net@tub.UUCP (Oliver Laumann) (05/04/90)
What is the correct method to check whether an exception (overflow, invalid arguments, etc.) occurred during the execution of a C math library function (such as sqrt() or sin())? Do I have to test the result against NaN, or do I have to check errno (EDOM)? If the latter is true, is errno cleared automatically on success or do I have to clear it before calling the function? Thanks, -- Oliver Laumann net@TUB.BITNET net@tub.cs.tu-berlin.de net@tub.UUCP
henry@utzoo.uucp (Henry Spencer) (05/08/90)
In article <1323@tub.UUCP> net@tub.UUCP (Oliver Laumann) writes: >What is the correct method to check whether an exception (overflow, >invalid arguments, etc.) occurred during the execution of a C math >library function (such as sqrt() or sin())? > >Do I have to test the result against NaN, or do I have to check errno >(EDOM)? If the latter is true, is errno cleared automatically on >success or do I have to clear it before calling the function? There is no simple, uniform, standard way of testing for trouble in the math functions. You have to look at the return value of each function in combination with errno. No, the functions do not clear errno on success, and in certain circumstances they are not guaranteed to set it on failure. No, they do not return NaN, since that is not a fully portable concept. The whole area is kind of a mess, especially since even the unsatisfactory existing state of things does not satisfy the high-performance fanatics, who would prefer (last I heard) to get back something like a NaN rather than having to test values and flags every time. There are people looking at better approaches. -- If OSI is the answer, what is | Henry Spencer at U of Toronto Zoology the question?? -Rolf Nordhagen| uunet!attcan!utzoo!henry henry@zoo.toronto.edu
decot@hpisod2.HP.COM (Dave Decot) (05/08/90)
> What is the correct method to check whether an exception (overflow, > invalid arguments, etc.) occurred during the execution of a C math > library function (such as sqrt() or sin())? > > Do I have to test the result against NaN, or do I have to check errno > (EDOM)? If the latter is true, is errno cleared automatically on > success or do I have to clear it before calling the function? Given the current confused state of affairs... For System V conforming systems, errno is set appropriately, and NaN (or infinities, or zero, where appropriate) may or may not be returned. For IEEE 754/854 conforming systems, NaN, appropriate infinities, and zero are returned as required, but errno may or may not be set. matherr() may or may not be called if provided by the application ...I would recommend following the suggestions given in X/Open Portability Guide, Issue 3 (XPG3); namely: An application wishing to check for error situations should set errno to zero before calling the function. If errno is set on return, or the return value is NaN (or for some functions, zero, or HUGE_VAL, where appropriate), an error has occurred. Under error conditions on some implementations, an error message is printed on stderr. On such implementations, this error handling can be suppressed by providing a matherr() function, such as the following, as part of the application code: #include <math.h> #include <errno.h> #ifdef DOMAIN int matherr(except) struct exception *except; { switch (except->type) { case DOMAIN: case SING: errno = EDOM; return 1; defauilt: errno = ERANGE; return 1; } } #endif /* DOMAIN */ Dave Decot HP