[renggl0j]

Hi,

I'm using your interpolators in my code. Unfortunately when x is out of bounds they return 0, which is not easy to discriminate if a valid interpolations may also return 0.

I'm therefore proposing this change so the caller may chose a different invalid value to return (typically std::numeric_limits<float>::quiet_NaN()), whilst guaranteeing backward-compatilibity.

[CD3]

Thank you for the pull request.

There was another request to allow extrapolation for out of bounds values. I've pushed a branch that allows configuring the out-of-bounds value when extrapolation is disabled and does a "reasonable" extrapolation when enabled. The interface looks like this

TEST_CASE("Extrapolation settings") {
    _1D::CubicSplineInterpolator<double> interp;

    int N = 100;
    double xmin = 0, xmax = 2 * M_PI;
    double dx = (xmax - xmin) / (N - 1);

    _1D::CubicSplineInterpolator<double>::VectorType xx(N), yy(N);

    CHECK_THROWS_AS(interp(1), std::logic_error);

    for (int i = 0; i < N; i++) {
        xx(i) = dx * i;
        yy(i) = sin(xx(i));
    }
    interp.setData(xx, yy);

    CHECK(interp(-1) == 0);
    CHECK(interp(8) == 0);
    CHECK(!std::isnan(interp(-1)));
    CHECK(!std::isnan(interp(-4)));

    interp.setOutOfBoundsValue(1);

    CHECK(interp(-1) == 1);
    CHECK(interp(8) == 1);
    CHECK(!std::isnan(interp(-1)));
    CHECK(!std::isnan(interp(-4)));

    interp.setOutOfBoundsNaN();

    CHECK(interp(-1) != std::numeric_limits<double>::quiet_NaN());
    CHECK(interp(8) != std::numeric_limits<double>::quiet_NaN());
    CHECK(std::isnan(interp(-1)));
    CHECK(std::isnan(interp(-4)));

    interp.enableExtrapolation();

    CHECK(interp(-1) == Approx(-1));
    CHECK(interp(8) == Approx(8 - 2 * M_PI));
}

Does this do what you are looking for?

[CD3]

#35 would be resolved with feat-extrapolation-configuration branch too.

[renggl0j]

Yes, that would be an even better interface than my quick proposal, thanks.