline,
+        splev, splrep, splprep, splder, splantider, sproot, splint, insert,
+        CubicSpline, NdBSpline, make_smoothing_spline, RegularGridInterpolator,
+)
+import scipy.linalg as sl
+import scipy.sparse.linalg as ssl
+
+from scipy.interpolate._bsplines import (_not_a_knot, _augknt,
+                                        _woodbury_algorithm, _periodic_knots,
+                                         _make_interp_per_full_matr)
+
+from scipy.interpolate import generate_knots, make_splrep, make_splprep
+
+import scipy.interpolate._fitpack_impl as _impl
+from scipy._lib._util import AxisError
+from scipy._lib._testutils import _run_concurrent_barrier
+
+# XXX: move to the interpolate namespace
+from scipy.interpolate._ndbspline import make_ndbspl
+
+from scipy.interpolate import _dfitpack as dfitpack
+from scipy.interpolate import _bsplines as _b
+from scipy.interpolate import _dierckx
+
+
+class TestBSpline:
+
+    def test_ctor(self):
+        # knots should be an ordered 1-D array of finite real numbers
+        assert_raises((TypeError, ValueError), BSpline,
+                **dict(t=[1, 1.j], c=[1.], k=0))
+        with np.errstate(invalid='ignore'):
+            assert_raises(ValueError, BSpline, **dict(t=[1, np.nan], c=[1.], k=0))
+        assert_raises(ValueError, BSpline, **dict(t=[1, np.inf], c=[1.], k=0))
+        assert_raises(ValueError, BSpline, **dict(t=[1, -1], c=[1.], k=0))
+        assert_raises(ValueError, BSpline, **dict(t=[[1], [1]], c=[1.], k=0))
+
+        # for n+k+1 knots and degree k need at least n coefficients
+        assert_raises(ValueError, BSpline, **dict(t=[0, 1, 2], c=[1], k=0))
+        assert_raises(ValueError, BSpline,
+                **dict(t=[0, 1, 2, 3, 4], c=[1., 1.], k=2))
+
+        # non-integer orders
+        assert_raises(TypeError, BSpline,
+                **dict(t=[0., 0., 1., 2., 3., 4.], c=[1., 1., 1.], k="cubic"))
+        assert_raises(TypeError, BSpline,
+                **dict(t=[0., 0., 1., 2., 3., 4.], c=[1., 1., 1.], k=2.5))
+
+        # basic interval cannot have measure zero (here: [1..1])
+        assert_raises(ValueError, BSpline,
+                **dict(t=[0., 0, 1, 1, 2, 3], c=[1., 1, 1], k=2))
+
+        # tck vs self.tck
+        n, k = 11, 3
+        t = np.arange(n+k+1, dtype=np.float64)
+        c = np.random.random(n)
+        b = BSpline(t, c, k)
+
+        xp_assert_close(t, b.t)
+        xp_assert_close(c, b.c)
+        assert k == b.k
+
+    def test_tck(self):
+        b = _make_random_spline()
+        tck = b.tck
+
+        xp_assert_close(b.t, tck[0], atol=1e-15, rtol=1e-15)
+        xp_assert_close(b.c, tck[1], atol=1e-15, rtol=1e-15)
+        assert b.k == tck[2]
+
+        # b.tck is read-only
+        with pytest.raises(AttributeError):
+            b.tck = 'foo'
+
+    def test_degree_0(self):
+        xx = np.linspace(0, 1, 10)
+
+        b = BSpline(t=[0, 1], c=[3.], k=0)
+        xp_assert_close(b(xx), np.ones_like(xx) * 3.0)
+
+        b = BSpline(t=[0, 0.35, 1], c=[3, 4], k=0)
+        xp_assert_close(b(xx), np.where(xx < 0.35, 3.0, 4.0))
+
+    def test_degree_1(self):
+        t = [0, 1, 2, 3, 4]
+        c = [1, 2, 3]
+        k = 1
+        b = BSpline(t, c, k)
+
+        x = np.linspace(1, 3, 50)
+        xp_assert_close(c[0]*B_012(x) + c[1]*B_012(x-1) + c[2]*B_012(x-2),
+                        b(x), atol=1e-14)
+        xp_assert_close(splev(x, (t, c, k)), b(x), atol=1e-14)
+
+    def test_bernstein(self):
+        # a special knot vector: Bernstein polynomials
+        k = 3
+        t = np.asarray([0]*(k+1) + [1]*(k+1))
+        c = np.asarray([1., 2., 3., 4.])
+        bp = BPoly(c.reshape(-1, 1), [0, 1])
+        bspl = BSpline(t, c, k)
+
+        xx = np.linspace(-1., 2., 10)
+        xp_assert_close(bp(xx, extrapolate=True),
+                        bspl(xx, extrapolate=True), atol=1e-14)
+        xp_assert_close(splev(xx, (t, c, k)),
+                        bspl(xx), atol=1e-14)
+
+    def test_rndm_naive_eval(self):
+        # test random coefficient spline *on the base interval*,
+        # t[k] <= x < t[-k-1]
+        b = _make_random_spline()
+        t, c, k = b.tck
+        xx = np.linspace(t[k], t[-k-1], 50)
+        y_b = b(xx)
+
+        y_n = [_naive_eval(x, t, c, k) for x in xx]
+        xp_assert_close(y_b, y_n, atol=1e-14)
+
+        y_n2 = [_naive_eval_2(x, t, c, k) for x in xx]
+        xp_assert_close(y_b, y_n2, atol=1e-14)
+
+    def test_rndm_splev(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        xx = np.linspace(t[k], t[-k-1], 50)
+        xp_assert_close(b(xx), splev(xx, (t, c, k)), atol=1e-14)
+
+    def test_rndm_splrep(self):
+        rng = np.random.RandomState(1234)
+        x = np.sort(rng.random(20))
+        y = rng.random(20)
+
+        tck = splrep(x, y)
+        b = BSpline(*tck)
+
+        t, k = b.t, b.k
+        xx = np.linspace(t[k], t[-k-1], 80)
+        xp_assert_close(b(xx), splev(xx, tck), atol=1e-14)
+
+    def test_rndm_unity(self):
+        b = _make_random_spline()
+        b.c = np.ones_like(b.c)
+        xx = np.linspace(b.t[b.k], b.t[-b.k-1], 100)
+        xp_assert_close(b(xx), np.ones_like(xx))
+
+    def test_vectorization(self):
+        rng = np.random.RandomState(1234)
+        n, k = 22, 3
+        t = np.sort(rng.random(n))
+        c = rng.random(size=(n, 6, 7))
+        b = BSpline(t, c, k)
+        tm, tp = t[k], t[-k-1]
+        xx = tm + (tp - tm) * rng.random((3, 4, 5))
+        assert b(xx).shape == (3, 4, 5, 6, 7)
+
+    def test_len_c(self):
+        # for n+k+1 knots, only first n coefs are used.
+        # and BTW this is consistent with FITPACK
+        rng = np.random.RandomState(1234)
+        n, k = 33, 3
+        t = np.sort(rng.random(n+k+1))
+        c = rng.random(n)
+
+        # pad coefficients with random garbage
+        c_pad = np.r_[c, rng.random(k+1)]
+
+        b, b_pad = BSpline(t, c, k), BSpline(t, c_pad, k)
+
+        dt = t[-1] - t[0]
+        xx = np.linspace(t[0] - dt, t[-1] + dt, 50)
+        xp_assert_close(b(xx), b_pad(xx), atol=1e-14)
+        xp_assert_close(b(xx), splev(xx, (t, c, k)), atol=1e-14)
+        xp_assert_close(b(xx), splev(xx, (t, c_pad, k)), atol=1e-14)
+
+    def test_endpoints(self, num_parallel_threads):
+        # base interval is closed
+        b = _make_random_spline()
+        t, _, k = b.tck
+        tm, tp = t[k], t[-k-1]
+        # atol = 1e-9 if num_parallel_threads == 1 else 1e-7
+        for extrap in (True, False):
+            xp_assert_close(b([tm, tp], extrap),
+                            b([tm + 1e-10, tp - 1e-10], extrap), atol=1e-9, rtol=1e-7)
+
+    def test_continuity(self, num_parallel_threads):
+        # assert continuity at internal knots
+        b = _make_random_spline()
+        t, _, k = b.tck
+        xp_assert_close(b(t[k+1:-k-1] - 1e-10), b(t[k+1:-k-1] + 1e-10),
+                atol=1e-9)
+
+    def test_extrap(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        dt = t[-1] - t[0]
+        xx = np.linspace(t[k] - dt, t[-k-1] + dt, 50)
+        mask = (t[k] < xx) & (xx < t[-k-1])
+
+        # extrap has no effect within the base interval
+        xp_assert_close(b(xx[mask], extrapolate=True),
+                        b(xx[mask], extrapolate=False))
+
+        # extrapolated values agree with FITPACK
+        xp_assert_close(b(xx, extrapolate=True),
+                splev(xx, (t, c, k), ext=0))
+
+    def test_default_extrap(self):
+        # BSpline defaults to extrapolate=True
+        b = _make_random_spline()
+        t, _, k = b.tck
+        xx = [t[0] - 1, t[-1] + 1]
+        yy = b(xx)
+        assert not np.all(np.isnan(yy))
+
+    def test_periodic_extrap(self):
+        rng = np.random.RandomState(1234)
+        t = np.sort(rng.random(8))
+        c = rng.random(4)
+        k = 3
+        b = BSpline(t, c, k, extrapolate='periodic')
+        n = t.size - (k + 1)
+
+        dt = t[-1] - t[0]
+        xx = np.linspace(t[k] - dt, t[n] + dt, 50)
+        xy = t[k] + (xx - t[k]) % (t[n] - t[k])
+        xp_assert_close(b(xx), splev(xy, (t, c, k)))
+
+        # Direct check
+        xx = [-1, 0, 0.5, 1]
+        xy = t[k] + (xx - t[k]) % (t[n] - t[k])
+        xp_assert_equal(b(xx, extrapolate='periodic'), b(xy, extrapolate=True))
+
+    def test_ppoly(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        pp = PPoly.from_spline((t, c, k))
+
+        xx = np.linspace(t[k], t[-k], 100)
+        xp_assert_close(b(xx), pp(xx), atol=1e-14, rtol=1e-14)
+
+    def test_derivative_rndm(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        xx = np.linspace(t[0], t[-1], 50)
+        xx = np.r_[xx, t]
+
+        for der in range(1, k+1):
+            yd = splev(xx, (t, c, k), der=der)
+            xp_assert_close(yd, b(xx, nu=der), atol=1e-14)
+
+        # higher derivatives all vanish
+        xp_assert_close(b(xx, nu=k+1), np.zeros_like(xx), atol=1e-14)
+
+    def test_derivative_jumps(self):
+        # example from de Boor, Chap IX, example (24)
+        # NB: knots augmented & corresp coefs are zeroed out
+        # in agreement with the convention (29)
+        k = 2
+        t = [-1, -1, 0, 1, 1, 3, 4, 6, 6, 6, 7, 7]
+        rng = np.random.RandomState(1234)
+        c = np.r_[0, 0, rng.random(5), 0, 0]
+        b = BSpline(t, c, k)
+
+        # b is continuous at x != 6 (triple knot)
+        x = np.asarray([1, 3, 4, 6])
+        xp_assert_close(b(x[x != 6] - 1e-10),
+                        b(x[x != 6] + 1e-10))
+        assert not np.allclose(b(6.-1e-10), b(6+1e-10))
+
+        # 1st derivative jumps at double knots, 1 & 6:
+        x0 = np.asarray([3, 4])
+        xp_assert_close(b(x0 - 1e-10, nu=1),
+                        b(x0 + 1e-10, nu=1))
+        x1 = np.asarray([1, 6])
+        assert not np.allclose(b(x1 - 1e-10, nu=1), b(x1 + 1e-10, nu=1))
+
+        # 2nd derivative is not guaranteed to be continuous either
+        assert not np.allclose(b(x - 1e-10, nu=2), b(x + 1e-10, nu=2))
+
+    def test_basis_element_quadratic(self):
+        xx = np.linspace(-1, 4, 20)
+        b = BSpline.basis_element(t=[0, 1, 2, 3])
+        xp_assert_close(b(xx),
+                        splev(xx, (b.t, b.c, b.k)), atol=1e-14)
+        xp_assert_close(b(xx),
+                        B_0123(xx), atol=1e-14)
+
+        b = BSpline.basis_element(t=[0, 1, 1, 2])
+        xx = np.linspace(0, 2, 10)
+        xp_assert_close(b(xx),
+                np.where(xx < 1, xx*xx, (2.-xx)**2), atol=1e-14)
+
+    def test_basis_element_rndm(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        xx = np.linspace(t[k], t[-k-1], 20)
+        xp_assert_close(b(xx), _sum_basis_elements(xx, t, c, k), atol=1e-14)
+
+    def test_cmplx(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        cc = c * (1. + 3.j)
+
+        b = BSpline(t, cc, k)
+        b_re = BSpline(t, b.c.real, k)
+        b_im = BSpline(t, b.c.imag, k)
+
+        xx = np.linspace(t[k], t[-k-1], 20)
+        xp_assert_close(b(xx).real, b_re(xx), atol=1e-14)
+        xp_assert_close(b(xx).imag, b_im(xx), atol=1e-14)
+
+    def test_nan(self):
+        # nan in, nan out.
+        b = BSpline.basis_element([0, 1, 1, 2])
+        assert np.isnan(b(np.nan))
+
+    def test_derivative_method(self):
+        b = _make_random_spline(k=5)
+        t, c, k = b.tck
+        b0 = BSpline(t, c, k)
+        xx = np.linspace(t[k], t[-k-1], 20)
+        for j in range(1, k):
+            b = b.derivative()
+            xp_assert_close(b0(xx, j), b(xx), atol=1e-12, rtol=1e-12)
+
+    def test_antiderivative_method(self):
+        b = _make_random_spline()
+        t, c, k = b.tck
+        xx = np.linspace(t[k], t[-k-1], 20)
+        xp_assert_close(b.antiderivative().derivative()(xx),
+                        b(xx), atol=1e-14, rtol=1e-14)
+
+        # repeat with N-D array for c
+        c = np.c_[c, c, c]
+        c = np.dstack((c, c))
+        b = BSpline(t, c, k)
+        xp_assert_close(b.antiderivative().derivative()(xx),
+                        b(xx), atol=1e-14, rtol=1e-14)
+
+    def test_integral(self):
+        b = BSpline.basis_element([0, 1, 2])  # x for x < 1 else 2 - x
+        xp_assert_close(b.integrate(0, 1), np.asarray(0.5))
+        xp_assert_close(b.integrate(1, 0), np.asarray(-1 * 0.5))
+        xp_assert_close(b.integrate(1, 0), np.asarray(-0.5))
+
+        # extrapolate or zeros outside of [0, 2]; default is yes
+        xp_assert_close(b.integrate(-1, 1), np.asarray(0.0))
+        xp_assert_close(b.integrate(-1, 1, extrapolate=True), np.asarray(0.0))
+        xp_assert_close(b.integrate(-1, 1, extrapolate=False), np.asarray(0.5))
+        xp_assert_close(b.integrate(1, -1, extrapolate=False), np.asarray(-1 * 0.5))
+
+        # Test ``_fitpack._splint()``
+        xp_assert_close(b.integrate(1, -1, extrapolate=False),
+                        np.asarray(_impl.splint(1, -1, b.tck)))
+
+        # Test ``extrapolate='periodic'``.
+        b.extrapolate = 'periodic'
+        i = b.antiderivative()
+        period_int = np.asarray(i(2) - i(0))
+
+        xp_assert_close(b.integrate(0, 2), period_int)
+        xp_assert_close(b.integrate(2, 0), np.asarray(-1 * period_int))
+        xp_assert_close(b.integrate(-9, -7), period_int)
+        xp_assert_close(b.integrate(-8, -4), np.asarray(2 * period_int))
+
+        xp_assert_close(b.integrate(0.5, 1.5),
+                        np.asarray(i(1.5) - i(0.5)))
+        xp_assert_close(b.integrate(1.5, 3),
+                        np.asarray(i(1) - i(0) + i(2) - i(1.5)))
+        xp_assert_close(b.integrate(1.5 + 12, 3 + 12),
+                        np.asarray(i(1) - i(0) + i(2) - i(1.5)))
+        xp_assert_close(b.integrate(1.5, 3 + 12),
+                        np.asarray(i(1) - i(0) + i(2) - i(1.5) + 6 * period_int))
+
+        xp_assert_close(b.integrate(0, -1), np.asarray(i(0) - i(1)))
+        xp_assert_close(b.integrate(-9, -10), np.asarray(i(0) - i(1)))
+        xp_assert_close(b.integrate(0, -9),
+                        np.asarray(i(1) - i(2) - 4 * period_int))
+
+    def test_integrate_ppoly(self):
+        # test .integrate method to be consistent with PPoly.integrate
+        x = [0, 1, 2, 3, 4]
+        b = make_interp_spline(x, x)
+        b.extrapolate = 'periodic'
+        p = PPoly.from_spline(b)
+
+        for x0, x1 in [(-5, 0.5), (0.5, 5), (-4, 13)]:
+            xp_assert_close(b.integrate(x0, x1),
+                            p.integrate(x0, x1))
+
+    def test_integrate_0D_always(self):
+        # make sure the result is always a 0D array (not a python scalar)
+        b = BSpline.basis_element([0, 1, 2])
+        for extrapolate in (True, False):
+            res = b.integrate(0, 1, extrapolate=extrapolate)
+            assert isinstance(res, np.ndarray)
+            assert res.ndim == 0
+
+    def test_subclassing(self):
+        # classmethods should not decay to the base class
+        class B(BSpline):
+            pass
+
+        b = B.basis_element([0, 1, 2, 2])
+        assert b.__class__ == B
+        assert b.derivative().__class__ == B
+        assert b.antiderivative().__class__ == B
+
+    @pytest.mark.parametrize('axis', range(-4, 4))
+    def test_axis(self, axis):
+        n, k = 22, 3
+        t = np.linspace(0, 1, n + k + 1)
+        sh = [6, 7, 8]
+        # We need the positive axis for some of the indexing and slices used
+        # in this test.
+        pos_axis = axis % 4
+        sh.insert(pos_axis, n)   # [22, 6, 7, 8] etc
+        sh = tuple(sh)
+        rng = np.random.RandomState(1234)
+        c = rng.random(size=sh)
+        b = BSpline(t, c, k, axis=axis)
+        assert b.c.shape == (sh[pos_axis],) + sh[:pos_axis] + sh[pos_axis+1:]
+
+        xp = rng.random((3, 4, 5))
+        assert b(xp).shape == sh[:pos_axis] + xp.shape + sh[pos_axis+1:]
+
+        # -c.ndim <= axis < c.ndim
+        for ax in [-c.ndim - 1, c.ndim]:
+            assert_raises(AxisError, BSpline,
+                          **dict(t=t, c=c, k=k, axis=ax))
+
+        # derivative, antiderivative keeps the axis
+        for b1 in [BSpline(t, c, k, axis=axis).derivative(),
+                   BSpline(t, c, k, axis=axis).derivative(2),
+                   BSpline(t, c, k, axis=axis).antiderivative(),
+                   BSpline(t, c, k, axis=axis).antiderivative(2)]:
+            assert b1.axis == b.axis
+
+    def test_neg_axis(self):
+        k = 2
+        t = [0, 1, 2, 3, 4, 5, 6]
+        c = np.array([[-1, 2, 0, -1], [2, 0, -3, 1]])
+
+        spl = BSpline(t, c, k, axis=-1)
+        spl0 = BSpline(t, c[0], k)
+        spl1 = BSpline(t, c[1], k)
+        xp_assert_equal(spl(2.5), [spl0(2.5), spl1(2.5)])
+
+    @pytest.mark.thread_unsafe
+    def test_design_matrix_bc_types(self):
+        '''
+        Splines with different boundary conditions are built on different
+        types of vectors of knots. As far as design matrix depends only on
+        vector of knots, `k` and `x` it is useful to make tests for different
+        boundary conditions (and as following different vectors of knots).
