Agena

agena >> `The Power of Procedural Programming` 6.7.4 Isfahan, February 08, 2026 - With univariate functions, `calc.differ` now calls `dual.diff` internally to return high-accuracy results if possible. Should the function to be differentiated include functions and operators not supporting dual numbers, `calc.differ` automatically reverts to the classic differentiating functions `calc.eulerdiff`, `calc.xpdiff`, `calc.diff`. - `calc.extrema`, `calc.minimum`, `calc.maximum`, `calc.inflect`, `calc.saddles` and `calc.curvature` now call `dual.diff` internally if possible to return high-accuracy results, too. - The `+`, `-`, `*`, `/`, `^` and `**` could freeze when given `undefined` or `infinity` as an argument. This has been fixed. - With negative dual numbers, the `^` operator no longer returns `undefined` but mostly a finite result. `x^0` and `x**0` now either correctly yield `undefined` or `1`. - The `^` and `**` operators could not correctly compute derivatives at the origin and with an integral power. This has been fixed, also benefitting `dual.diff`. - There has been a general problem with the Agena numbers `undefined` or `infinity`, or these non-finite values packed into dual numbers, passed to dual-number functions and operators, causing segmentation faults or freezes. This has been fixed. 6.7.3 Isfahan, February 06, 2026 - On Intel platforms, the precision of `calc.Psi` with the first and second derivative has been improved by three to four digits, now calculating the results with 80-bit precision. Thus, the function now fully utilizes 80-bit floating-point math. - The precision of the `gamma` and `fact` functions has been improved a tiny bit in the real domain with larger arguments on Intel platforms. - On Intel platforms, all trigonometric and hyperbolic functions internally call the same underlying 80-bit precision C functions, also to protect the interpreter against unpredictable and peculiar behaviour in OS/2. - `math.lnbeta` is now better detecting poles. - On Raspberry Pi, `long.csch`, `long.sech` and `long.coth` have become 50 percent faster by calling the respective Sun Microsystems implementations instead of the built-in GCC ones