Turbo C 2.01

/*------------------------------------------------------------------------ * filename - matherr.c * * function(s) * matherr - user-modifiable math error handler *-----------------------------------------------------------------------*/ /*[]---------------------------------------------------[]*/ /*| |*/ /*| Turbo C Run Time Library - Version 2.0 |*/ /*| |*/ /*| |*/ /*| Copyright (c) 1987,1988 by Borland International|*/ /*| All Rights Reserved. |*/ /*| |*/ /*[]---------------------------------------------------[]*/ #include <math.h> #ifdef UNIX_matherr #include <stdio.h> #include <process.h> char *whyS [] = { "argument domain error", "argument singularity ", "overflow range error ", "underflow range error", "total loss of significance", "partial loss of significance" }; /*------------------------------------------------------------------------* Name matherr - user-modifiable math error handler Usage #include <math.h> int matherr(struct exception *e); Prototype in math.h Description When exceptions are detected in the math library then a call is made to _matherr() with all the available information. That function does very little, except to map the exception "why" into either ERANGE or EDOMAIN in errno. Its main purpose is to act as a focal point for changes in error handling. For example, if you were writing a spreadsheet you might replace this function with one which pops up an error window explaining something like: "log (-2.0) caused domain error, in cell J7" and then longjmp() to a reset state in the spreadsheet and await the next command from the user. The default version of Turbo C's matherr routine masks underflow and precision errors; others errors are considered fatal. It serves as a hook that you can replace when writing your own math error handling routine. The rationale for masking underflow and precision errors is that these are not errors according to the ANSI C spec. Consequently, you will get exp(-1000) = 0 sin(1e100) = NAN without any error or warning, even though there is a total loss of precision in both cases. You can trap these errors by modifying matherr. The possible errors are DOMAIN, SING, OVERFLOW, UNDERFLOW, TLOSS, PLOSS and listed in <math.h>. As explained above, UNDERFLOW and TLOSS are masked by the default matherr. PLOSS is not supported by TC and is not generated by any library functions. The remaining errors, DOMAIN, SING, and OVERFLOW, are fatal with the default matherr. You can modify matherr to be a custom error handling routine (such as one that catches and resolves certain type of errors); the modified matherr should return 0 if it failed to resolve the error, or non-zero if the error was resolved. When matherr returns non-zero, no error message is printed, and errno is not changed. The important thing is that we don't know what error handling you want, but you are assured that all errors will arrive at matherr() with all the information you need to design a custom format. We do not ship as standard the function named matherr() which may be familiar to UNIX users, since the ANSI x3j11 draft specifies an incompatible style. This version is as close as we could get without breaking the ANSI rules. You can, however, convert this version to the UNIX style if you prefer. The necessary code is included but switched off. Return value The default return value for matherr is simply 0. matherr can also modify e->retval, which propagates through _matherr back to the original caller. When matherr returns 0, (indicating that it was not able to resolve the error) _matherr sets errno and prints an error message. When matherr returns non-zero, (indicating that it was able to resolve the error) errno is not set and no messages are printed. *-------------------------------------------------------------------------*/ int matherr (struct exception *e) { fprintf (stderr, "%s (%8g,%8g): %s\n", e->name, e->arg1, e->arg2, whyS [e->type - 1]); exit (1); } #else int matherr(struct exception *e) { if (e->type == UNDERFLOW) { /* flush underflow to 0 */ e->retval = 0; return 1; } if (e->type == TLOSS) { /* total loss of precision, but ignore the problem */ return 1; } /* all other errors are fatal */ return 0; } #endif