+        '''
+        def run_design_matrix_tests(n, k, bc_type):
+            '''
+            To avoid repetition of code the following function is provided.
+            '''
+            rng = np.random.RandomState(1234)
+            x = np.sort(rng.random_sample(n) * 40 - 20)
+            y = rng.random_sample(n) * 40 - 20
+            if bc_type == "periodic":
+                y[0] = y[-1]
+
+            bspl = make_interp_spline(x, y, k=k, bc_type=bc_type)
+
+            c = np.eye(len(bspl.t) - k - 1)
+            des_matr_def = BSpline(bspl.t, c, k)(x)
+            des_matr_csr = BSpline.design_matrix(x,
+                                                 bspl.t,
+                                                 k).toarray()
+            xp_assert_close(des_matr_csr @ bspl.c, y, atol=1e-14)
+            xp_assert_close(des_matr_def, des_matr_csr, atol=1e-14)
+
+        # "clamped" and "natural" work only with `k = 3`
+        n = 11
+        k = 3
+        for bc in ["clamped", "natural"]:
+            run_design_matrix_tests(n, k, bc)
+
+        # "not-a-knot" works with odd `k`
+        for k in range(3, 8, 2):
+            run_design_matrix_tests(n, k, "not-a-knot")
+
+        # "periodic" works with any `k` (even more than `n`)
+        n = 5  # smaller `n` to test `k > n` case
+        for k in range(2, 7):
+            run_design_matrix_tests(n, k, "periodic")
+
+    @pytest.mark.parametrize('extrapolate', [False, True, 'periodic'])
+    @pytest.mark.parametrize('degree', range(5))
+    def test_design_matrix_same_as_BSpline_call(self, extrapolate, degree):
+        """Test that design_matrix(x) is equivalent to BSpline(..)(x)."""
+        rng = np.random.RandomState(1234)
+        x = rng.random_sample(10 * (degree + 1))
+        xmin, xmax = np.amin(x), np.amax(x)
+        k = degree
+        t = np.r_[np.linspace(xmin - 2, xmin - 1, degree),
+                  np.linspace(xmin, xmax, 2 * (degree + 1)),
+                  np.linspace(xmax + 1, xmax + 2, degree)]
+        c = np.eye(len(t) - k - 1)
+        bspline = BSpline(t, c, k, extrapolate)
+        xp_assert_close(
+            bspline(x), BSpline.design_matrix(x, t, k, extrapolate).toarray()
+        )
+
+        # extrapolation regime
+        x = np.array([xmin - 10, xmin - 1, xmax + 1.5, xmax + 10])
+        if not extrapolate:
+            with pytest.raises(ValueError):
+                BSpline.design_matrix(x, t, k, extrapolate)
+        else:
+            xp_assert_close(
+                bspline(x),
+                BSpline.design_matrix(x, t, k, extrapolate).toarray()
+            )
+
+    def test_design_matrix_x_shapes(self):
+        # test for different `x` shapes
+        rng = np.random.RandomState(1234)
+        n = 10
+        k = 3
+        x = np.sort(rng.random_sample(n) * 40 - 20)
+        y = rng.random_sample(n) * 40 - 20
+
+        bspl = make_interp_spline(x, y, k=k)
+        for i in range(1, 4):
+            xc = x[:i]
+            yc = y[:i]
+            des_matr_csr = BSpline.design_matrix(xc,
+                                                 bspl.t,
+                                                 k).toarray()
+            xp_assert_close(des_matr_csr @ bspl.c, yc, atol=1e-14)
+
+    def test_design_matrix_t_shapes(self):
+        # test for minimal possible `t` shape
+        t = [1., 1., 1., 2., 3., 4., 4., 4.]
+        des_matr = BSpline.design_matrix(2., t, 3).toarray()
+        xp_assert_close(des_matr,
+                        [[0.25, 0.58333333, 0.16666667, 0.]],
+                        atol=1e-14)
+
+    def test_design_matrix_asserts(self):
+        rng = np.random.RandomState(1234)
+        n = 10
+        k = 3
+        x = np.sort(rng.random_sample(n) * 40 - 20)
+        y = rng.random_sample(n) * 40 - 20
+        bspl = make_interp_spline(x, y, k=k)
+        # invalid vector of knots (should be a 1D non-descending array)
+        # here the actual vector of knots is reversed, so it is invalid
+        with assert_raises(ValueError):
+            BSpline.design_matrix(x, bspl.t[::-1], k)
+        k = 2
+        t = [0., 1., 2., 3., 4., 5.]
+        x = [1., 2., 3., 4.]
+        # out of bounds
+        with assert_raises(ValueError):
+            BSpline.design_matrix(x, t, k)
+
+    @pytest.mark.parametrize('bc_type', ['natural', 'clamped',
+                                         'periodic', 'not-a-knot'])
+    def test_from_power_basis(self, bc_type):
+        rng = np.random.RandomState(1234)
+        x = np.sort(rng.random(20))
+        y = rng.random(20)
+        if bc_type == 'periodic':
+            y[-1] = y[0]
+        cb = CubicSpline(x, y, bc_type=bc_type)
+        bspl = BSpline.from_power_basis(cb, bc_type=bc_type)
+        xx = np.linspace(0, 1, 20)
+        xp_assert_close(cb(xx), bspl(xx), atol=1e-15)
+        bspl_new = make_interp_spline(x, y, bc_type=bc_type)
+        xp_assert_close(bspl.c, bspl_new.c, atol=1e-15)
+
+    @pytest.mark.parametrize('bc_type', ['natural', 'clamped',
+                                         'periodic', 'not-a-knot'])
+    def test_from_power_basis_complex(self, bc_type):
+        rng = np.random.RandomState(1234)
+        x = np.sort(rng.random(20))
+        y = rng.random(20) + rng.random(20) * 1j
+        if bc_type == 'periodic':
+            y[-1] = y[0]
+        cb = CubicSpline(x, y, bc_type=bc_type)
+        bspl = BSpline.from_power_basis(cb, bc_type=bc_type)
+        bspl_new_real = make_interp_spline(x, y.real, bc_type=bc_type)
+        bspl_new_imag = make_interp_spline(x, y.imag, bc_type=bc_type)
+        xp_assert_close(bspl.c, bspl_new_real.c + 1j * bspl_new_imag.c, atol=1e-15)
+
+    def test_from_power_basis_exmp(self):
+        '''
+        For x = [0, 1, 2, 3, 4] and y = [1, 1, 1, 1, 1]
+        the coefficients of Cubic Spline in the power basis:
+
+        $[[0, 0, 0, 0, 0],\\$
+        $[0, 0, 0, 0, 0],\\$
+        $[0, 0, 0, 0, 0],\\$
+        $[1, 1, 1, 1, 1]]$
+
+        It could be shown explicitly that coefficients of the interpolating
+        function in B-spline basis are c = [1, 1, 1, 1, 1, 1, 1]
+        '''
+        x = np.array([0, 1, 2, 3, 4])
+        y = np.array([1, 1, 1, 1, 1])
+        bspl = BSpline.from_power_basis(CubicSpline(x, y, bc_type='natural'),
+                                        bc_type='natural')
+        xp_assert_close(bspl.c, [1.0, 1, 1, 1, 1, 1, 1], atol=1e-15)
+
+    def test_read_only(self):
+        # BSpline must work on read-only knots and coefficients.
+        t = np.array([0, 1])
+        c = np.array([3.0])
+        t.setflags(write=False)
+        c.setflags(write=False)
+
+        xx = np.linspace(0, 1, 10)
+        xx.setflags(write=False)
+
+        b = BSpline(t=t, c=c, k=0)
+        xp_assert_close(b(xx), np.ones_like(xx) * 3.0)
+
+    @pytest.mark.thread_unsafe
+    def test_concurrency(self):
+        # Check that no segfaults appear with concurrent access to BSpline
+        b = _make_random_spline()
+
+        def worker_fn(_, b):
+            t, _, k = b.tck
+            xx = np.linspace(t[k], t[-k-1], 10000)
+            b(xx)
+
+        _run_concurrent_barrier(10, worker_fn, b)
+
+
+    def test_memmap(self, tmpdir):
+        # Make sure that memmaps can be used as t and c atrributes after the
+        # spline has been constructed. This is similar to what happens in a
+        # scikit-learn context, where joblib can create read-only memmap to
+        # share objects between workers. For more details, see
+        # https://github.com/scipy/scipy/issues/22143
+        b = _make_random_spline()
+        xx = np.linspace(0, 1, 10)
+
+        expected = b(xx)
+
+        tid = threading.get_native_id()
+        t_mm = np.memmap(str(tmpdir.join(f't{tid}.dat')), mode='w+',
+                         dtype=b.t.dtype, shape=b.t.shape)
+        t_mm[:] = b.t
+        c_mm = np.memmap(str(tmpdir.join(f'c{tid}.dat')), mode='w+',
+                         dtype=b.c.dtype, shape=b.c.shape)
+        c_mm[:] = b.c
+        b.t = t_mm
+        b.c = c_mm
+
+        xp_assert_close(b(xx), expected)
+
+
+class TestInsert:
+
+    @pytest.mark.parametrize('xval', [0.0, 1.0, 2.5, 4, 6.5, 7.0])
+    def test_insert(self, xval):
+        # insert a knot, incl edges (0.0, 7.0) and exactly at an existing knot (4.0)
+        x = np.arange(8)
+        y = np.sin(x)**3
+        spl = make_interp_spline(x, y, k=3)
+
+        spl_1f = insert(xval, spl)     # FITPACK
+        spl_1 = spl.insert_knot(xval)
+
+        xp_assert_close(spl_1.t, spl_1f.t, atol=1e-15)
+        xp_assert_close(spl_1.c, spl_1f.c[:-spl.k-1], atol=1e-15)
+
+        # knot insertion preserves values, unless multiplicity >= k+1
+        xx = x if xval != x[-1] else x[:-1]
+        xx = np.r_[xx, 0.5*(x[1:] + x[:-1])]
+        xp_assert_close(spl(xx), spl_1(xx), atol=1e-15)
+
+        # ... repeat with ndim > 1
+        y1 = np.cos(x)**3
+        spl_y1 = make_interp_spline(x, y1, k=3)
+        spl_yy = make_interp_spline(x, np.c_[y, y1], k=3)
+        spl_yy1 = spl_yy.insert_knot(xval)
+
+        xp_assert_close(spl_yy1.t, spl_1.t, atol=1e-15)
+        xp_assert_close(spl_yy1.c, np.c_[spl.insert_knot(xval).c,
+                                         spl_y1.insert_knot(xval).c], atol=1e-15)
+
+        xx = x if xval != x[-1] else x[:-1]
+        xx = np.r_[xx, 0.5*(x[1:] + x[:-1])]
+        xp_assert_close(spl_yy(xx), spl_yy1(xx), atol=1e-15)
+
+
+    @pytest.mark.parametrize(
+        'xval, m', [(0.0, 2), (1.0, 3), (1.5, 5), (4, 2), (7.0, 2)]
+    )
+    def test_insert_multi(self, xval, m):
+        x = np.arange(8)
+        y = np.sin(x)**3
+        spl = make_interp_spline(x, y, k=3)
+
+        spl_1f = insert(xval, spl, m=m)
+        spl_1 = spl.insert_knot(xval, m)
+
+        xp_assert_close(spl_1.t, spl_1f.t, atol=1e-15)
+        xp_assert_close(spl_1.c, spl_1f.c[:-spl.k-1], atol=1e-15)
+
+        xx = x if xval != x[-1] else x[:-1]
+        xx = np.r_[xx, 0.5*(x[1:] + x[:-1])]
+        xp_assert_close(spl(xx), spl_1(xx), atol=1e-15)
+
+    def test_insert_random(self):
+        rng = np.random.default_rng(12345)
+        n, k = 11, 3
+
+        t = np.sort(rng.uniform(size=n+k+1))
+        c = rng.uniform(size=(n, 3, 2))
+        spl = BSpline(t, c, k)
+
+        xv = rng.uniform(low=t[k+1], high=t[-k-1])
+        spl_1 = spl.insert_knot(xv)
+
+        xx = rng.uniform(low=t[k+1], high=t[-k-1], size=33)
+        xp_assert_close(spl(xx), spl_1(xx), atol=1e-15)
+
+    @pytest.mark.parametrize('xv', [0, 0.1, 2.0, 4.0, 4.5,      # l.h. edge
+                                    5.5, 6.0, 6.1, 7.0]         # r.h. edge
+    )
+    def test_insert_periodic(self, xv):
+        x = np.arange(8)
+        y = np.sin(x)**3
+        tck = splrep(x, y, k=3)
+        spl = BSpline(*tck, extrapolate="periodic")
+
+        spl_1 = spl.insert_knot(xv)
+        tf, cf, k = insert(xv, spl.tck, per=True)
+
+        xp_assert_close(spl_1.t, tf, atol=1e-15)
+        xp_assert_close(spl_1.c[:-k-1], cf[:-k-1], atol=1e-15)
+
+        xx = np.random.default_rng(1234).uniform(low=0, high=7, size=41)
+        xp_assert_close(spl_1(xx), splev(xx, (tf, cf, k)), atol=1e-15)
+
+    @pytest.mark.parametrize('extrapolate', [None, 'periodic'])
+    def test_complex(self, extrapolate):
+        x = np.arange(8)*2*np.pi
+        y_re, y_im = np.sin(x), np.cos(x)
+
+        spl = make_interp_spline(x, y_re + 1j*y_im, k=3)
+        spl.extrapolate = extrapolate
+
+        spl_re = make_interp_spline(x, y_re, k=3)
+        spl_re.extrapolate = extrapolate
+
+        spl_im = make_interp_spline(x, y_im, k=3)
+        spl_im.extrapolate = extrapolate
+
+        xv = 3.5
+        spl_1 = spl.insert_knot(xv)
+        spl_1re = spl_re.insert_knot(xv)
+        spl_1im = spl_im.insert_knot(xv)
+
+        xp_assert_close(spl_1.t, spl_1re.t, atol=1e-15)
+        xp_assert_close(spl_1.t, spl_1im.t, atol=1e-15)
+        xp_assert_close(spl_1.c, spl_1re.c + 1j*spl_1im.c, atol=1e-15)
+
+    def test_insert_periodic_too_few_internal_knots(self):
+        # both FITPACK and spl.insert_knot raise when there's not enough
+        # internal knots to make a periodic extension.
+        # Below the internal knots are 2, 3,    , 4, 5
+        #                                     ^
+        #                              2, 3, 3.5, 4, 5
+        #   so two knots from each side from the new one, while need at least
+        #   from either left or right.
+        xv = 3.5
+        k = 3
+        t = np.array([0]*(k+1) + [2, 3, 4, 5] + [7]*(k+1))
+        c = np.ones(len(t) - k - 1)
+        spl = BSpline(t, c, k, extrapolate="periodic")
+
+        with assert_raises(ValueError):
+            insert(xv, (t, c, k), per=True)
+
+        with assert_raises(ValueError):
+            spl.insert_knot(xv)
+
+    def test_insert_no_extrap(self):
+        k = 3
+        t = np.array([0]*(k+1) + [2, 3, 4, 5] + [7]*(k+1))
+        c = np.ones(len(t) - k - 1)
+        spl = BSpline(t, c, k)
+
+        with assert_raises(ValueError):
+            spl.insert_knot(-1)
+
+        with assert_raises(ValueError):
+            spl.insert_knot(8)
+
+        with assert_raises(ValueError):
+            spl.insert_knot(3, m=0)
+
+
+def test_knots_multiplicity():
+    # Take a spline w/ random coefficients, throw in knots of varying
+    # multiplicity.
+
+    def check_splev(b, j, der=0, atol=1e-14, rtol=1e-14):
+        # check evaluations against FITPACK, incl extrapolations
+        t, c, k = b.tck
+        x = np.unique(t)
+        x = np.r_[t[0]-0.1, 0.5*(x[1:] + x[:1]), t[-1]+0.1]
+        xp_assert_close(splev(x, (t, c, k), der), b(x, der),
+                atol=atol, rtol=rtol, err_msg=f'der = {der}  k = {b.k}')
+
+    # test loop itself
+    # [the index `j` is for interpreting the traceback in case of a failure]
+    for k in [1, 2, 3, 4, 5]:
+        b = _make_random_spline(k=k)
+        for j, b1 in enumerate(_make_multiples(b)):
+            check_splev(b1, j)
+            for der in range(1, k+1):
+                check_splev(b1, j, der, 1e-12, 1e-12)
+
+
+### stolen from @pv, verbatim
+def _naive_B(x, k, i, t):
+    """
+    Naive way to compute B-spline basis functions. Useful only for testing!
+    computes B(x; t[i],..., t[i+k+1])
+    """
+    if k == 0:
+        return 1.0 if t[i] <= x < t[i+1] else 0.0
+    if t[i+k] == t[i]:
+        c1 = 0.0
+    else:
+        c1 = (x - t[i])/(t[i+k] - t[i]) * _naive_B(x, k-1, i, t)
+    if t[i+k+1] == t[i+1]:
+        c2 = 0.0
+    else:
+        c2 = (t[i+k+1] - x)/(t[i+k+1] - t[i+1]) * _naive_B(x, k-1, i+1, t)
+    return (c1 + c2)
+
+
+### stolen from @pv, verbatim
+def _naive_eval(x, t, c, k):
+    """
+    Naive B-spline evaluation. Useful only for testing!
+    """
+    if x == t[k]:
+        i = k
+    else:
+        i = np.searchsorted(t, x) - 1
+    assert t[i] <= x <= t[i+1]
+    assert i >= k and i < len(t) - k
+    return sum(c[i-j] * _naive_B(x, k, i-j, t) for j in range(0, k+1))
+
+
+def _naive_eval_2(x, t, c, k):
+    """Naive B-spline evaluation, another way."""
+    n = len(t) - (k+1)
+    assert n >= k+1
+    assert len(c) >= n
+    assert t[k] <= x <= t[n]
+    return sum(c[i] * _naive_B(x, k, i, t) for i in range(n))
+
+
+def _sum_basis_elements(x, t, c, k):
+    n = len(t) - (k+1)
+    assert n >= k+1
+    assert len(c) >= n
+    s = 0.
+    for i in range(n):
+        b = BSpline.basis_element(t[i:i+k+2], extrapolate=False)(x)
+        s += c[i] * np.nan_to_num(b)   # zero out out-of-bounds elements
+    return s
+
+
+def B_012(x):
+    """ A linear B-spline function B(x | 0, 1, 2)."""
+    x = np.atleast_1d(x)
+    return np.piecewise(x, [(x < 0) | (x > 2),
+                            (x >= 0) & (x < 1),
+                            (x >= 1) & (x <= 2)],
+                           [lambda x: 0., lambda x: x, lambda x: 2.-x])
+
+
+def B_0123(x, der=0):
+    """A quadratic B-spline function B(x | 0, 1, 2, 3)."""
