Description:

• The command implementconstant implements the constant expression expr in arbitrary precision. More precisely, it generates the source code (written in C, and using MPFR) of a C function const_something with the following signature: void const_something (mpfr_ptr y, mp_prec_t prec) Let us denote by c the exact mathematical value of the constant defined by the expression expr. When called with arguments y and prec (where the variable y is supposed to be already initialized), the function mpfr_const_something sets the precision of y to a suitable precision and stores in it an approximate value of c such that |y-c| <= |c|*2^(1-prec).
• When no filename filename is given or if default is given as filename, the source code produced by implementconstant is printed on standard output. Otherwise, when filename is given as a string of characters, the source code is output to a file named filename. If that file cannot be opened and/or written to, implementconstant fails and has no other effect.
• When functionname is given as an argument to implementconstant and functionname evaluates to a string of characters, the default name for the C function const_something is replaced by functionname. When default is given as functionname, the default name is used nevertheless, as if no functionname argument were given. When choosing a character sequence for functionname, the user should keep attention to the fact that functionname must be a valid C identifier in order to enable error-free compilation of the produced code.
• If expr refers to a constant defined with libraryconstant, the produced code uses the external code implementing this constant. The user should keep in mind that it is up to them to make sure the symbol for that external code can get resolved when the newly generated code is to be loaded.
• If a subexpression of expr evaluates to 0, implementconstant will most likely fail with an error message.
• implementconstant is unable to implement constant expressions expr that contain procedure-based functions, i.e. functions created from Sollya procedures using the function construct. If expr contains such a procedure-based function, implementconstant prints a warning and fails silently. The reason for this lack of functionality is that the produced C source code, which is supposed to be compiled, would have to call back to the Sollya interpreter in order to evaluate the procedure-based function.
• Similarly, implementconstant is currently unable to implement constant expressions expr that contain library-based functions, i.e. functions dynamically bound to Sollya using the library construct. If expr contains such a library-based function, implementconstant prints a warning and fails silently. Support for this feature is in principle feasible from a technical standpoint and might be added in a future release of Sollya.
• Currently, non-differentiable functions such as double, doubledouble, tripledouble, single, halfprecision, quad, doubleextended, floor, ceil, nearestint are not supported by implementconstant. If implementconstant encounters one of them, a warning message is displayed and no code is produced. However, if autosimplify equals on, it is possible that Sollya silently simplifies subexpressions of expr containing such functions and that implementconstant successfully produces code for evaluating expr.
• While it produces an MPFR-based C source code for expr, implementconstant takes architectural and system-dependent parameters into account. For example, it checks whether literal constants figuring in expr can be represented on a C long int type or if they must be stored in a different manner not to affect their accuracy. These tests, performed by Sollya during execution of implementconstant, depend themselves on the architecture Sollya is running on. Users should keep this matter in mind, especially when trying to compile source code on one machine whilst it has been produced on another.