Version=0.2.2 (07/2026)

## Bug Fixes
- adapt demos to new notations
- add a new class "bspline_basis" including parameter order_deriv
- derivative of constant polynomials and bsplines,
- bspline_basis, bspline_base_der and bs-direct until order 0

## Changes
- add an option to 'makepp' function to render PP polynomial callable
- add the printable method for the new class "callable_pp"
and "non_callable_pp"
- modify 'makespline' function with a flag the chose callable
or non-callable output


Version=0.2.1 (07/2026)
=============
## Bug Fixes
- Deprecated notation in CVXR
- Small bug preventing the regression when the number of knots is minimal
(only one piece)
- eval_pp : add extroplating values out of the knots range
- Fixed issues with `der3cons` vector lengths for different degrees
- Fixed third derivative constraints for cubic splines
- Complete constraints if not enough are provided if degree=3
- Update `print.callable_spline` method


## Changes
- Updated CVXR syntax, removed deprecated functions and cleaned up code
- Improved solver selection logic with fallback mechanism
- Stabilizing bspline_eval and eval_pp functions for unique interval case
- Allow bspline_eval and eval_pp functions to extrapolate values outside the knots range
- Quantile_spline() :   callable=TRUE is now default value
- New classes as output of quantile_spline: callable_spline and non_callable_spline
- Improved `make_spline` and 'spline_eval' function to handle both list and callable objects



Version=0.2.0 (07/2026)
=============
0) minor updates
- Removed "beta" status
- Added system requirements (Rust/Cargo for Linux) to README
- Updated CRAN badges
1) Concerning knots
- Replaced the variable "knots" by the variable "knot" everywhere to avoid confusion with knots() R build-in function
- changed notation paradigm: variable "knot" refers to knots including ends (replace int_knots in version 1.0.1)
variable "ext_knot" refers to extended knot partition (replace knots in version 1.0.1)
3) Spline Evaluation
-Added case degree=0 which caused problems since some variables (coeff) lose 1 dimension. R automaticaly reduces the dimension of the arrays.
-Rendered the spline container (list of knots, coefficients, degree) callable
-Allowed the call of a spline to pass the spline basis coefficient to accelerate the computations (as well as in the function eval_spline)
4) Add other degree for constrained regression
- quartic splines using Karlin-studen technique for cubic polynomial (monotone) and quadratic polynomial (convex).
- quadratic and linear splines. No major theoretical difficulty.
- Unify all regression functions in a quantile_spline function with selection of the degree as a parameter