+    x = np.atleast_1d(x)
+    conds = [x < 1, (x > 1) & (x < 2), x > 2]
+    if der == 0:
+        funcs = [lambda x: x*x/2.,
+                 lambda x: 3./4 - (x-3./2)**2,
+                 lambda x: (3.-x)**2 / 2]
+    elif der == 2:
+        funcs = [lambda x: 1.,
+                 lambda x: -2.,
+                 lambda x: 1.]
+    else:
+        raise ValueError(f'never be here: der={der}')
+    pieces = np.piecewise(x, conds, funcs)
+    return pieces
+
+
+def _make_random_spline(n=35, k=3):
+    rng = np.random.RandomState(123)
+    t = np.sort(rng.random(n+k+1))
+    c = rng.random(n)
+    return BSpline.construct_fast(t, c, k)
+
+
+def _make_multiples(b):
+    """Increase knot multiplicity."""
+    c, k = b.c, b.k
+
+    t1 = b.t.copy()
+    t1[17:19] = t1[17]
+    t1[22] = t1[21]
+    yield BSpline(t1, c, k)
+
+    t1 = b.t.copy()
+    t1[:k+1] = t1[0]
+    yield BSpline(t1, c, k)
+
+    t1 = b.t.copy()
+    t1[-k-1:] = t1[-1]
+    yield BSpline(t1, c, k)
+
+
+class TestInterop:
+    #
+    # Test that FITPACK-based spl* functions can deal with BSpline objects
+    #
+    def setup_method(self):
+        xx = np.linspace(0, 4.*np.pi, 41)
+        yy = np.cos(xx)
+        b = make_interp_spline(xx, yy)
+        self.tck = (b.t, b.c, b.k)
+        self.xx, self.yy, self.b = xx, yy, b
+
+        self.xnew = np.linspace(0, 4.*np.pi, 21)
+
+        c2 = np.c_[b.c, b.c, b.c]
+        self.c2 = np.dstack((c2, c2))
+        self.b2 = BSpline(b.t, self.c2, b.k)
+
+    def test_splev(self):
+        xnew, b, b2 = self.xnew, self.b, self.b2
+
+        # check that splev works with 1-D array of coefficients
+        # for array and scalar `x`
+        xp_assert_close(splev(xnew, b),
+                        b(xnew), atol=1e-15, rtol=1e-15)
+        xp_assert_close(splev(xnew, b.tck),
+                        b(xnew), atol=1e-15, rtol=1e-15)
+        xp_assert_close(np.asarray([splev(x, b) for x in xnew]),
+                        b(xnew), atol=1e-15, rtol=1e-15)
+
+        # With N-D coefficients, there's a quirck:
+        # splev(x, BSpline) is equivalent to BSpline(x)
+        with assert_raises(ValueError, match="Calling splev.. with BSpline"):
+            splev(xnew, b2)
+
+        # However, splev(x, BSpline.tck) needs some transposes. This is because
+        # BSpline interpolates along the first axis, while the legacy FITPACK
+        # wrapper does list(map(...)) which effectively interpolates along the
+        # last axis. Like so:
+        sh = tuple(range(1, b2.c.ndim)) + (0,)   # sh = (1, 2, 0)
+        cc = b2.c.transpose(sh)
+        tck = (b2.t, cc, b2.k)
+        xp_assert_close(np.asarray(splev(xnew, tck)),
+                        b2(xnew).transpose(sh), atol=1e-15, rtol=1e-15)
+
+    def test_splrep(self):
+        x, y = self.xx, self.yy
+        # test that "new" splrep is equivalent to _impl.splrep
+        tck = splrep(x, y)
+        t, c, k = _impl.splrep(x, y)
+        xp_assert_close(tck[0], t, atol=1e-15)
+        xp_assert_close(tck[1], c, atol=1e-15)
+        assert tck[2] == k
+
+        # also cover the `full_output=True` branch
+        tck_f, _, _, _ = splrep(x, y, full_output=True)
+        xp_assert_close(tck_f[0], t, atol=1e-15)
+        xp_assert_close(tck_f[1], c, atol=1e-15)
+        assert tck_f[2] == k
+
+        # test that the result of splrep roundtrips with splev:
+        # evaluate the spline on the original `x` points
+        yy = splev(x, tck)
+        xp_assert_close(y, yy, atol=1e-15)
+
+        # ... and also it roundtrips if wrapped in a BSpline
+        b = BSpline(*tck)
+        xp_assert_close(y, b(x), atol=1e-15)
+
+    def test_splrep_errors(self):
+        # test that both "old" and "new" splrep raise for an N-D ``y`` array
+        # with n > 1
+        x, y = self.xx, self.yy
+        y2 = np.c_[y, y]
+        with assert_raises(ValueError):
+            splrep(x, y2)
+        with assert_raises(ValueError):
+            _impl.splrep(x, y2)
+
+        # input below minimum size
+        with assert_raises(TypeError, match="m > k must hold"):
+            splrep(x[:3], y[:3])
+        with assert_raises(TypeError, match="m > k must hold"):
+            _impl.splrep(x[:3], y[:3])
+
+    def test_splprep(self):
+        x = np.arange(15, dtype=np.float64).reshape((3, 5))
+        b, u = splprep(x)
+        tck, u1 = _impl.splprep(x)
+
+        # test the roundtrip with splev for both "old" and "new" output
+        xp_assert_close(u, u1, atol=1e-15)
+        xp_assert_close(np.asarray(splev(u, b)), x, atol=1e-15)
+        xp_assert_close(np.asarray(splev(u, tck)), x, atol=1e-15)
+
+        # cover the ``full_output=True`` branch
+        (b_f, u_f), _, _, _ = splprep(x, s=0, full_output=True)
+        xp_assert_close(u, u_f, atol=1e-15)
+        xp_assert_close(np.asarray(splev(u_f, b_f)), x, atol=1e-15)
+
+    def test_splprep_errors(self):
+        # test that both "old" and "new" code paths raise for x.ndim > 2
+        x = np.arange(3*4*5).reshape((3, 4, 5))
+        with assert_raises(ValueError, match="too many values to unpack"):
+            splprep(x)
+        with assert_raises(ValueError, match="too many values to unpack"):
+            _impl.splprep(x)
+
+        # input below minimum size
+        x = np.linspace(0, 40, num=3)
+        with assert_raises(TypeError, match="m > k must hold"):
+            splprep([x])
+        with assert_raises(TypeError, match="m > k must hold"):
+            _impl.splprep([x])
+
+        # automatically calculated parameters are non-increasing
+        # see gh-7589
+        x = [-50.49072266, -50.49072266, -54.49072266, -54.49072266]
+        with assert_raises(ValueError, match="Invalid inputs"):
+            splprep([x])
+        with assert_raises(ValueError, match="Invalid inputs"):
+            _impl.splprep([x])
+
+        # given non-increasing parameter values u
+        x = [1, 3, 2, 4]
+        u = [0, 0.3, 0.2, 1]
+        with assert_raises(ValueError, match="Invalid inputs"):
+            splprep(*[[x], None, u])
+
+    def test_sproot(self):
+        b, b2 = self.b, self.b2
+        roots = np.array([0.5, 1.5, 2.5, 3.5])*np.pi
+        # sproot accepts a BSpline obj w/ 1-D coef array
+        xp_assert_close(sproot(b), roots, atol=1e-7, rtol=1e-7)
+        xp_assert_close(sproot((b.t, b.c, b.k)), roots, atol=1e-7, rtol=1e-7)
+
+        # ... and deals with trailing dimensions if coef array is N-D
+        with assert_raises(ValueError, match="Calling sproot.. with BSpline"):
+            sproot(b2, mest=50)
+
+        # and legacy behavior is preserved for a tck tuple w/ N-D coef
+        c2r = b2.c.transpose(1, 2, 0)
+        rr = np.asarray(sproot((b2.t, c2r, b2.k), mest=50))
+        assert rr.shape == (3, 2, 4)
+        xp_assert_close(rr - roots, np.zeros_like(rr), atol=1e-12)
+
+    def test_splint(self):
+        # test that splint accepts BSpline objects
+        b, b2 = self.b, self.b2
+
+        xp_assert_close(splint(0, 1, b),
+                        splint(0, 1, b.tck), atol=1e-14, check_0d=False)
+        xp_assert_close(splint(0, 1, b),
+                        b.integrate(0, 1), atol=1e-14, check_0d=False)
+
+        # ... and deals with N-D arrays of coefficients
+        with assert_raises(ValueError, match="Calling splint.. with BSpline"):
+            splint(0, 1, b2)
+
+        # and the legacy behavior is preserved for a tck tuple w/ N-D coef
+        c2r = b2.c.transpose(1, 2, 0)
+        integr = np.asarray(splint(0, 1, (b2.t, c2r, b2.k)))
+        assert integr.shape == (3, 2)
+        xp_assert_close(integr,
+                        splint(0, 1, b), atol=1e-14, check_shape=False)
+
+    def test_splder(self):
+        for b in [self.b, self.b2]:
+            # pad the c array (FITPACK convention)
+            ct = len(b.t) - len(b.c)
+            b_c = b.c.copy()
+            if ct > 0:
+                b_c = np.r_[b_c, np.zeros((ct,) + b_c.shape[1:])]
+
+            for n in [1, 2, 3]:
+                bd = splder(b)
+                tck_d = _impl.splder((b.t.copy(), b_c, b.k))
+                xp_assert_close(bd.t, tck_d[0], atol=1e-15)
+                xp_assert_close(bd.c, tck_d[1], atol=1e-15)
+                assert bd.k == tck_d[2]
+                assert isinstance(bd, BSpline)
+                assert isinstance(tck_d, tuple)  # back-compat: tck in and out
+
+    def test_splantider(self):
+        for b in [self.b, self.b2]:
+            # pad the c array (FITPACK convention)
+            ct = len(b.t) - len(b.c)
+            b_c = b.c.copy()
+            if ct > 0:
+                b_c = np.r_[b_c, np.zeros((ct,) + b_c.shape[1:])]
+
+            for n in [1, 2, 3]:
+                bd = splantider(b)
+                tck_d = _impl.splantider((b.t.copy(), b_c, b.k))
+                xp_assert_close(bd.t, tck_d[0], atol=1e-15)
+                xp_assert_close(bd.c, tck_d[1], atol=1e-15)
+                assert bd.k == tck_d[2]
+                assert isinstance(bd, BSpline)
+                assert isinstance(tck_d, tuple)  # back-compat: tck in and out
+
+    def test_insert(self):
+        b, b2, xx = self.b, self.b2, self.xx
+
+        j = b.t.size // 2
+        tn = 0.5*(b.t[j] + b.t[j+1])
+
+        bn, tck_n = insert(tn, b), insert(tn, (b.t, b.c, b.k))
+        xp_assert_close(splev(xx, bn),
+                        splev(xx, tck_n), atol=1e-15)
+        assert isinstance(bn, BSpline)
+        assert isinstance(tck_n, tuple)   # back-compat: tck in, tck out
+
+        # for N-D array of coefficients, BSpline.c needs to be transposed
+        # after that, the results are equivalent.
+        sh = tuple(range(b2.c.ndim))
+        c_ = b2.c.transpose(sh[1:] + (0,))
+        tck_n2 = insert(tn, (b2.t, c_, b2.k))
+
+        bn2 = insert(tn, b2)
+
+        # need a transpose for comparing the results, cf test_splev
+        xp_assert_close(np.asarray(splev(xx, tck_n2)).transpose(2, 0, 1),
+                        bn2(xx), atol=1e-15)
+        assert isinstance(bn2, BSpline)
+        assert isinstance(tck_n2, tuple)   # back-compat: tck in, tck out
+
+
+class TestInterp:
+    #
+    # Test basic ways of constructing interpolating splines.
+    #
+    xx = np.linspace(0., 2.*np.pi)
+    yy = np.sin(xx)
+
+    def test_non_int_order(self):
+        with assert_raises(TypeError):
+            make_interp_spline(self.xx, self.yy, k=2.5)
+
+    def test_order_0(self):
+        b = make_interp_spline(self.xx, self.yy, k=0)
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        b = make_interp_spline(self.xx, self.yy, k=0, axis=-1)
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+
+    def test_linear(self):
+        b = make_interp_spline(self.xx, self.yy, k=1)
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        b = make_interp_spline(self.xx, self.yy, k=1, axis=-1)
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+
+    @pytest.mark.parametrize('k', [0, 1, 2, 3])
+    def test_incompatible_x_y(self, k):
+        x = [0, 1, 2, 3, 4, 5]
+        y = [0, 1, 2, 3, 4, 5, 6, 7]
+        with assert_raises(ValueError, match="Shapes of x"):
+            make_interp_spline(x, y, k=k)
+
+    @pytest.mark.parametrize('k', [0, 1, 2, 3])
+    def test_broken_x(self, k):
+        x = [0, 1, 1, 2, 3, 4]      # duplicates
+        y = [0, 1, 2, 3, 4, 5]
+        with assert_raises(ValueError, match="x to not have duplicates"):
+            make_interp_spline(x, y, k=k)
+
+        x = [0, 2, 1, 3, 4, 5]      # unsorted
+        with assert_raises(ValueError, match="Expect x to be a 1D strictly"):
+            make_interp_spline(x, y, k=k)
+
+        x = [0, 1, 2, 3, 4, 5]
+        x = np.asarray(x).reshape((1, -1))     # 1D
+        with assert_raises(ValueError, match="Expect x to be a 1D strictly"):
+            make_interp_spline(x, y, k=k)
+
+    def test_not_a_knot(self):
+        for k in [2, 3, 4, 5, 6, 7]:
+            b = make_interp_spline(self.xx, self.yy, k)
+            xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+
+    def test_periodic(self):
+        # k = 5 here for more derivatives
+        b = make_interp_spline(self.xx, self.yy, k=5, bc_type='periodic')
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        # in periodic case it is expected equality of k-1 first
+        # derivatives at the boundaries
+        for i in range(1, 5):
+            xp_assert_close(b(self.xx[0], nu=i), b(self.xx[-1], nu=i), atol=1e-11)
+        # tests for axis=-1
+        b = make_interp_spline(self.xx, self.yy, k=5, bc_type='periodic', axis=-1)
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        for i in range(1, 5):
+            xp_assert_close(b(self.xx[0], nu=i), b(self.xx[-1], nu=i), atol=1e-11)
+
+    @pytest.mark.parametrize('k', [2, 3, 4, 5, 6, 7])
+    def test_periodic_random(self, k):
+        # tests for both cases (k > n and k <= n)
+        n = 5
+        rng = np.random.RandomState(1234)
+        x = np.sort(rng.random_sample(n) * 10)
+        y = rng.random_sample(n) * 100
+        y[0] = y[-1]
+        b = make_interp_spline(x, y, k=k, bc_type='periodic')
+        xp_assert_close(b(x), y, atol=1e-14)
+
+    def test_periodic_axis(self):
+        n = self.xx.shape[0]
+        rng = np.random.RandomState(1234)
+        x = rng.random_sample(n) * 2 * np.pi
+        x = np.sort(x)
+        x[0] = 0.
+        x[-1] = 2 * np.pi
+        y = np.zeros((2, n))
+        y[0] = np.sin(x)
+        y[1] = np.cos(x)
+        b = make_interp_spline(x, y, k=5, bc_type='periodic', axis=1)
+        for i in range(n):
+            xp_assert_close(b(x[i]), y[:, i], atol=1e-14)
+        xp_assert_close(b(x[0]), b(x[-1]), atol=1e-14)
+
+    def test_periodic_points_exception(self):
+        # first and last points should match when periodic case expected
+        rng = np.random.RandomState(1234)
+        k = 5
+        n = 8
+        x = np.sort(rng.random_sample(n))
+        y = rng.random_sample(n)
+        y[0] = y[-1] - 1  # to be sure that they are not equal
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, k=k, bc_type='periodic')
+
+    def test_periodic_knots_exception(self):
+        # `periodic` case does not work with passed vector of knots
+        rng = np.random.RandomState(1234)
+        k = 3
+        n = 7
+        x = np.sort(rng.random_sample(n))
+        y = rng.random_sample(n)
+        t = np.zeros(n + 2 * k)
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, k, t, 'periodic')
+
+    @pytest.mark.parametrize('k', [2, 3, 4, 5])
+    def test_periodic_splev(self, k):
+        # comparison values of periodic b-spline with splev
+        b = make_interp_spline(self.xx, self.yy, k=k, bc_type='periodic')
+        tck = splrep(self.xx, self.yy, per=True, k=k)
+        spl = splev(self.xx, tck)
+        xp_assert_close(spl, b(self.xx), atol=1e-14)
+
+        # comparison derivatives of periodic b-spline with splev
+        for i in range(1, k):
+            spl = splev(self.xx, tck, der=i)
+            xp_assert_close(spl, b(self.xx, nu=i), atol=1e-10)
+
+    def test_periodic_cubic(self):
+        # comparison values of cubic periodic b-spline with CubicSpline
+        b = make_interp_spline(self.xx, self.yy, k=3, bc_type='periodic')
+        cub = CubicSpline(self.xx, self.yy, bc_type='periodic')
+        xp_assert_close(b(self.xx), cub(self.xx), atol=1e-14)
+
+        # edge case: Cubic interpolation on 3 points
+        rng = np.random.RandomState(1234)
+        n = 3
+        x = np.sort(rng.random_sample(n) * 10)
+        y = rng.random_sample(n) * 100
+        y[0] = y[-1]
+        b = make_interp_spline(x, y, k=3, bc_type='periodic')
+        cub = CubicSpline(x, y, bc_type='periodic')
+        xp_assert_close(b(x), cub(x), atol=1e-14)
+
+    def test_periodic_full_matrix(self):
+        # comparison values of cubic periodic b-spline with
+        # solution of the system with full matrix
+        k = 3
+        b = make_interp_spline(self.xx, self.yy, k=k, bc_type='periodic')
+        t = _periodic_knots(self.xx, k)
+        c = _make_interp_per_full_matr(self.xx, self.yy, t, k)
+        b1 = np.vectorize(lambda x: _naive_eval(x, t, c, k))
+        xp_assert_close(b(self.xx), b1(self.xx), atol=1e-14)
+
+    def test_quadratic_deriv(self):
+        der = [(1, 8.)]  # order, value: f'(x) = 8.
+
+        # derivative at right-hand edge
+        b = make_interp_spline(self.xx, self.yy, k=2, bc_type=(None, der))
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(
+            b(self.xx[-1], 1), der[0][1], atol=1e-14, rtol=1e-14, check_0d=False
+        )
+
+        # derivative at left-hand edge
+        b = make_interp_spline(self.xx, self.yy, k=2, bc_type=(der, None))
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(
+            b(self.xx[0], 1), der[0][1], atol=1e-14, rtol=1e-14, check_0d=False
+        )
+
+    def test_cubic_deriv(self):
+        k = 3
+
+        # first derivatives at left & right edges:
+        der_l, der_r = [(1, 3.)], [(1, 4.)]
+        b = make_interp_spline(self.xx, self.yy, k, bc_type=(der_l, der_r))
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(np.asarray([b(self.xx[0], 1), b(self.xx[-1], 1)]),
+                        np.asarray([der_l[0][1], der_r[0][1]]), atol=1e-14, rtol=1e-14)
+
+        # 'natural' cubic spline, zero out 2nd derivatives at the boundaries
+        der_l, der_r = [(2, 0)], [(2, 0)]
+        b = make_interp_spline(self.xx, self.yy, k, bc_type=(der_l, der_r))
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+
+    def test_quintic_derivs(self):
+        k, n = 5, 7
+        x = np.arange(n).astype(np.float64)
+        y = np.sin(x)
+        der_l = [(1, -12.), (2, 1)]
+        der_r = [(1, 8.), (2, 3.)]
+        b = make_interp_spline(x, y, k=k, bc_type=(der_l, der_r))
+        xp_assert_close(b(x), y, atol=1e-14, rtol=1e-14)
+        xp_assert_close(np.asarray([b(x[0], 1), b(x[0], 2)]),
+                        np.asarray([val for (nu, val) in der_l]))
+        xp_assert_close(np.asarray([b(x[-1], 1), b(x[-1], 2)]),
+                        np.asarray([val for (nu, val) in der_r]))
+
+    @pytest.mark.xfail(reason='unstable')
+    def test_cubic_deriv_unstable(self):
+        # 1st and 2nd derivative at x[0], no derivative information at x[-1]
+        # The problem is not that it fails [who would use this anyway],
+        # the problem is that it fails *silently*, and I've no idea
+        # how to detect this sort of instability.
+        # In this particular case: it's OK for len(t) < 20, goes haywire
+        # at larger `len(t)`.
+        k = 3
+        t = _augknt(self.xx, k)
+
+        der_l = [(1, 3.), (2, 4.)]
+        b = make_interp_spline(self.xx, self.yy, k, t, bc_type=(der_l, None))
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+
+    def test_knots_not_data_sites(self):
+        # Knots need not coincide with the data sites.
+        # use a quadratic spline, knots are at data averages,
+        # two additional constraints are zero 2nd derivatives at edges
+        k = 2
+        t = np.r_[(self.xx[0],)*(k+1),
+                  (self.xx[1:] + self.xx[:-1]) / 2.,
+                  (self.xx[-1],)*(k+1)]
+        b = make_interp_spline(self.xx, self.yy, k, t,
+                               bc_type=([(2, 0)], [(2, 0)]))
+
+        xp_assert_close(b(self.xx), self.yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(b(self.xx[0], 2), np.asarray(0.0), atol=1e-14)
+        xp_assert_close(b(self.xx[-1], 2), np.asarray(0.0), atol=1e-14)
+
+    def test_minimum_points_and_deriv(self):
+        # interpolation of f(x) = x**3 between 0 and 1. f'(x) = 3 * xx**2 and
+        # f'(0) = 0, f'(1) = 3.
+        k = 3
+        x = [0., 1.]
+        y = [0., 1.]
+        b = make_interp_spline(x, y, k, bc_type=([(1, 0.)], [(1, 3.)]))
+
+        xx = np.linspace(0., 1.)
+        yy = xx**3
+        xp_assert_close(b(xx), yy, atol=1e-14, rtol=1e-14)
+
+    def test_deriv_spec(self):
+        # If one of the derivatives is omitted, the spline definition is
+        # incomplete.
+        x = y = [1.0, 2, 3, 4, 5, 6]
+
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, bc_type=([(1, 0.)], None))
+
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, bc_type=(1, 0.))
+
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, bc_type=[(1, 0.)])
+
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, bc_type=42)
+
+        # CubicSpline expects`bc_type=(left_pair, right_pair)`, while
+        # here we expect `bc_type=(iterable, iterable)`.
+        l, r = (1, 0.0), (1, 0.0)
+        with assert_raises(ValueError):
+            make_interp_spline(x, y, bc_type=(l, r))
+
+    def test_deriv_order_too_large(self):
+        x = np.arange(7)
+        y = x**2
+        l, r = [(6, 0)], [(1, 0)]    # 6th derivative = 0 at x[0] for k=3
+        with assert_raises(ValueError, match="Bad boundary conditions at 0."):
+            # cannot fix 6th derivative at x[0]: does not segfault
+            make_interp_spline(x, y, bc_type=(l, r))
+
+        l, r = [(1, 0)], [(-6, 0)]    # derivative order < 0 at x[-1]
+        with assert_raises(ValueError, match="Bad boundary conditions at 6."):
+            # does not segfault
+            make_interp_spline(x, y, bc_type=(l, r))
+
+    def test_complex(self):
+        k = 3
+        xx = self.xx
+        yy = self.yy + 1.j*self.yy
+
+        # first derivatives at left & right edges:
+        der_l, der_r = [(1, 3.j)], [(1, 4.+2.j)]
+        b = make_interp_spline(xx, yy, k, bc_type=(der_l, der_r))
+        xp_assert_close(b(xx), yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(
+            b(xx[0], 1), der_l[0][1], atol=1e-14, rtol=1e-14, check_0d=False
+        )
+        xp_assert_close(
+            b(xx[-1], 1), der_r[0][1], atol=1e-14, rtol=1e-14, check_0d=False
+        )
+
+        # also test zero and first order
+        for k in (0, 1):
+            b = make_interp_spline(xx, yy, k=k)
+            xp_assert_close(b(xx), yy, atol=1e-14, rtol=1e-14)
+
+    def test_int_xy(self):
+        x = np.arange(10).astype(int)
+        y = np.arange(10).astype(int)
+
+        # Cython chokes on "buffer type mismatch" (construction) or
+        # "no matching signature found" (evaluation)
+        for k in (0, 1, 2, 3):
+            b = make_interp_spline(x, y, k=k)
+            b(x)
+
+    def test_sliced_input(self):
+        # Cython code chokes on non C contiguous arrays
+        xx = np.linspace(-1, 1, 100)
+
+        x = xx[::5]
+        y = xx[::5]
+
+        for k in (0, 1, 2, 3):
+            make_interp_spline(x, y, k=k)
+
+    def test_check_finite(self):
+        # check_finite defaults to True; nans and such trigger a ValueError
+        x = np.arange(10).astype(float)
+        y = x**2
+
+        for z in [np.nan, np.inf, -np.inf]:
+            y[-1] = z
+            assert_raises(ValueError, make_interp_spline, x, y)
+
+    @pytest.mark.parametrize('k', [1, 2, 3, 5])
+    def test_list_input(self, k):
+        # regression test for gh-8714: TypeError for x, y being lists and k=2
+        x = list(range(10))
+        y = [a**2 for a in x]
+        make_interp_spline(x, y, k=k)
+
+    def test_multiple_rhs(self):
+        yy = np.c_[np.sin(self.xx), np.cos(self.xx)]
+        der_l = [(1, [1., 2.])]
+        der_r = [(1, [3., 4.])]
+
+        b = make_interp_spline(self.xx, yy, k=3, bc_type=(der_l, der_r))
+        xp_assert_close(b(self.xx), yy, atol=1e-14, rtol=1e-14)
+        xp_assert_close(b(self.xx[0], 1), der_l[0][1], atol=1e-14, rtol=1e-14)
+        xp_assert_close(b(self.xx[-1], 1), der_r[0][1], atol=1e-14, rtol=1e-14)
+
+    def test_shapes(self):
+        rng = np.random.RandomState(1234)
+        k, n = 3, 22
+        x = np.sort(rng.random(size=n))
+        y = rng.random(size=(n, 5, 6, 7))
+
+        b = make_interp_spline(x, y, k)
+        assert b.c.shape == (n, 5, 6, 7)
+
+        # now throw in some derivatives
+        d_l = [(1, rng.random((5, 6, 7)))]
+        d_r = [(1, rng.random((5, 6, 7)))]
+        b = make_interp_spline(x, y, k, bc_type=(d_l, d_r))
+        assert b.c.shape == (n + k - 1, 5, 6, 7)
+
+    def test_string_aliases(self):
+        yy = np.sin(self.xx)
+
+        # a single string is duplicated
+        b1 = make_interp_spline(self.xx, yy, k=3, bc_type='natural')
+        b2 = make_interp_spline(self.xx, yy, k=3, bc_type=([(2, 0)], [(2, 0)]))
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+        # two strings are handled
+        b1 = make_interp_spline(self.xx, yy, k=3,
+                                bc_type=('natural', 'clamped'))
+        b2 = make_interp_spline(self.xx, yy, k=3,
+                                bc_type=([(2, 0)], [(1, 0)]))
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+        # one-sided BCs are OK
+        b1 = make_interp_spline(self.xx, yy, k=2, bc_type=(None, 'clamped'))
+        b2 = make_interp_spline(self.xx, yy, k=2, bc_type=(None, [(1, 0.0)]))
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+        # 'not-a-knot' is equivalent to None
+        b1 = make_interp_spline(self.xx, yy, k=3, bc_type='not-a-knot')
+        b2 = make_interp_spline(self.xx, yy, k=3, bc_type=None)
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+        # unknown strings do not pass
+        with assert_raises(ValueError):
+            make_interp_spline(self.xx, yy, k=3, bc_type='typo')
+
+        # string aliases are handled for 2D values
+        yy = np.c_[np.sin(self.xx), np.cos(self.xx)]
+        der_l = [(1, [0., 0.])]
+        der_r = [(2, [0., 0.])]
+        b2 = make_interp_spline(self.xx, yy, k=3, bc_type=(der_l, der_r))
+        b1 = make_interp_spline(self.xx, yy, k=3,
+                                bc_type=('clamped', 'natural'))
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+        # ... and for N-D values:
+        rng = np.random.RandomState(1234)
+        k, n = 3, 22
+        x = np.sort(rng.random(size=n))
+        y = rng.random(size=(n, 5, 6, 7))
+
+        # now throw in some derivatives
+        d_l = [(1, np.zeros((5, 6, 7)))]
+        d_r = [(1, np.zeros((5, 6, 7)))]
+        b1 = make_interp_spline(x, y, k, bc_type=(d_l, d_r))
+        b2 = make_interp_spline(x, y, k, bc_type='clamped')
+        xp_assert_close(b1.c, b2.c, atol=1e-15)
+
+    def test_full_matrix(self):
+        rng = np.random.RandomState(1234)
+        k, n = 3, 7
+        x = np.sort(rng.random(size=n))
+        y = rng.random(size=n)
+        t = _not_a_knot(x, k)
+
+        b = make_interp_spline(x, y, k, t)
+        cf = make_interp_full_matr(x, y, t, k)
+        xp_assert_close(b.c, cf, atol=1e-14, rtol=1e-14)
+
+    def test_woodbury(self):
+        '''
+        Random elements in diagonal matrix with blocks in the
+        left lower and right upper corners checking the
+        implementation of Woodbury algorithm.
+        '''
+        rng = np.random.RandomState(1234)
+        n = 201
+        for k in range(3, 32, 2):
+            offset = int((k - 1) / 2)
+            a = np.diagflat(rng.random((1, n)))
+            for i in range(1, offset + 1):
+                a[:-i, i:] += np.diagflat(rng.random((1, n - i)))
+                a[i:, :-i] += np.diagflat(rng.random((1, n - i)))
+            ur = rng.random((offset, offset))
+            a[:offset, -offset:] = ur
+            ll = rng.random((offset, offset))
+            a[-offset:, :offset] = ll
+            d = np.zeros((k, n))
+            for i, j in enumerate(range(offset, -offset - 1, -1)):
+                if j < 0:
+                    d[i, :j] = np.diagonal(a, offset=j)
+                else:
+                    d[i, j:] = np.diagonal(a, offset=j)
+            b = rng.random(n)
+            xp_assert_close(_woodbury_algorithm(d, ur, ll, b, k),
+                            np.linalg.solve(a, b), atol=1e-14)
+
+
+def make_interp_full_matr(x, y, t, k):
+    """Assemble an spline order k with knots t to interpolate
+    y(x) using full matrices.
+    Not-a-knot BC only.
+
+    This routine is here for testing only (even though it's functional).
+    """
+    assert x.size == y.size
+    assert t.size == x.size + k + 1
+    n = x.size
+
+    A = np.zeros((n, n), dtype=np.float64)
+
+    for j in range(n):
+        xval = x[j]
+        if xval == t[k]:
+            left = k
+        else:
+            left = np.searchsorted(t, xval) - 1
+
+        # fill a row
+        bb = _dierckx.evaluate_all_bspl(t, k, xval, left)
+        A[j, left-k:left+1] = bb
+
+    c = sl.solve(A, y)
+    return c
+
+
+def make_lsq_full_matrix(x, y, t, k=3):
+    """Make the least-square spline, full matrices."""
+    x, y, t = map(np.asarray, (x, y, t))
+    m = x.size
+    n = t.size - k - 1
+
+    A = np.zeros((m, n), dtype=np.float64)
+
+    for j in range(m):
+        xval = x[j]
+        # find interval
+        if xval == t[k]:
+            left = k
+        else:
+            left = np.searchsorted(t, xval) - 1
+
+        # fill a row
+        bb = _dierckx.evaluate_all_bspl(t, k, xval, left)
+        A[j, left-k:left+1] = bb
+
+    # have observation matrix, can solve the LSQ problem
+    B = np.dot(A.T, A)
+    Y = np.dot(A.T, y)
+    c = sl.solve(B, Y)
+
+    return c, (A, Y)
+
+
+parametrize_lsq_methods = pytest.mark.parametrize("method", ["norm-eq", "qr"])
+
+class TestLSQ:
+    #
+    # Test make_lsq_spline
+    #
+    rng = np.random.RandomState(1234)
+    n, k = 13, 3
+    x = np.sort(rng.random(n))
+    y = rng.random(n)
+    t = _augknt(np.linspace(x[0], x[-1], 7), k)
+
+    @parametrize_lsq_methods
+    def test_lstsq(self, method):
+        # check LSQ construction vs a full matrix version
+        x, y, t, k = self.x, self.y, self.t, self.k
+
+        c0, AY = make_lsq_full_matrix(x, y, t, k)
+        b = make_lsq_spline(x, y, t, k, method=method)
+
+        xp_assert_close(b.c, c0)
+        assert b.c.shape == (t.size - k - 1,)
+
+        # also check against numpy.lstsq
+        aa, yy = AY
+        c1, _, _, _ = np.linalg.lstsq(aa, y, rcond=-1)
+        xp_assert_close(b.c, c1)
+
+    @parametrize_lsq_methods
+    def test_weights(self, method):
+        # weights = 1 is same as None
+        x, y, t, k = self.x, self.y, self.t, self.k
+        w = np.ones_like(x)
+
+        b = make_lsq_spline(x, y, t, k, method=method)
+        b_w = make_lsq_spline(x, y, t, k, w=w, method=method)
+
+        xp_assert_close(b.t, b_w.t, atol=1e-14)
+        xp_assert_close(b.c, b_w.c, atol=1e-14)
+        assert b.k == b_w.k
+
+    def test_weights_same(self):
+        # both methods treat weights
+        x, y, t, k = self.x, self.y, self.t, self.k
+        w = np.random.default_rng(1234).uniform(size=x.shape[0])
+
+        b_ne = make_lsq_spline(x, y, t, k, w=w, method="norm-eq")
+        b_qr = make_lsq_spline(x, y, t, k, w=w, method="qr")
+        b_no_w = make_lsq_spline(x, y, t, k, method="qr")
+
+        xp_assert_close(b_ne.c, b_qr.c, atol=1e-14)
+        assert not np.allclose(b_no_w.c, b_qr.c, atol=1e-14)
+
+    @parametrize_lsq_methods
+    def test_multiple_rhs(self, method):
+        x, t, k, n = self.x, self.t, self.k, self.n
+        rng = np.random.RandomState(1234)
+        y = rng.random(size=(n, 5, 6, 7))
+        b = make_lsq_spline(x, y, t, k, method=method)
+        assert b.c.shape == (t.size-k-1, 5, 6, 7)
+
+    @parametrize_lsq_methods
+    def test_multiple_rhs_2(self, method):
+        x, t, k, n = self.x, self.t, self.k, self.n
+        nrhs = 3
+        rng = np.random.RandomState(1234)
+        y = rng.random(size=(n, nrhs))
+        b = make_lsq_spline(x, y, t, k, method=method)
+
+        bb = [make_lsq_spline(x, y[:, i], t, k, method=method)
+              for i in range(nrhs)]
+        coefs = np.vstack([bb[i].c for i in range(nrhs)]).T
+
+        xp_assert_close(coefs, b.c, atol=1e-15)
+
+    def test_multiple_rhs_3(self):
+        x, t, k, n = self.x, self.t, self.k, self.n
+        nrhs = 3
+        y = np.random.random(size=(n, nrhs))
+        b_qr = make_lsq_spline(x, y, t, k, method="qr")
+        b_neq = make_lsq_spline(x, y, t, k, method="norm-eq")
+        xp_assert_close(b_qr.c, b_neq.c, atol=1e-15)
+
+    @parametrize_lsq_methods
+    def test_complex(self, method):
+        # cmplx-valued `y`
+        x, t, k = self.x, self.t, self.k
+        yc = self.y * (1. + 2.j)
+
+        b = make_lsq_spline(x, yc, t, k, method=method)
+        b_re = make_lsq_spline(x, yc.real, t, k, method=method)
+        b_im = make_lsq_spline(x, yc.imag, t, k, method=method)
+
+        xp_assert_close(b(x), b_re(x) + 1.j*b_im(x), atol=1e-15, rtol=1e-15)
+
+    def test_complex_2(self):
+        # test complex-valued y with y.ndim > 1
+
+        x, t, k = self.x, self.t, self.k
+        yc = self.y * (1. + 2.j)
+        yc = np.stack((yc, yc), axis=1)
+
+        b = make_lsq_spline(x, yc, t, k)
+        b_re = make_lsq_spline(x, yc.real, t, k)
+        b_im = make_lsq_spline(x, yc.imag, t, k)
+
+        xp_assert_close(b(x), b_re(x) + 1.j*b_im(x), atol=1e-15, rtol=1e-15)
+
+        # repeat with num_trailing_dims > 1 : yc.shape[1:] = (2, 2)
+        yc = np.stack((yc, yc), axis=1)
+
+        b = make_lsq_spline(x, yc, t, k)
+        b_re = make_lsq_spline(x, yc.real, t, k)
+        b_im = make_lsq_spline(x, yc.imag, t, k)
+
+        xp_assert_close(b(x), b_re(x) + 1.j*b_im(x), atol=1e-15, rtol=1e-15)
+
+    @parametrize_lsq_methods
+    def test_int_xy(self, method):
+        x = np.arange(10).astype(int)
+        y = np.arange(10).astype(int)
+        t = _augknt(x, k=1)
+        # Cython chokes on "buffer type mismatch"
+        make_lsq_spline(x, y, t, k=1, method=method)
+
+    @parametrize_lsq_methods
+    def test_f32_xy(self, method):
+        x = np.arange(10, dtype=np.float32)
+        y = np.arange(10, dtype=np.float32)
+        t = _augknt(x, k=1)
+        spl_f32 = make_lsq_spline(x, y, t, k=1, method=method)
+        spl_f64 = make_lsq_spline(
+            x.astype(float), y.astype(float), t.astype(float), k=1, method=method
+        )
+
+        x2 = (x[1:] + x[:-1]) / 2.0
+        xp_assert_close(spl_f32(x2), spl_f64(x2), atol=1e-15)
+
+    @parametrize_lsq_methods
+    def test_sliced_input(self, method):
+        # Cython code chokes on non C contiguous arrays
+        xx = np.linspace(-1, 1, 100)
+
+        x = xx[::3]
+        y = xx[::3]
+        t = _augknt(x, 1)
+        make_lsq_spline(x, y, t, k=1, method=method)
+
+    @parametrize_lsq_methods
+    def test_checkfinite(self, method):
+        # check_finite defaults to True; nans and such trigger a ValueError
+        x = np.arange(12).astype(float)
+        y = x**2
+        t = _augknt(x, 3)
+
+        for z in [np.nan, np.inf, -np.inf]:
+            y[-1] = z
+            assert_raises(ValueError, make_lsq_spline, x, y, t, method=method)
+
+    @parametrize_lsq_methods
+    def test_read_only(self, method):
+        # Check that make_lsq_spline works with read only arrays
+        x, y, t = self.x, self.y, self.t
+        x.setflags(write=False)
+        y.setflags(write=False)
+        t.setflags(write=False)
+        make_lsq_spline(x=x, y=y, t=t, method=method)
+
+    @pytest.mark.parametrize('k', list(range(1, 7)))
+    def test_qr_vs_norm_eq(self, k):
+        # check that QR and normal eq solutions match
+        x, y = self.x, self.y
+        t = _augknt(np.linspace(x[0], x[-1], 7), k)
+        spl_norm_eq = make_lsq_spline(x, y, t, k=k, method='norm-eq')
+        spl_qr = make_lsq_spline(x, y, t, k=k, method='qr')
+
+        xx = (x[1:] + x[:-1]) / 2.0
+        xp_assert_close(spl_norm_eq(xx), spl_qr(xx), atol=1e-15)
+
+    def test_duplicates(self):
+        # method="qr" can handle duplicated data points
+        x = np.repeat(self.x, 2)
+        y = np.repeat(self.y, 2)
+        spl_1 = make_lsq_spline(self.x, self.y, self.t, k=3, method='qr')
+        spl_2 = make_lsq_spline(x, y, self.t, k=3, method='qr')
+
+        xx = (x[1:] + x[:-1]) / 2.0
+        xp_assert_close(spl_1(xx), spl_2(xx), atol=1e-15)
+
+
+class PackedMatrix:
+    """A simplified CSR format for when non-zeros in each row are consecutive.
+
+    Assuming that each row of an `(m, nc)` matrix 1) only has `nz` non-zeros, and
+    2) these non-zeros are consecutive, we only store an `(m, nz)` matrix of
+    non-zeros and a 1D array of row offsets. This way, a row `i` of the original
+    matrix A is ``A[i, offset[i]: offset[i] + nz]``.
+
+    """
+    def __init__(self, a, offset, nc):
+        self.a = a
+        self.offset = offset
+        self.nc = nc
+
+        assert a.ndim == 2
+        assert offset.ndim == 1
+        assert a.shape[0] == offset.shape[0]
+
+    @property
+    def shape(self):
+        return self.a.shape[0], self.nc
+
+    def todense(self):
+        out = np.zeros(self.shape)
+        nelem = self.a.shape[1]
+        for i in range(out.shape[0]):
+            nel = min(self.nc - self.offset[i], nelem)
+            out[i, self.offset[i]:self.offset[i] + nel] = self.a[i, :nel]
+        return out
+
+
+def _qr_reduce_py(a_p, y, startrow=1):
+    """This is a python counterpart of the `_qr_reduce` routine,
+    declared in interpolate/src/__fitpack.h
+    """
+    from scipy.linalg.lapack import dlartg
+
+    # unpack the packed format
+    a = a_p.a
+    offset = a_p.offset
+    nc = a_p.nc
+
+    m, nz = a.shape
+
+    assert y.shape[0] == m
+    R = a.copy()
+    y1 = y.copy()
+
+    for i in range(startrow, m):
+        oi = offset[i]
+        for j in range(oi, nc):
+            # rotate only the lower diagonal
+            if j >= min(i, nc):
+                break
+
+            # In dense format: diag a1[j, j] vs a1[i, j]
+            c, s, r = dlartg(R[j, 0], R[i, 0])
+
+            # rotate l.h.s.
+            R[j, 0] = r
+            for l in range(1, nz):
+                R[j, l], R[i, l-1] = fprota(c, s, R[j, l], R[i, l])
+            R[i, -1] = 0.0
+
+            # rotate r.h.s.
+            for l in range(y1.shape[1]):
+                y1[j, l], y1[i, l] = fprota(c, s, y1[j, l], y1[i, l])
+
+    # convert to packed
+    offs = list(range(R.shape[0]))
+    R_p = PackedMatrix(R, np.array(offs, dtype=np.int64), nc)
+
+    return R_p, y1
+
+
+def fprota(c, s, a, b):
+    """Givens rotate [a, b].
+
+    [aa] = [ c s] @ [a]
+    [bb]   [-s c]   [b]
+
+    """
+    aa =  c*a + s*b
+    bb = -s*a + c*b
+    return aa, bb
+
+
+def fpback(R_p, y):
+    """Backsubsitution solve upper triangular banded `R @ c = y.`
+
+    `R` is in the "packed" format: `R[i, :]` is `a[i, i:i+k+1]`
+    """
+    R = R_p.a
+    _, nz = R.shape
+    nc = R_p.nc
+    assert y.shape[0] == R.shape[0]
+
+    c = np.zeros_like(y[:nc])
+    c[nc-1, ...] = y[nc-1] / R[nc-1, 0]
+    for i in range(nc-2, -1, -1):
+        nel = min(nz, nc-i)
+        # NB: broadcast R across trailing dimensions of `c`.
+        summ = (R[i, 1:nel, None] * c[i+1:i+nel, ...]).sum(axis=0)
+        c[i, ...] = ( y[i] - summ ) / R[i, 0]
+    return c
+
+
+class TestGivensQR:
+    # Test row-by-row QR factorization, used for the LSQ spline construction.
+    # This is implementation detail; still test it separately.
+    def _get_xyt(self, n):
+        k = 3
+        x = np.arange(n, dtype=float)
+        y = x**3 + 1/(1+x)
+        t = _not_a_knot(x, k)
+        return x, y, t, k
+
+    def test_vs_full(self):
+        n = 10
+        x, y, t, k = self._get_xyt(n)
+
+        # design matrix
+        a_csr = BSpline.design_matrix(x, t, k)
+
+        # dense QR
+        q, r = sl.qr(a_csr.todense())
+        qTy = q.T @ y
+
+        # prepare the PackedMatrix to factorize
+        # convert to "packed" format
+        m, nc = a_csr.shape
+        assert nc == t.shape[0] - k - 1
+
+        offset = a_csr.indices[::(k+1)]
+        offset = np.ascontiguousarray(offset, dtype=np.int64)
+        A = a_csr.data.reshape(m, k+1)
+
+        R = PackedMatrix(A, offset, nc)
+        y_ = y[:, None]     # _qr_reduce requires `y` a 2D array
+        _dierckx.qr_reduce(A, offset, nc, y_)      # modifies arguments in-place
+
+        # signs may differ
+        xp_assert_close(np.minimum(R.todense() + r,
+                                   R.todense() - r), np.zeros_like(r), atol=1e-15)
+        xp_assert_close(np.minimum(abs(qTy - y_[:, 0]),
+                                   abs(qTy + y_[:, 0])), np.zeros_like(qTy), atol=2e-13)
+
+        # sign changes are consistent between Q and R:
+        c_full = sl.solve(r, qTy)
+        c_banded = _dierckx.fpback(R.a, R.nc, y_)
+        xp_assert_close(c_full, c_banded[:, 0], atol=5e-13)
+
+    def test_py_vs_compiled(self):
+        # test _qr_reduce vs a python implementation
+        n = 10
+        x, y, t, k = self._get_xyt(n)
+
+        # design matrix
+        a_csr = BSpline.design_matrix(x, t, k)
+        m, nc = a_csr.shape
+        assert nc == t.shape[0] - k - 1
+
+        offset = a_csr.indices[::(k+1)]
+        offset = np.ascontiguousarray(offset, dtype=np.int64)
+        A = a_csr.data.reshape(m, k+1)
+
+        R = PackedMatrix(A, offset, nc)
+        y_ = y[:, None]
+
+        RR, yy = _qr_reduce_py(R, y_)
+        _dierckx.qr_reduce(A, offset, nc , y_)   # in-place
+
+        xp_assert_close(RR.a, R.a, atol=1e-15)
+        xp_assert_equal(RR.offset, R.offset, check_dtype=False)
+        assert RR.nc == R.nc
+        xp_assert_close(yy, y_, atol=1e-15)
+
+    # Test C-level construction of the design matrix
+
+    def test_data_matrix(self):
+        n = 10
+        x, y, t, k = self._get_xyt(n)
+        w = np.arange(1, n+1, dtype=float)
+
+        A, offset, nc = _dierckx.data_matrix(x, t, k, w)
+
+        m = x.shape[0]
+        a_csr = BSpline.design_matrix(x, t, k)
+        a_w = (a_csr * w[:, None]).tocsr()
+        A_ = a_w.data.reshape((m, k+1))
+        offset_ = a_w.indices[::(k+1)].astype(np.int64)
+
+        xp_assert_close(A, A_, atol=1e-15)
+        xp_assert_equal(offset, offset_)
+        assert nc == t.shape[0] - k - 1
+
+    def test_fpback(self):
+        n = 10
+        x, y, t, k = self._get_xyt(n)
+        y = np.c_[y, y**2]
+        A, offset, nc = _dierckx.data_matrix(x, t, k, np.ones_like(x))
+        R = PackedMatrix(A, offset, nc)
+        _dierckx.qr_reduce(A, offset, nc, y)
+
+        c = fpback(R, y)
+        cc = _dierckx.fpback(A, nc, y)
+
+        xp_assert_close(cc, c, atol=1e-14)
+
+
+def data_file(basename):
+    return os.path.join(os.path.abspath(os.path.dirname(__file__)),
+                        'data', basename)
+
+
+class TestSmoothingSpline:
+    #
+    # test make_smoothing_spline
+    #
+    def test_invalid_input(self):
+        rng = np.random.RandomState(1234)
+        n = 100
+        x = np.sort(rng.random_sample(n) * 4 - 2)
+        y = x**2 * np.sin(4 * x) + x**3 + rng.normal(0., 1.5, n)
+
+        # ``x`` and ``y`` should have same shapes (1-D array)
+        with assert_raises(ValueError):
+            make_smoothing_spline(x, y[1:])
+        with assert_raises(ValueError):
+            make_smoothing_spline(x[1:], y)
+        with assert_raises(ValueError):
+            make_smoothing_spline(x.reshape(1, n), y)
+
+        # ``x`` should be an ascending array
+        with assert_raises(ValueError):
+            make_smoothing_spline(x[::-1], y)
+
+        x_dupl = np.copy(x)
+        x_dupl[0] = x_dupl[1]
+
+        with assert_raises(ValueError):
+            make_smoothing_spline(x_dupl, y)
+
+        # x and y length must be >= 5
+        x = np.arange(4)
+        y = np.ones(4)
+        exception_message = "``x`` and ``y`` length must be at least 5"
+        with pytest.raises(ValueError, match=exception_message):
+            make_smoothing_spline(x, y)
+
+    def test_compare_with_GCVSPL(self):
+        """
+        Data is generated in the following way:
+        >>> np.random.seed(1234)
+        >>> n = 100
+        >>> x = np.sort(np.random.random_sample(n) * 4 - 2)
+        >>> y = np.sin(x) + np.random.normal(scale=.5, size=n)
+        >>> np.savetxt('x.csv', x)
+        >>> np.savetxt('y.csv', y)
+
+        We obtain the result of performing the GCV smoothing splines
+        package (by Woltring, gcvspl) on the sample data points
+        using its version for Octave (https://github.com/srkuberski/gcvspl).
+        In order to use this implementation, one should clone the repository
+        and open the folder in Octave.
+        In Octave, we load up ``x`` and ``y`` (generated from Python code
+        above):
+
+        >>> x = csvread('x.csv');
+        >>> y = csvread('y.csv');
+
+        Then, in order to access the implementation, we compile gcvspl files in
+        Octave:
+
+        >>> mex gcvsplmex.c gcvspl.c
+        >>> mex spldermex.c gcvspl.c
+
+        The first function computes the vector of unknowns from the dataset
+        (x, y) while the second one evaluates the spline in certain points
+        with known vector of coefficients.
+
+        >>> c = gcvsplmex( x, y, 2 );
+        >>> y0 = spldermex( x, c, 2, x, 0 );
+
+        If we want to compare the results of the gcvspl code, we can save
+        ``y0`` in csv file:
+
+        >>> csvwrite('y0.csv', y0);
+
+        """
+        # load the data sample
+        with np.load(data_file('gcvspl.npz')) as data:
+            # data points
+            x = data['x']
+            y = data['y']
+
+            y_GCVSPL = data['y_GCVSPL']
+        y_compr = make_smoothing_spline(x, y)(x)
+
+        # such tolerance is explained by the fact that the spline is built
+        # using an iterative algorithm for minimizing the GCV criteria. These
+        # algorithms may vary, so the tolerance should be rather low.
+        # Not checking dtypes as gcvspl.npz stores little endian arrays, which
+        # result in conflicting dtypes on big endian systems.
+        xp_assert_close(y_compr, y_GCVSPL, atol=1e-4, rtol=1e-4, check_dtype=False)
+
+    def test_non_regularized_case(self):
+        """
+        In case the regularization parameter is 0, the resulting spline
+        is an interpolation spline with natural boundary conditions.
+        """
+        # create data sample
+        rng = np.random.RandomState(1234)
+        n = 100
+        x = np.sort(rng.random_sample(n) * 4 - 2)
+        y = x**2 * np.sin(4 * x) + x**3 + rng.normal(0., 1.5, n)
+
+        spline_GCV = make_smoothing_spline(x, y, lam=0.)
+        spline_interp = make_interp_spline(x, y, 3, bc_type='natural')
+
+        grid = np.linspace(x[0], x[-1], 2 * n)
+        xp_assert_close(spline_GCV(grid),
+                        spline_interp(grid),
+                        atol=1e-15)
+
+    @pytest.mark.fail_slow(2)
+    def test_weighted_smoothing_spline(self):
+        # create data sample
+        rng = np.random.RandomState(1234)
+        n = 100
+        x = np.sort(rng.random_sample(n) * 4 - 2)
+        y = x**2 * np.sin(4 * x) + x**3 + rng.normal(0., 1.5, n)
+
+        spl = make_smoothing_spline(x, y)
+
+        # in order not to iterate over all of the indices, we select 10 of
+        # them randomly
+        for ind in rng.choice(range(100), size=10):
+            w = np.ones(n)
+            w[ind] = 30.
+            spl_w = make_smoothing_spline(x, y, w)
+            # check that spline with weight in a certain point is closer to the
+            # original point than the one without weights
+            orig = abs(spl(x[ind]) - y[ind])
+            weighted = abs(spl_w(x[ind]) - y[ind])
+
+            if orig < weighted:
+                raise ValueError(f'Spline with weights should be closer to the'
+                                 f' points than the original one: {orig:.4} < '
+                                 f'{weighted:.4}')
+
+
+################################
+# NdBSpline tests
+def bspline2(xy, t, c, k):
+    """A naive 2D tensort product spline evaluation."""
+    x, y = xy
+    tx, ty = t
+    nx = len(tx) - k - 1
+    assert (nx >= k+1)
+    ny = len(ty) - k - 1
+    assert (ny >= k+1)
+    res = sum(c[ix, iy] * B(x, k, ix, tx) * B(y, k, iy, ty)
+              for ix in range(nx) for iy in range(ny))
+    return np.asarray(res)
+
+
+def B(x, k, i, t):
+    if k == 0:
+        return 1.0 if t[i] <= x < t[i+1] else 0.0
+    if t[i+k] == t[i]:
+        c1 = 0.0
+    else:
+        c1 = (x - t[i])/(t[i+k] - t[i]) * B(x, k-1, i, t)
+    if t[i+k+1] == t[i+1]:
+        c2 = 0.0
+    else:
+        c2 = (t[i+k+1] - x)/(t[i+k+1] - t[i+1]) * B(x, k-1, i+1, t)
+    return c1 + c2
+
+
+def bspline(x, t, c, k):
+    n = len(t) - k - 1
+    assert (n >= k+1) and (len(c) >= n)
+    return sum(c[i] * B(x, k, i, t) for i in range(n))
+
+
+class NdBSpline0:
+    def __init__(self, t, c, k=3):
+        """Tensor product spline object.
+
+        c[i1, i2, ..., id] * B(x1, i1) * B(x2, i2) * ... * B(xd, id)
+
+        Parameters
+        ----------
+        c : ndarray, shape (n1, n2, ..., nd, ...)
+            b-spline coefficients
+        t : tuple of 1D ndarrays
+            knot vectors in directions 1, 2, ... d
+            ``len(t[i]) == n[i] + k + 1``
+        k : int or length-d tuple of integers
+            spline degrees.
+        """
+        ndim = len(t)
+        assert ndim <= len(c.shape)
+
+        try:
+            len(k)
+        except TypeError:
+            # make k a tuple
+            k = (k,)*ndim
+
+        self.k = tuple(operator.index(ki) for ki in k)
+        self.t = tuple(np.asarray(ti, dtype=float) for ti in t)
+        self.c = c
+
+    def __call__(self, x):
+        ndim = len(self.t)
+        # a single evaluation point: `x` is a 1D array_like, shape (ndim,)
+        assert len(x) == ndim
+
+        # get the indices in an ndim-dimensional vector
+        i = ['none', ]*ndim
+        for d in range(ndim):
+            td, xd = self.t[d], x[d]
+            k = self.k[d]
+
+            # find the index for x[d]
+            if xd == td[k]:
+                i[d] = k
+            else:
+                i[d] = np.searchsorted(td, xd) - 1
+            assert td[i[d]] <= xd <= td[i[d]+1]
+            assert i[d] >= k and i[d] < len(td) - k
+        i = tuple(i)
+
+        # iterate over the dimensions, form linear combinations of
+        # products B(x_1) * B(x_2) * ... B(x_N) of (k+1)**N b-splines
+        # which are non-zero at `i = (i_1, i_2, ..., i_N)`.
+        result = 0
+        iters = [range(i[d] - self.k[d], i[d] + 1) for d in range(ndim)]
+        for idx in itertools.product(*iters):
+            term = self.c[idx] * np.prod([B(x[d], self.k[d], idx[d], self.t[d])
+                                          for d in range(ndim)])
+            result += term
+        return np.asarray(result)
+
+
+class TestNdBSpline:
+
+    def test_1D(self):
+        # test ndim=1 agrees with BSpline
+        rng = np.random.default_rng(12345)
+        n, k = 11, 3
+        n_tr = 7
+        t = np.sort(rng.uniform(size=n + k + 1))
+        c = rng.uniform(size=(n, n_tr))
+
+        b = BSpline(t, c, k)
+        nb = NdBSpline((t,), c, k)
+
+        xi = rng.uniform(size=21)
+        # NdBSpline expects xi.shape=(npts, ndim)
+        xp_assert_close(nb(xi[:, None]),
+                        b(xi), atol=1e-14)
+        assert nb(xi[:, None]).shape == (xi.shape[0], c.shape[1])
+
+    def make_2d_case(self):
+        # make a 2D separable spline
+        x = np.arange(6)
+        y = x**3
+        spl = make_interp_spline(x, y, k=3)
+
+        y_1 = x**3 + 2*x
+        spl_1 = make_interp_spline(x, y_1, k=3)
+
+        t2 = (spl.t, spl_1.t)
+        c2 = spl.c[:, None] * spl_1.c[None, :]
+
+        return t2, c2, 3
+
+    def make_2d_mixed(self):
+        # make a 2D separable spline w/ kx=3, ky=2
+        x = np.arange(6)
+        y = x**3
+        spl = make_interp_spline(x, y, k=3)
+
+        x = np.arange(5) + 1.5
+        y_1 = x**2 + 2*x
+        spl_1 = make_interp_spline(x, y_1, k=2)
+
+        t2 = (spl.t, spl_1.t)
+        c2 = spl.c[:, None] * spl_1.c[None, :]
+
+        return t2, c2, spl.k, spl_1.k
+
+    def test_2D_separable(self):
+        xi = [(1.5, 2.5), (2.5, 1), (0.5, 1.5)]
+        t2, c2, k = self.make_2d_case()
+        target = [x**3 * (y**3 + 2*y) for (x, y) in xi]
+
+        # sanity check: bspline2 gives the product as constructed
+        xp_assert_close(np.asarray([bspline2(xy, t2, c2, k) for xy in xi]),
+                        np.asarray(target),
+                        check_shape=False,
+                        atol=1e-14)
+
+        # check evaluation on a 2D array: the 1D array of 2D points
+        bspl2 = NdBSpline(t2, c2, k=3)
+        assert bspl2(xi).shape == (len(xi), )
+        xp_assert_close(bspl2(xi),
+                        target, atol=1e-14)
+
+        # test that a nan in -> nan out
+        xi = np.asarray(xi)
+        xi[0, 1] = np.nan
+        xp_assert_equal(np.isnan(bspl2(xi)), np.asarray([True, False, False]))
+
+        # now check on a multidim xi
+        rng = np.random.default_rng(12345)
+        xi = rng.uniform(size=(4, 3, 2)) * 5
+        result = bspl2(xi)
+        assert result.shape == (4, 3)
+
+        # also check the values
+        x, y = xi.reshape((-1, 2)).T
+        xp_assert_close(result.ravel(),
+                        x**3 * (y**3 + 2*y), atol=1e-14)
+
+    def test_2D_separable_2(self):
+        # test `c` with trailing dimensions, i.e. c.ndim > ndim
+        ndim = 2
+        xi = [(1.5, 2.5), (2.5, 1), (0.5, 1.5)]
+        target = [x**3 * (y**3 + 2*y) for (x, y) in xi]
+
+        t2, c2, k = self.make_2d_case()
+        c2_4 = np.dstack((c2, c2, c2, c2))   # c22.shape = (6, 6, 4)
+
+        xy = (1.5, 2.5)
+        bspl2_4 = NdBSpline(t2, c2_4, k=3)
+        result = bspl2_4(xy)
+        val_single = NdBSpline(t2, c2, k)(xy)
+        assert result.shape == (4,)
+        xp_assert_close(result,
+                        [val_single, ]*4, atol=1e-14)
+
+        # now try the array xi : the output.shape is (3, 4) where 3
+        # is the number of points in xi and 4 is the trailing dimension of c
+        assert bspl2_4(xi).shape == np.shape(xi)[:-1] + bspl2_4.c.shape[ndim:]
+        xp_assert_close(bspl2_4(xi),  np.asarray(target)[:, None],
+                        check_shape=False,
+                        atol=5e-14)
+
+        # two trailing dimensions
+        c2_22 = c2_4.reshape((6, 6, 2, 2))
+        bspl2_22 = NdBSpline(t2, c2_22, k=3)
+
+        result = bspl2_22(xy)
+        assert result.shape == (2, 2)
+        xp_assert_close(result,
+                        [[val_single, val_single],
+                         [val_single, val_single]], atol=1e-14)
+
+        # now try the array xi : the output shape is (3, 2, 2)
+        # for 3 points in xi and c trailing dimensions being (2, 2)
+        assert (bspl2_22(xi).shape ==
+                np.shape(xi)[:-1] + bspl2_22.c.shape[ndim:])
+        xp_assert_close(bspl2_22(xi), np.asarray(target)[:, None, None],
+                        check_shape=False,
+                        atol=5e-14)
+
+
+    def test_2D_separable_2_complex(self):
+        # test `c` with c.dtype == complex, with and w/o trailing dims
+        xi = [(1.5, 2.5), (2.5, 1), (0.5, 1.5)]
+        target = [x**3 * (y**3 + 2*y) for (x, y) in xi]
+
+        target = [t + 2j*t for t in target]
+
+        t2, c2, k = self.make_2d_case()
+        c2 = c2 * (1 + 2j)
+        c2_4 = np.dstack((c2, c2, c2, c2))   # c2_4.shape = (6, 6, 4)
+
+        xy = (1.5, 2.5)
+        bspl2_4 = NdBSpline(t2, c2_4, k=3)
+        result = bspl2_4(xy)
+        val_single = NdBSpline(t2, c2, k)(xy)
+        assert result.shape == (4,)
+        xp_assert_close(result,
+                        [val_single, ]*4, atol=1e-14)
+
+    def test_2D_random(self):
+        rng = np.random.default_rng(12345)
+        k = 3
+        tx = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7)) * 3, 3, 3, 3, 3]
+        ty = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        c = rng.uniform(size=(tx.size-k-1, ty.size-k-1))
+
+        spl = NdBSpline((tx, ty), c, k=k)
+
+        xi = (1., 1.)
+        xp_assert_close(spl(xi),
+                        bspline2(xi, (tx, ty), c, k), atol=1e-14)
+
+        xi = np.c_[[1, 1.5, 2],
+                   [1.1, 1.6, 2.1]]
+        xp_assert_close(spl(xi),
+                        [bspline2(xy, (tx, ty), c, k) for xy in xi],
+                        atol=1e-14)
+
+    def test_2D_mixed(self):
+        t2, c2, kx, ky = self.make_2d_mixed()
+        xi = [(1.4, 4.5), (2.5, 2.4), (4.5, 3.5)]
+        target = [x**3 * (y**2 + 2*y) for (x, y) in xi]
+        bspl2 = NdBSpline(t2, c2, k=(kx, ky))
+        assert bspl2(xi).shape == (len(xi), )
+        xp_assert_close(bspl2(xi),
+                        target, atol=1e-14)
+
+    def test_2D_derivative(self):
+        t2, c2, kx, ky = self.make_2d_mixed()
+        xi = [(1.4, 4.5), (2.5, 2.4), (4.5, 3.5)]
+        bspl2 = NdBSpline(t2, c2, k=(kx, ky))
+
+        der = bspl2(xi, nu=(1, 0))
+        xp_assert_close(der,
+                        [3*x**2 * (y**2 + 2*y) for x, y in xi], atol=1e-14)
+
+        der = bspl2(xi, nu=(1, 1))
+        xp_assert_close(der,
+                        [3*x**2 * (2*y + 2) for x, y in xi], atol=1e-14)
+
+        der = bspl2(xi, nu=(0, 0))
+        xp_assert_close(der,
+                        [x**3 * (y**2 + 2*y) for x, y in xi], atol=1e-14)
+
+        with assert_raises(ValueError):
+            # all(nu >= 0)
+            der = bspl2(xi, nu=(-1, 0))
+
+        with assert_raises(ValueError):
+            # len(nu) == ndim
+            der = bspl2(xi, nu=(-1, 0, 1))
+
+    def test_2D_mixed_random(self):
+        rng = np.random.default_rng(12345)
+        kx, ky = 2, 3
+        tx = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7)) * 3, 3, 3, 3, 3]
+        ty = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        c = rng.uniform(size=(tx.size - kx - 1, ty.size - ky - 1))
+
+        xi = np.c_[[1, 1.5, 2],
+                   [1.1, 1.6, 2.1]]
+
+        bspl2 = NdBSpline((tx, ty), c, k=(kx, ky))
+        bspl2_0 = NdBSpline0((tx, ty), c, k=(kx, ky))
+
+        xp_assert_close(bspl2(xi),
+                        [bspl2_0(xp) for xp in xi], atol=1e-14)
+
+    def test_tx_neq_ty(self):
+        # 2D separable spline w/ len(tx) != len(ty)
+        x = np.arange(6)
+        y = np.arange(7) + 1.5
+
+        spl_x = make_interp_spline(x, x**3, k=3)
+        spl_y = make_interp_spline(y, y**2 + 2*y, k=3)
+        cc = spl_x.c[:, None] * spl_y.c[None, :]
+        bspl = NdBSpline((spl_x.t, spl_y.t), cc, (spl_x.k, spl_y.k))
+
+        values = (x**3)[:, None] * (y**2 + 2*y)[None, :]
+        rgi = RegularGridInterpolator((x, y), values)
+
+        xi = [(a, b) for a, b in itertools.product(x, y)]
+        bxi = bspl(xi)
+
+        assert not np.isnan(bxi).any()
+        xp_assert_close(bxi, rgi(xi), atol=1e-14)
+        xp_assert_close(bxi.reshape(values.shape), values, atol=1e-14)
+
+    def make_3d_case(self):
+        # make a 3D separable spline
+        x = np.arange(6)
+        y = x**3
+        spl = make_interp_spline(x, y, k=3)
+
+        y_1 = x**3 + 2*x
+        spl_1 = make_interp_spline(x, y_1, k=3)
+
+        y_2 = x**3 + 3*x + 1
+        spl_2 = make_interp_spline(x, y_2, k=3)
+
+        t2 = (spl.t, spl_1.t, spl_2.t)
+        c2 = (spl.c[:, None, None] *
+              spl_1.c[None, :, None] *
+              spl_2.c[None, None, :])
+
+        return t2, c2, 3
+
+    def test_3D_separable(self):
+        rng = np.random.default_rng(12345)
+        x, y, z = rng.uniform(size=(3, 11)) * 5
+        target = x**3 * (y**3 + 2*y) * (z**3 + 3*z + 1)
+
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3)
+
+        xi = [_ for _ in zip(x, y, z)]
+        result = bspl3(xi)
+        assert result.shape == (11,)
+        xp_assert_close(result, target, atol=1e-14)
+
+    def test_3D_derivative(self):
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3)
+        rng = np.random.default_rng(12345)
+        x, y, z = rng.uniform(size=(3, 11)) * 5
+        xi = [_ for _ in zip(x, y, z)]
+
+        xp_assert_close(bspl3(xi, nu=(1, 0, 0)),
+                        3*x**2 * (y**3 + 2*y) * (z**3 + 3*z + 1), atol=1e-14)
+
+        xp_assert_close(bspl3(xi, nu=(2, 0, 0)),
+                        6*x * (y**3 + 2*y) * (z**3 + 3*z + 1), atol=1e-14)
+
+        xp_assert_close(bspl3(xi, nu=(2, 1, 0)),
+                        6*x * (3*y**2 + 2) * (z**3 + 3*z + 1), atol=1e-14)
+
+        xp_assert_close(bspl3(xi, nu=(2, 1, 3)),
+                        6*x * (3*y**2 + 2) * (6), atol=1e-14)
+
+        xp_assert_close(bspl3(xi, nu=(2, 1, 4)),
+                        np.zeros(len(xi)), atol=1e-14)
+
+    def test_3D_random(self):
+        rng = np.random.default_rng(12345)
+        k = 3
+        tx = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7)) * 3, 3, 3, 3, 3]
+        ty = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        tz = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        c = rng.uniform(size=(tx.size-k-1, ty.size-k-1, tz.size-k-1))
+
+        spl = NdBSpline((tx, ty, tz), c, k=k)
+        spl_0 = NdBSpline0((tx, ty, tz), c, k=k)
+
+        xi = (1., 1., 1)
+        xp_assert_close(spl(xi), spl_0(xi), atol=1e-14)
+
+        xi = np.c_[[1, 1.5, 2],
+                   [1.1, 1.6, 2.1],
+                   [0.9, 1.4, 1.9]]
+        xp_assert_close(spl(xi), [spl_0(xp) for xp in xi], atol=1e-14)
+
+    def test_3D_random_complex(self):
+        rng = np.random.default_rng(12345)
+        k = 3
+        tx = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7)) * 3, 3, 3, 3, 3]
+        ty = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        tz = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        c = (rng.uniform(size=(tx.size-k-1, ty.size-k-1, tz.size-k-1)) +
+             rng.uniform(size=(tx.size-k-1, ty.size-k-1, tz.size-k-1))*1j)
+
+        spl = NdBSpline((tx, ty, tz), c, k=k)
+        spl_re = NdBSpline((tx, ty, tz), c.real, k=k)
+        spl_im = NdBSpline((tx, ty, tz), c.imag, k=k)
+
+        xi = np.c_[[1, 1.5, 2],
+                   [1.1, 1.6, 2.1],
+                   [0.9, 1.4, 1.9]]
+        xp_assert_close(spl(xi),
+                        spl_re(xi) + 1j*spl_im(xi), atol=1e-14)
+
+    @pytest.mark.parametrize('cls_extrap', [None, True])
+    @pytest.mark.parametrize('call_extrap', [None, True])
+    def test_extrapolate_3D_separable(self, cls_extrap, call_extrap):
+        # test that extrapolate=True does extrapolate
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3, extrapolate=cls_extrap)
+
+        # evaluate out of bounds
+        x, y, z = [-2, -1, 7], [-3, -0.5, 6.5], [-1, -1.5, 7.5]
+        x, y, z = map(np.asarray, (x, y, z))
+        xi = [_ for _ in zip(x, y, z)]
+        target = x**3 * (y**3 + 2*y) * (z**3 + 3*z + 1)
+
+        result = bspl3(xi, extrapolate=call_extrap)
+        xp_assert_close(result, target, atol=1e-14)
+
+    @pytest.mark.parametrize('extrap', [(False, True), (True, None)])
+    def test_extrapolate_3D_separable_2(self, extrap):
+        # test that call(..., extrapolate=None) defers to self.extrapolate,
+        # otherwise supersedes self.extrapolate
+        t3, c3, k = self.make_3d_case()
+        cls_extrap, call_extrap = extrap
+        bspl3 = NdBSpline(t3, c3, k=3, extrapolate=cls_extrap)
+
+        # evaluate out of bounds
+        x, y, z = [-2, -1, 7], [-3, -0.5, 6.5], [-1, -1.5, 7.5]
+        x, y, z = map(np.asarray, (x, y, z))
+        xi = [_ for _ in zip(x, y, z)]
+        target = x**3 * (y**3 + 2*y) * (z**3 + 3*z + 1)
+
+        result = bspl3(xi, extrapolate=call_extrap)
+        xp_assert_close(result, target, atol=1e-14)
+
+    def test_extrapolate_false_3D_separable(self):
+        # test that extrapolate=False produces nans for out-of-bounds values
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3)
+
+        # evaluate out of bounds and inside
+        x, y, z = [-2, 1, 7], [-3, 0.5, 6.5], [-1, 1.5, 7.5]
+        x, y, z = map(np.asarray, (x, y, z))
+        xi = [_ for _ in zip(x, y, z)]
+        target = x**3 * (y**3 + 2*y) * (z**3 + 3*z + 1)
+
+        result = bspl3(xi, extrapolate=False)
+        assert np.isnan(result[0])
+        assert np.isnan(result[-1])
+        xp_assert_close(result[1:-1], target[1:-1], atol=1e-14)
+
+    def test_x_nan_3D(self):
+        # test that spline(nan) is nan
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3)
+
+        # evaluate out of bounds and inside
+        x = np.asarray([-2, 3, np.nan, 1, 2, 7, np.nan])
+        y = np.asarray([-3, 3.5, 1, np.nan, 3, 6.5, 6.5])
+        z = np.asarray([-1, 3.5, 2, 3, np.nan, 7.5, 7.5])
+        xi = [_ for _ in zip(x, y, z)]
+        target = x**3 * (y**3 + 2*y) * (z**3 + 3*z + 1)
+        mask = np.isnan(x) | np.isnan(y) | np.isnan(z)
+        target[mask] = np.nan
+
+        result = bspl3(xi)
+        assert np.isnan(result[mask]).all()
+        xp_assert_close(result, target, atol=1e-14)
+
+    def test_non_c_contiguous(self):
+        # check that non C-contiguous inputs are OK
+        rng = np.random.default_rng(12345)
+        kx, ky = 3, 3
+        tx = np.sort(rng.uniform(low=0, high=4, size=16))
+        tx = np.r_[(tx[0],)*kx, tx, (tx[-1],)*kx]
+        ty = np.sort(rng.uniform(low=0, high=4, size=16))
+        ty = np.r_[(ty[0],)*ky, ty, (ty[-1],)*ky]
+
+        assert not tx[::2].flags.c_contiguous
+        assert not ty[::2].flags.c_contiguous
+
+        c = rng.uniform(size=(tx.size//2 - kx - 1, ty.size//2 - ky - 1))
+        c = c.T
+        assert not c.flags.c_contiguous
+
+        xi = np.c_[[1, 1.5, 2],
+                   [1.1, 1.6, 2.1]]
+
+        bspl2 = NdBSpline((tx[::2], ty[::2]), c, k=(kx, ky))
+        bspl2_0 = NdBSpline0((tx[::2], ty[::2]), c, k=(kx, ky))
+
+        xp_assert_close(bspl2(xi),
+                        [bspl2_0(xp) for xp in xi], atol=1e-14)
+
+    def test_readonly(self):
+        t3, c3, k = self.make_3d_case()
+        bspl3 = NdBSpline(t3, c3, k=3)
+
+        for i in range(3):
+            t3[i].flags.writeable = False
+        c3.flags.writeable = False
+
+        bspl3_ = NdBSpline(t3, c3, k=3)
+
+        assert bspl3((1, 2, 3)) == bspl3_((1, 2, 3))
+
+    def test_design_matrix(self):
+        t3, c3, k = self.make_3d_case()
+
+        xi = np.asarray([[1, 2, 3], [4, 5, 6]])
+        dm = NdBSpline(t3, c3, k).design_matrix(xi, t3, k)
+        dm1 = NdBSpline.design_matrix(xi, t3, [k, k, k])
+        assert dm.shape[0] == xi.shape[0]
+        xp_assert_close(dm.todense(), dm1.todense(), atol=1e-16)
+
+        with assert_raises(ValueError):
+            NdBSpline.design_matrix([1, 2, 3], t3, [k]*3)
+
+        with assert_raises(ValueError, match="Data and knots*"):
+            NdBSpline.design_matrix([[1, 2]], t3, [k]*3)
+
+    @pytest.mark.thread_unsafe
+    def test_concurrency(self):
+        rng = np.random.default_rng(12345)
+        k = 3
+        tx = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7)) * 3, 3, 3, 3, 3]
+        ty = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        tz = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=8)) * 4, 4, 4, 4, 4]
+        c = rng.uniform(size=(tx.size-k-1, ty.size-k-1, tz.size-k-1))
+
+        spl = NdBSpline((tx, ty, tz), c, k=k)
+
+        def worker_fn(_, spl):
+            xi = np.c_[[1, 1.5, 2],
+                       [1.1, 1.6, 2.1],
+                       [0.9, 1.4, 1.9]]
+            spl(xi)
+
+        _run_concurrent_barrier(10, worker_fn, spl)
+
+
+class TestMakeND:
+    def test_2D_separable_simple(self):
+        x = np.arange(6)
+        y = np.arange(6) + 0.5
+        values = x[:, None]**3 * (y**3 + 2*y)[None, :]
+        xi = [(a, b) for a, b in itertools.product(x, y)]
+
+        bspl = make_ndbspl((x, y), values, k=1)
+        xp_assert_close(bspl(xi), values.ravel(), atol=1e-15)
+
+        # test the coefficients vs outer product of 1D coefficients
+        spl_x = make_interp_spline(x, x**3, k=1)
+        spl_y = make_interp_spline(y, y**3 + 2*y, k=1)
+        cc = spl_x.c[:, None] * spl_y.c[None, :]
+        xp_assert_close(cc, bspl.c, atol=1e-11, rtol=0)
+
+        # test against RGI
+        from scipy.interpolate import RegularGridInterpolator as RGI
+        rgi = RGI((x, y), values, method='linear')
+        xp_assert_close(rgi(xi), bspl(xi), atol=1e-14)
+
+    def test_2D_separable_trailing_dims(self):
+        # test `c` with trailing dimensions, i.e. c.ndim > ndim
+        x = np.arange(6)
+        y = np.arange(6)
+        xi = [(a, b) for a, b in itertools.product(x, y)]
+
+        # make values4.shape = (6, 6, 4)
+        values = x[:, None]**3 * (y**3 + 2*y)[None, :]
+        values4 = np.dstack((values, values, values, values))
+        bspl = make_ndbspl((x, y), values4, k=3, solver=ssl.spsolve)
+
+        result = bspl(xi)
+        target = np.dstack((values, values, values, values)).astype(float)
+        assert result.shape == (36, 4)
+        xp_assert_close(result.reshape(6, 6, 4),
+                        target, atol=1e-14)
+
+        # now two trailing dimensions
+        values22 = values4.reshape((6, 6, 2, 2))
+        bspl = make_ndbspl((x, y), values22, k=3, solver=ssl.spsolve)
+
+        result = bspl(xi)
+        assert result.shape == (36, 2, 2)
+        xp_assert_close(result.reshape(6, 6, 2, 2),
+                        target.reshape((6, 6, 2, 2)), atol=1e-14)
+
+    @pytest.mark.parametrize('k', [(3, 3), (1, 1), (3, 1), (1, 3), (3, 5)])
+    def test_2D_mixed(self, k):
+        # make a 2D separable spline w/ len(tx) != len(ty)
+        x = np.arange(6)
+        y = np.arange(7) + 1.5
+        xi = [(a, b) for a, b in itertools.product(x, y)]
+
+        values = (x**3)[:, None] * (y**2 + 2*y)[None, :]
+        bspl = make_ndbspl((x, y), values, k=k, solver=ssl.spsolve)
+        xp_assert_close(bspl(xi), values.ravel(), atol=1e-15)
+
+    def test_2D_nans(self):
+        x = np.arange(6)
+        y = np.arange(6) + 0.5
+        y[-1] = np.nan
+        values = x[:, None]**3 * (y**3 + 2*y)[None, :]
+
+        with assert_raises(ValueError):
+            make_ndbspl((x, y), values, k=1)
+
+    def _get_sample_2d_data(self):
+        # from test_rgi.py::TestIntepN
+        x = np.array([.5, 2., 3., 4., 5.5, 6.])
+        y = np.array([.5, 2., 3., 4., 5.5, 6.])
+        z = np.array(
+            [
+                [1, 2, 1, 2, 1, 1],
+                [1, 2, 1, 2, 1, 1],
+                [1, 2, 3, 2, 1, 1],
+                [1, 2, 2, 2, 1, 1],
+                [1, 2, 1, 2, 1, 1],
+                [1, 2, 2, 2, 1, 1],
+            ]
+        )
+        return x, y, z
+
+    def test_2D_vs_RGI_linear(self):
+        x, y, z = self._get_sample_2d_data()
+        bspl = make_ndbspl((x, y), z, k=1)
+        rgi = RegularGridInterpolator((x, y), z, method='linear')
+
+        xi = np.array([[1, 2.3, 5.3, 0.5, 3.3, 1.2, 3],
+                       [1, 3.3, 1.2, 4.0, 5.0, 1.0, 3]]).T
+
+        xp_assert_close(bspl(xi), rgi(xi), atol=1e-14)
+
+    def test_2D_vs_RGI_cubic(self):
+        x, y, z = self._get_sample_2d_data()
+        bspl = make_ndbspl((x, y), z, k=3, solver=ssl.spsolve)
+        rgi = RegularGridInterpolator((x, y), z, method='cubic_legacy')
+
+        xi = np.array([[1, 2.3, 5.3, 0.5, 3.3, 1.2, 3],
+                       [1, 3.3, 1.2, 4.0, 5.0, 1.0, 3]]).T
+
+        xp_assert_close(bspl(xi), rgi(xi), atol=1e-14)
+
+    @pytest.mark.parametrize('solver', [ssl.gmres, ssl.gcrotmk])
+    def test_2D_vs_RGI_cubic_iterative(self, solver):
+        # same as `test_2D_vs_RGI_cubic`, only with an iterative solver.
+        # Note the need to add an explicit `rtol` solver_arg to achieve the
+        # target accuracy of 1e-14. (the relation between solver atol/rtol
+        # and the accuracy of the final result is not direct and needs experimenting)
+        x, y, z = self._get_sample_2d_data()
+        bspl = make_ndbspl((x, y), z, k=3, solver=solver, rtol=1e-6)
+        rgi = RegularGridInterpolator((x, y), z, method='cubic_legacy')
+
+        xi = np.array([[1, 2.3, 5.3, 0.5, 3.3, 1.2, 3],
+                       [1, 3.3, 1.2, 4.0, 5.0, 1.0, 3]]).T
+
+        xp_assert_close(bspl(xi), rgi(xi), atol=1e-14, rtol=1e-7)
+
+    def test_2D_vs_RGI_quintic(self):
+        x, y, z = self._get_sample_2d_data()
+        bspl = make_ndbspl((x, y), z, k=5, solver=ssl.spsolve)
+        rgi = RegularGridInterpolator((x, y), z, method='quintic_legacy')
+
+        xi = np.array([[1, 2.3, 5.3, 0.5, 3.3, 1.2, 3],
+                       [1, 3.3, 1.2, 4.0, 5.0, 1.0, 3]]).T
+
+        xp_assert_close(bspl(xi), rgi(xi), atol=1e-14)
+
+    @pytest.mark.parametrize(
+        'k, meth', [(1, 'linear'), (3, 'cubic_legacy'), (5, 'quintic_legacy')]
+    )
+    def test_3D_random_vs_RGI(self, k, meth):
+        rndm = np.random.default_rng(123456)
+        x = np.cumsum(rndm.uniform(size=6))
+        y = np.cumsum(rndm.uniform(size=7))
+        z = np.cumsum(rndm.uniform(size=8))
+        values = rndm.uniform(size=(6, 7, 8))
+
+        bspl = make_ndbspl((x, y, z), values, k=k, solver=ssl.spsolve)
+        rgi = RegularGridInterpolator((x, y, z), values, method=meth)
+
+        xi = np.random.uniform(low=0.7, high=2.1, size=(11, 3))
+        xp_assert_close(bspl(xi), rgi(xi), atol=1e-14)
+
+    def test_solver_err_not_converged(self):
+        x, y, z = self._get_sample_2d_data()
+        solver_args = {'maxiter': 1}
+        with assert_raises(ValueError, match='solver'):
+            make_ndbspl((x, y), z, k=3, **solver_args)
+
+        with assert_raises(ValueError, match='solver'):
+            make_ndbspl((x, y), np.dstack((z, z)), k=3, **solver_args)
+
+
+class TestFpchec:
+    # https://github.com/scipy/scipy/blob/main/scipy/interpolate/fitpack/fpchec.f
+
+    def test_1D_x_t(self):
+        k = 1
+        t = np.arange(12).reshape(2, 6)
+        x = np.arange(12)
+
+        with pytest.raises(ValueError, match="1D sequence"):
+            _b.fpcheck(x, t, k)
+
+        with pytest.raises(ValueError, match="1D sequence"):
+            _b.fpcheck(t, x, k)
+
+    def test_condition_1(self):
+        # c      1) k+1 <= n-k-1 <= m
+        k = 3
+        n  = 2*(k + 1) - 1    # not OK
+        m = n + 11            # OK
+        t = np.arange(n)
+        x = np.arange(m)
+
+        assert dfitpack.fpchec(x, t, k) == 10
+        with pytest.raises(ValueError, match="Need k+1*"):
+            _b.fpcheck(x, t, k)
+
+        n = 2*(k+1) + 1   # OK
+        m = n - k - 2     # not OK
+        t = np.arange(n)
+        x = np.arange(m)
+
+        assert dfitpack.fpchec(x, t, k) == 10
+        with pytest.raises(ValueError, match="Need k+1*"):
+            _b.fpcheck(x, t, k)
+
+    def test_condition_2(self):
+        # c      2) t(1) <= t(2) <= ... <= t(k+1)
+        # c         t(n-k) <= t(n-k+1) <= ... <= t(n)
+        k = 3
+        t = [0]*(k+1) + [2] + [5]*(k+1)   # this is OK
+        x = [1, 2, 3, 4, 4.5]
+
+        assert dfitpack.fpchec(x, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None    # does not raise
+
+        tt = t.copy()
+        tt[-1] = tt[0]   # not OK
+        assert dfitpack.fpchec(x, tt, k) == 20
+        with pytest.raises(ValueError, match="Last k knots*"):
+            _b.fpcheck(x, tt, k)
+
+        tt = t.copy()
+        tt[0] = tt[-1]   # not OK
+        assert dfitpack.fpchec(x, tt, k) == 20
+        with pytest.raises(ValueError, match="First k knots*"):
+            _b.fpcheck(x, tt, k)
+
+    def test_condition_3(self):
+        # c      3) t(k+1) < t(k+2) < ... < t(n-k)
+        k = 3
+        t = [0]*(k+1) + [2, 3] + [5]*(k+1)   # this is OK
+        x = [1, 2, 3, 3.5, 4, 4.5]
+        assert dfitpack.fpchec(x, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None
+
+        t = [0]*(k+1) + [2, 2] + [5]*(k+1)   # this is not OK
+        assert dfitpack.fpchec(x, t, k) == 30
+        with pytest.raises(ValueError, match="Internal knots*"):
+            _b.fpcheck(x, t, k)
+
+    def test_condition_4(self):
+        # c      4) t(k+1) <= x(i) <= t(n-k)
+        # NB: FITPACK's fpchec only checks x[0] & x[-1], so we follow.
+        k = 3
+        t = [0]*(k+1) + [5]*(k+1)
+        x = [1, 2, 3, 3.5, 4, 4.5]      # this is OK
+        assert dfitpack.fpchec(x, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None
+
+        xx = x.copy()
+        xx[0] = t[0]    # still OK
+        assert dfitpack.fpchec(xx, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None
+
+        xx = x.copy()
+        xx[0] = t[0] - 1    # not OK
+        assert dfitpack.fpchec(xx, t, k) == 40
+        with pytest.raises(ValueError, match="Out of bounds*"):
+            _b.fpcheck(xx, t, k)
+
+        xx = x.copy()
+        xx[-1] = t[-1] + 1    # not OK
+        assert dfitpack.fpchec(xx, t, k) == 40
+        with pytest.raises(ValueError, match="Out of bounds*"):
+            _b.fpcheck(xx, t, k)
+
+    # ### Test the S-W condition (no 5)
+    # c      5) the conditions specified by schoenberg and whitney must hold
+    # c         for at least one subset of data points, i.e. there must be a
+    # c         subset of data points y(j) such that
+    # c             t(j) < y(j) < t(j+k+1), j=1,2,...,n-k-1
+    def test_condition_5_x1xm(self):
+        # x(1).ge.t(k2) .or. x(m).le.t(nk1)
+        k = 1
+        t = [0, 0, 1, 2, 2]
+        x = [1.1, 1.1, 1.1]
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+        x = [0.5, 0.5, 0.5]
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+    def test_condition_5_k1(self):
+        # special case nk3 (== n - k - 2) < 2
+        k = 1
+        t = [0, 0, 1, 1]
+        x = [0.5, 0.6]
+        assert dfitpack.fpchec(x, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None
+
+    def test_condition_5_1(self):
+        # basically, there can't be an interval of t[j]..t[j+k+1] with no x
+        k = 3
+        t = [0]*(k+1) + [2] + [5]*(k+1)
+        x = [3]*5
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+        t = [0]*(k+1) + [2] + [5]*(k+1)
+        x = [1]*5
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+    def test_condition_5_2(self):
+        # same as _5_1, only the empty interval is in the middle
+        k = 3
+        t = [0]*(k+1) + [2, 3] + [5]*(k+1)
+        x = [1.1]*5 + [4]
+
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+        # and this one is OK
+        x = [1.1]*4 + [4, 4]
+        assert dfitpack.fpchec(x, t, k) == 0
+        assert _b.fpcheck(x, t, k) is None
+
+    def test_condition_5_3(self):
+        # similar to _5_2, covers a different failure branch
+        k = 1
+        t = [0, 0, 2, 3, 4, 5, 6, 7, 7]
+        x = [1, 1, 1, 5.2, 5.2, 5.2, 6.5]
+
+        assert dfitpack.fpchec(x, t, k) == 50
+        with pytest.raises(ValueError, match="Schoenberg-Whitney*"):
+            _b.fpcheck(x, t, k)
+
+
+# ### python replicas of generate_knots(...) implementation details, for testing.
+# ### see TestGenerateKnots::test_split_and_add_knot
+def _split(x, t, k, residuals):
+    """Split the knot interval into "runs".
+    """
+    ix = np.searchsorted(x, t[k:-k])
+    # sum half-open intervals
+    fparts = [residuals[ix[i]:ix[i+1]].sum() for i in range(len(ix)-1)]
+    carries = residuals[ix[1:-1]]
+
+    for i in range(len(carries)):     # split residuals at internal knots
+        carry = carries[i] / 2
+        fparts[i] += carry
+        fparts[i+1] -= carry
+
+    fparts[-1] += residuals[-1]       # add the contribution of the last knot
+
+    xp_assert_close(sum(fparts), sum(residuals), atol=1e-15)
+
+    return fparts, ix
+
+
+def _add_knot(x, t, k, residuals):
+    """Insert a new knot given reduals."""
+    fparts, ix = _split(x, t, k, residuals)
+
+    # find the interval with max fparts and non-zero number of x values inside
+    idx_max = -101
+    fpart_max = -1e100
+    for i in range(len(fparts)):
+        if ix[i+1] - ix[i] > 1 and fparts[i] > fpart_max:
+            idx_max = i
+            fpart_max = fparts[i]
+
+    if idx_max == -101:
+        raise ValueError("Internal error, please report it to SciPy developers.")
+
+    # round up, like Dierckx does? This is really arbitrary though.
+    idx_newknot = (ix[idx_max] + ix[idx_max+1] + 1) // 2
+    new_knot = x[idx_newknot]
+    idx_t = np.searchsorted(t, new_knot)
+    t_new = np.r_[t[:idx_t], new_knot, t[idx_t:]]
+    return t_new
+
+
+class TestGenerateKnots:
+    def test_split_add_knot(self):
+        # smoke test implementation details: insert a new knot given residuals
+        x = np.arange(8, dtype=float)
+        y = x**3 + 1./(1 + x)
+        k = 3
+        t = np.array([0.]*(k+1) + [7.]*(k+1))
+        spl = make_lsq_spline(x, y, k=k, t=t)
+        residuals = (spl(x) - y)**2
+
+        from scipy.interpolate import _fitpack_repro as _fr
+        new_t = _fr.add_knot(x, t, k, residuals)
+        new_t_py = _add_knot(x, t, k, residuals)
+
+        xp_assert_close(new_t, new_t_py, atol=1e-15)
+
+        # redo with new knots
+        spl2 = make_lsq_spline(x, y, k=k, t=new_t)
+        residuals2 = (spl2(x) - y)**2
+
+        new_t2 = _fr.add_knot(x, new_t, k, residuals2)
+        new_t2_py = _add_knot(x, new_t, k, residuals2)
+
+        xp_assert_close(new_t2, new_t2_py, atol=1e-15)
+
+    @pytest.mark.parametrize('k', [1, 2, 3, 4, 5])
+    def test_s0(self, k):
+        x = np.arange(8, dtype=np.float64)
+        y = np.sin(x*np.pi/8)
+        t = list(generate_knots(x, y, k=k, s=0))[-1]
+
+        tt = splrep(x, y, k=k, s=0)[0]
+        xp_assert_close(t, tt, atol=1e-15)
+
+    def test_s0_1(self):
+        # with these data, naive algorithm tries to insert >= nmax knots
+        n = 10
+        x = np.arange(n)
+        y = x**3
+        knots = list(generate_knots(x, y, k=3, s=0))   # does not error out
+        xp_assert_close(knots[-1], _not_a_knot(x, 3), atol=1e-15)
+
+    def test_s0_n20(self):
+        n = 20
+        x = np.arange(n)
+        y = x**3
+        knots = list(generate_knots(x, y, k=3, s=0))
+        xp_assert_close(knots[-1], _not_a_knot(x, 3), atol=1e-15)
+
+    def test_s0_nest(self):
+        # s=0 and non-default nest: not implemented, errors out
+        x = np.arange(10)
+        y = x**3
+        with assert_raises(ValueError):
+            list(generate_knots(x, y, k=3, s=0, nest=10))
+
+    def test_s_switch(self):
+        # test the process switching to interpolating knots when len(t) == m + k + 1
+        """
+        To generate the `wanted` list below apply the following diff and rerun
+        the test. The stdout will contain successive iterations of the `t`
+        array.
+
+$ git diff scipy/interpolate/fitpack/fpcurf.f
+diff --git a/scipy/interpolate/fitpack/fpcurf.f b/scipy/interpolate/fitpack/fpcurf.f
+index 1afb1900f1..d817e51ad8 100644
+--- a/scipy/interpolate/fitpack/fpcurf.f
++++ b/scipy/interpolate/fitpack/fpcurf.f
+@@ -216,6 +216,9 @@ c  t(j+k) <= x(i) <= t(j+k+1) and store it in fpint(j),j=1,2,...nrint.
+         do 190 l=1,nplus
+ c  add a new knot.
+           call fpknot(x,m,t,n,fpint,nrdata,nrint,nest,1)
++          print*, l, nest, ': ', t
++          print*, "n, nmax = ", n, nmax
++
+ c  if n=nmax we locate the knots as for interpolation.
+           if(n.eq.nmax) go to 10
+ c  test whether we cannot further increase the number of knots.
+        """  # NOQA: E501
+        x = np.arange(8)
+        y = np.sin(x*np.pi/8)
+        k = 3
+
+        knots = list(generate_knots(x, y, k=k, s=1e-7))
+        wanted = [[0., 0., 0., 0., 7., 7., 7., 7.],
+                  [0., 0., 0., 0., 4., 7., 7., 7., 7.],
+                  [0., 0., 0., 0., 2., 4., 7., 7., 7., 7.],
+                  [0., 0., 0., 0., 2., 4., 6., 7., 7., 7., 7.],
+                  [0., 0., 0., 0., 2., 3., 4., 5., 7, 7., 7., 7.]
+        ]
+
+        assert len(knots) == len(wanted)
+        for t, tt in zip(knots, wanted):
+            xp_assert_close(t, tt, atol=1e-15)
+
+        # also check that the last knot vector matches FITPACK
+        t, _, _ = splrep(x, y, k=k, s=1e-7)
+        xp_assert_close(knots[-1], t, atol=1e-15)
+
+    def test_list_input(self):
+        # test that list inputs are accepted
+        x = list(range(8))
+        gen = generate_knots(x, x, s=0.1, k=1)
+        next(gen)
+
+    def test_nest(self):
+        # test that nest < nmax stops the process early (and we get 10 knots not 12)
+        x = np.arange(8)
+        y = np.sin(x*np.pi/8)
+        s = 1e-7
+
+        knots = list(generate_knots(x, y, k=3, s=s, nest=10))
+        xp_assert_close(knots[-1],
+                        [0., 0., 0., 0., 2., 4., 7., 7., 7., 7.], atol=1e-15)
+
+        with assert_raises(ValueError):
+            # nest < 2*(k+1)
+            list(generate_knots(x, y, k=3, nest=4))
+
+    def test_weights(self):
+        x = np.arange(8)
+        y = np.sin(x*np.pi/8)
+
+        with assert_raises(ValueError):
+            list(generate_knots(x, y, w=np.arange(11)))   # len(w) != len(x)
+
+        with assert_raises(ValueError):
+            list(generate_knots(x, y, w=-np.ones(8)))    # w < 0
+
+    @pytest.mark.parametrize("npts", [30, 50, 100])
+    @pytest.mark.parametrize("s", [0.1, 1e-2, 0])
+    def test_vs_splrep(self, s, npts):
+        # XXX this test is brittle: differences start apearing for k=3 and s=1e-6,
+        # also for k != 3. Might be worth investigating at some point.
+        # I think we do not really guarantee exact agreement with splrep. Instead,
+        # we guarantee it is the same *in most cases*; otherwise slight differences
+        # are allowed. There is no theorem, it is al heuristics by P. Dierckx.
+        # The best we can do it to best-effort reproduce it.
+        rndm = np.random.RandomState(12345)
+        x = 10*np.sort(rndm.uniform(size=npts))
+        y = np.sin(x*np.pi/10) + np.exp(-(x-6)**2)
+
+        k = 3
+        t = splrep(x, y, k=k, s=s)[0]
+        tt = list(generate_knots(x, y, k=k, s=s))[-1]
+
+        xp_assert_close(tt, t, atol=1e-15)
+
+    @pytest.mark.thread_unsafe
+    def test_s_too_small(self):
+        n = 14
+        x = np.arange(n)
+        y = x**3
+
+        # XXX splrep warns that "s too small": ier=2
+        knots = list(generate_knots(x, y, k=3, s=1e-50))
+
+        with suppress_warnings() as sup:
+            r = sup.record(RuntimeWarning)
+            tck = splrep(x, y, k=3, s=1e-50)
+            assert len(r) == 1
+        xp_assert_equal(knots[-1], tck[0])
+
+
+def disc_naive(t, k):
+    """Straitforward way to compute the discontinuity matrix. For testing ONLY.
+
+    This routine returns a dense matrix, while `_fitpack_repro.disc` returns
+    a packed one.
+    """
+    n = t.shape[0]
+
+    delta = t[n - k - 1] - t[k]
+    nrint = n - 2*k - 1
+
+    ti = t[k+1:n-k-1]   # internal knots
+    tii = np.repeat(ti, 2)
+    tii[::2] += 1e-10
+    tii[1::2] -= 1e-10
+    m = BSpline(t, np.eye(n - k - 1), k)(tii, nu=k)
+
+    matr = np.empty((nrint-1, m.shape[1]), dtype=float)
+    for i in range(0, m.shape[0], 2):
+        matr[i//2, :] = m[i, :] - m[i+1, :]
+
+    matr *= (delta/nrint)**k / math.factorial(k)
+    return matr
+
+
+class F_dense:
+    """ The r.h.s. of ``f(p) = s``, an analog of _fitpack_repro.F
+    Uses full matrices, so is for tests only.
+    """
+    def __init__(self, x, y, t, k, s, w=None):
+        self.x = x
+        self.y = y
+        self.t = t
+        self.k = k
+        self.w = np.ones_like(x, dtype=float) if w is None else w
+        assert self.w.ndim == 1
+
+        # lhs
+        a_dense = BSpline(t, np.eye(t.shape[0] - k - 1), k)(x)
+        self.a_dense = a_dense * self.w[:, None]
+
+        from scipy.interpolate import _fitpack_repro as _fr
+        self.b_dense = PackedMatrix(*_fr.disc(t, k)).todense()
+
+        # rhs
+        assert y.ndim == 1
+        yy = y * self.w
+        self.yy = np.r_[yy, np.zeros(self.b_dense.shape[0])]
+
+        self.s = s
+
+    def __call__(self, p):
+        ab = np.vstack((self.a_dense, self.b_dense / p))
+
+        # LSQ solution of ab @ c = yy
+        from scipy.linalg import qr, solve
+        q, r = qr(ab, mode='economic')
+
+        qy = q.T @ self.yy
+
+        nc = r.shape[1]
+        c = solve(r[:nc, :nc], qy[:nc])
+
+        spl = BSpline(self.t, c, self.k)
+        fp = np.sum(self.w**2 * (spl(self.x) - self.y)**2)
+
+        self.spl = spl   # store it
+
+        return fp - self.s
+
+
+class TestMakeSplrep:
+    def test_input_errors(self):
+        x = np.linspace(0, 10, 11)
+        y = np.linspace(0, 10, 12)
+        with assert_raises(ValueError):
+            # len(x) != len(y)
+            make_splrep(x, y)
+
+        with assert_raises(ValueError):
+            # 0D inputs
+            make_splrep(1, 2, s=0.1)
+
+        with assert_raises(ValueError):
+            # y.ndim > 2
+            y = np.ones((x.size, 2, 2, 2))
+            make_splrep(x, y, s=0.1)
+
+        w = np.ones(12)
+        with assert_raises(ValueError):
+            # len(weights) != len(x)
+            make_splrep(x, x**3, w=w, s=0.1)
+
+        w = -np.ones(12)
+        with assert_raises(ValueError):
+            # w < 0
+            make_splrep(x, x**3, w=w, s=0.1)
+
+        w = np.ones((x.shape[0], 2))
+        with assert_raises(ValueError):
+            # w.ndim != 1
+            make_splrep(x, x**3, w=w, s=0.1)
+
+        with assert_raises(ValueError):
+            # x not ordered
+            make_splrep(x[::-1], x**3, s=0.1)
+
+        with assert_raises(TypeError):
+            # k != int(k)
+            make_splrep(x, x**3, k=2.5, s=0.1)
+
+        with assert_raises(ValueError):
+            # s < 0
+            make_splrep(x, x**3, s=-1)
+
+        with assert_raises(ValueError):
+            # nest < 2*k + 2
+            make_splrep(x, x**3, k=3, nest=2, s=0.1)
+
+        with assert_raises(ValueError):
+            # nest not None and s==0
+            make_splrep(x, x**3, s=0, nest=11)
+
+        with assert_raises(ValueError):
+            # len(x) != len(y)
+            make_splrep(np.arange(8), np.arange(9), s=0.1)
+
+    def _get_xykt(self):
+        x = np.linspace(0, 5, 11)
+        y  = np.sin(x*3.14 / 5)**2
+        k = 3
+        s = 1.7e-4
+        tt = np.array([0]*(k+1) + [2.5, 4.0] + [5]*(k+1))
+
+        return x, y, k, s, tt
+
+    def test_fitpack_F(self):
+        # test an implementation detail: banded/packed linalg vs full matrices
+        from scipy.interpolate._fitpack_repro import F
+
+        x, y, k, s, t = self._get_xykt()
+        f = F(x, y[:, None], t, k, s)    # F expects y to be 2D
+        f_d = F_dense(x, y, t, k, s)
+        for p in [1, 10, 100]:
+            xp_assert_close(f(p), f_d(p), atol=1e-15)
+
+    def test_fitpack_F_with_weights(self):
+        # repeat test_fitpack_F, with weights
+        from scipy.interpolate._fitpack_repro import F
+
+        x, y, k, s, t = self._get_xykt()
+        w = np.arange(x.shape[0], dtype=float)
+        fw = F(x, y[:, None], t, k, s, w=w)       # F expects y to be 2D
+        fw_d = F_dense(x, y, t, k, s, w=w)
+
+        f_d = F_dense(x, y, t, k, s)   # no weights
+
+        for p in [1, 10, 100]:
+            xp_assert_close(fw(p), fw_d(p), atol=1e-15)
+            assert not np.allclose(f_d(p), fw_d(p), atol=1e-15)
+
+    def test_disc_matrix(self):
+        # test an implementation detail: discontinuity matrix
+        # (jumps of k-th derivative at knots)
+        import scipy.interpolate._fitpack_repro as _fr
+
+        rng = np.random.default_rng(12345)
+        t = np.r_[0, 0, 0, 0, np.sort(rng.uniform(size=7))*5, 5, 5, 5, 5]
+
+        n, k = len(t), 3
+        D = PackedMatrix(*_fr.disc(t, k)).todense()
+        D_dense = disc_naive(t, k)
+        assert D.shape[0] == n - 2*k - 2   # number of internal knots
+        xp_assert_close(D, D_dense, atol=1e-15)
+
+    def test_simple_vs_splrep(self):
+        x, y, k, s, tt = self._get_xykt()
+        tt = np.array([0]*(k+1) + [2.5, 4.0] + [5]*(k+1))
+
+        t,c,k = splrep(x, y, k=k, s=s)
+        assert all(t == tt)
+
+        spl = make_splrep(x, y, k=k, s=s)
+        xp_assert_close(c[:spl.c.size], spl.c, atol=1e-15)
+
+    def test_with_knots(self):
+        x, y, k, s, _ = self._get_xykt()
+
+        t = list(generate_knots(x, y, k=k, s=s))[-1]
+
+        spl_auto = make_splrep(x, y, k=k, s=s)
+        spl_t = make_splrep(x, y, t=t, k=k, s=s)
+
+        xp_assert_close(spl_auto.t, spl_t.t, atol=1e-15)
+        xp_assert_close(spl_auto.c, spl_t.c, atol=1e-15)
+        assert spl_auto.k == spl_t.k
+
+    def test_no_internal_knots(self):
+        # should not fail if there are no internal knots
+        n = 10
+        x = np.arange(n)
+        y = x**3
+        k = 3
+        spl = make_splrep(x, y, k=k, s=1)
+        assert spl.t.shape[0] == 2*(k+1)
+
+    def test_default_s(self):
+        n = 10
+        x = np.arange(n)
+        y = x**3
+        spl = make_splrep(x, y, k=3)
+        spl_i = make_interp_spline(x, y, k=3)
+
+        xp_assert_close(spl.c, spl_i.c, atol=1e-15)
+
+    @pytest.mark.thread_unsafe
+    def test_s_too_small(self):
+        # both splrep and make_splrep warn that "s too small": ier=2
+        n = 14
+        x = np.arange(n)
+        y = x**3
+
+        with suppress_warnings() as sup:
+            r = sup.record(RuntimeWarning)
+            tck = splrep(x, y, k=3, s=1e-50)
+            spl = make_splrep(x, y, k=3, s=1e-50)
+            assert len(r) == 2
+            xp_assert_equal(spl.t, tck[0])
+            xp_assert_close(np.r_[spl.c, [0]*(spl.k+1)],
+                            tck[1], atol=5e-13)
+
+    def test_issue_22704(self):
+        # Reference - https://github.com/scipy/scipy/issues/22704
+        x = np.asarray([20.00, 153.81, 175.57, 202.47, 237.11,
+             253.61, 258.56, 273.40, 284.54, 293.61,
+             298.56, 301.86, 305.57, 307.22, 308.45,
+             310.10, 310.10, 310.50], dtype=np.float64)
+        y = np.asarray([53.00, 49.50, 48.60, 46.80, 43.20,
+             40.32, 39.60, 36.00, 32.40, 28.80,
+             25.20, 21.60, 18.00, 14.40, 10.80,
+             7.20, 3.60, 0.0], dtype=np.float64)
+        w = np.asarray([1.38723] * y.shape[0], dtype=np.float64)
+        with assert_raises(ValueError):
+            make_splrep(x, y, w=w, k=2, s=12)
+
+    def test_shape(self):
+        # make sure coefficients have the right shape (not extra dims)
+        n, k = 10, 3
+        x = np.arange(n)
+        y = x**3
+
+        spl = make_splrep(x, y, k=k)
+        spl_1 = make_splrep(x, y, k=k, s=1e-5)
+
+        assert spl.c.ndim == 1
+        assert spl_1.c.ndim == 1
+
+        # force the general code path, not shortcuts
+        spl_2 = make_splrep(x, y + 1/(1+y), k=k, s=1e-5)
+        assert spl_2.c.ndim == 1
+
+    def test_s0_vs_not(self):
+        # check that the shapes are consistent
+        n, k = 10, 3
+        x = np.arange(n)
+        y = x**3
+
+        spl_0 = make_splrep(x, y, k=3, s=0)
+        spl_1 = make_splrep(x, y, k=3, s=1)
+
+        assert spl_0.c.ndim == 1
+        assert spl_1.c.ndim == 1
+
+        assert spl_0.t.shape[0] == n + k + 1
+        assert spl_1.t.shape[0] == 2 * (k + 1)
+
+
+class TestMakeSplprep:
+    def _get_xyk(self, m=10, k=3):
+        x = np.arange(m) * np.pi / m
+        y = [np.sin(x), np.cos(x)]
+        return x, y, k
+
+    @pytest.mark.parametrize('s', [0, 0.1, 1e-3, 1e-5])
+    def test_simple_vs_splprep(self, s):
+        # Check/document the interface vs splPrep
+        # The four values of `s` are to probe all code paths and shortcuts
+        m, k = 10, 3
+        x = np.arange(m) * np.pi / m
+        y = [np.sin(x), np.cos(x)]
+
+        # the number of knots depends on `s` (this is by construction)
+        num_knots = {0: 14, 0.1: 8, 1e-3: 8 + 1, 1e-5: 8 + 2}
+
+        # construct the splines
+        (t, c, k), u_ = splprep(y, s=s)
+        spl, u = make_splprep(y, s=s)
+
+        # parameters
+        xp_assert_close(u, u_, atol=1e-15)
+
+        # knots
+        xp_assert_close(spl.t, t, atol=1e-15)
+        assert len(t) == num_knots[s]
+
+        # coefficients: note the transpose
+        cc = np.asarray(c).T
+        xp_assert_close(spl.c, cc, atol=1e-15)
+
+        # values: note axis=1
+        xp_assert_close(spl(u),
+                        BSpline(t, c, k, axis=1)(u), atol=1e-15)
+
+    @pytest.mark.parametrize('s', [0, 0.1, 1e-3, 1e-5])
+    def test_array_not_list(self, s):
+        # the argument of splPrep is either a list of arrays or a 2D array (sigh)
+        _, y, _ = self._get_xyk()
+        assert isinstance(y, list)
+        assert np.shape(y)[0] == 2
+
+        # assert the behavior of FITPACK's splrep
+        tck, u = splprep(y, s=s)
+        tck_a, u_a = splprep(np.asarray(y), s=s)
+        xp_assert_close(u, u_a, atol=s)
+        xp_assert_close(tck[0], tck_a[0], atol=1e-15)
+        assert len(tck[1]) == len(tck_a[1])
+        for c1, c2 in zip(tck[1], tck_a[1]):
+            xp_assert_close(c1, c2, atol=1e-15)
+        assert tck[2] == tck_a[2]
+        assert np.shape(splev(u, tck)) == np.shape(y)
+
+        spl, u = make_splprep(y, s=s)
+        xp_assert_close(u, u_a, atol=1e-15)
+        xp_assert_close(spl.t, tck_a[0], atol=1e-15)
+        xp_assert_close(spl.c.T, tck_a[1], atol=1e-15)
+        assert spl.k == tck_a[2]
+        assert spl(u).shape == np.shape(y)
+
+        spl, u = make_splprep(np.asarray(y), s=s)
+        xp_assert_close(u, u_a, atol=1e-15)
+        xp_assert_close(spl.t, tck_a[0], atol=1e-15)
+        xp_assert_close(spl.c.T, tck_a[1], atol=1e-15)
+        assert spl.k == tck_a[2]
+        assert spl(u).shape == np.shape(y)
+
+        with assert_raises(ValueError):
+            make_splprep(np.asarray(y).T, s=s)
+
+    def test_default_s_is_zero(self):
+        x, y, k = self._get_xyk(m=10)
+
+        spl, u = make_splprep(y)
+        xp_assert_close(spl(u), y, atol=1e-15)
+
+    def test_s_zero_vs_near_zero(self):
+        # s=0 and s \approx 0 are consistent
+        x, y, k = self._get_xyk(m=10)
+
+        spl_i, u_i = make_splprep(y, s=0)
+        spl_n, u_n = make_splprep(y, s=1e-15)
+
+        xp_assert_close(u_i, u_n, atol=1e-15)
+        xp_assert_close(spl_i(u_i), y, atol=1e-15)
+        xp_assert_close(spl_n(u_n), y, atol=1e-7)
+        assert spl_i.axis == spl_n.axis
+        assert spl_i.c.shape == spl_n.c.shape
+
+    def test_1D(self):
+        x = np.arange(8, dtype=float)
+        with assert_raises(ValueError):
+            splprep(x)
+
+        with assert_raises(ValueError):
+            make_splprep(x, s=0)
+
+        with assert_raises(ValueError):
+            make_splprep(x, s=0.1)
+
+        tck, u_ = splprep([x], s=1e-5)
+        spl, u = make_splprep([x], s=1e-5)
+
+        assert spl(u).shape == (1, 8)
+        xp_assert_close(spl(u), [x], atol=1e-15)
+
+
+class BatchSpline:
+    # BSpline-line class with reference batch behavior
+    def __init__(self, x, y, axis, *, spline, **kwargs):
+        y = np.moveaxis(y, axis, -1)
+        self._batch_shape = y.shape[:-1]
+        self._splines = [spline(x, yi, **kwargs) for yi in y.reshape(-1, y.shape[-1])]
+        self._axis = axis
+
+    def __call__(self, x):
+        y = [spline(x) for spline in self._splines]
+        y = np.reshape(y, self._batch_shape + x.shape)
+        return np.moveaxis(y, -1, self._axis) if x.shape else y
+
+    def integrate(self, a, b, extrapolate=None):
+        y = [spline.integrate(a, b, extrapolate) for spline in self._splines]
+        return np.reshape(y, self._batch_shape)
+
+    def derivative(self, nu):
+        res = copy.deepcopy(self)
+        res._splines = [spline.derivative(nu) for spline in res._splines]
+        return res
+
+    def antiderivative(self, nu):
+        res = copy.deepcopy(self)
+        res._splines = [spline.antiderivative(nu) for spline in res._splines]
+        return res
+
+
+class TestBatch:
+    @pytest.mark.parametrize('make_spline, kwargs',
+        [(make_interp_spline, {}),
+         (make_smoothing_spline, {}),
+         (make_smoothing_spline, {'lam': 1.0}),
+         (make_lsq_spline, {'method': "norm-eq"}),
+         (make_lsq_spline, {'method': "qr"}),
+         ])
+    @pytest.mark.parametrize('eval_shape', [(), (1,), (3,)])
+    @pytest.mark.parametrize('axis', [-1, 0, 1])
+    def test_batch(self, make_spline, kwargs, axis, eval_shape):
+        rng = np.random.default_rng(4329872134985134)
+        n = 10
+        shape = (2, 3, 4, n)
+        domain = (0, 10)
+
+        x = np.linspace(*domain, n)
+        y = np.moveaxis(rng.random(shape), -1, axis)
+
+        if make_spline == make_lsq_spline:
+            k = 3  # spline degree, if needed
+            t = (x[0],) * (k + 1) + (x[-1],) * (k + 1)  # valid knots, if needed
+            kwargs = kwargs | dict(t=t, k=k)
+
+        res = make_spline(x, y, axis=axis, **kwargs)
+        ref = BatchSpline(x, y, axis=axis, spline=make_spline, **kwargs)
+
+        x = rng.uniform(*domain, size=eval_shape)
+        np.testing.assert_allclose(res(x), ref(x))
+
+        res, ref = res.antiderivative(1), ref.antiderivative(1)
+        np.testing.assert_allclose(res(x), ref(x))
+
+        res, ref = res.derivative(2), ref.derivative(2)
+        np.testing.assert_allclose(res(x), ref(x))
+
+        np.testing.assert_allclose(res.integrate(*domain), ref.integrate(*domain))