isinf() - comp.lang.c

The GNU Autoconf manual provides a work-around for missing isinf() and isnan() functions, which exist in C99 and many, but not all, pre-C99 implementations: http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html The given definition of isinf(x) boils down to return isnan (x - x), which will return true for NaNs. Surely this is wrong? -- Richard -- Please remember to mention me / in tapes you leave behind.

Richard Tobin wrote: > The GNU Autoconf manual provides a work-around for missing isinf() > and isnan() functions, which exist in C99 and many, but not all, > pre-C99 implementations: > > http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html > > The given definition of isinf(x) boils down to return isnan (x - x), > which will return true for NaNs. Surely this is wrong? It seems so to me.

"Malcolm McLean" <regniztar@btinternet.com> writes: > "Richard Tobin" <richard@cogsci.ed.ac.uk> wrote in message > > int isinf(double x) > { > if(x > DBL_MAX) > return 1; > else > return 0; > } That fails for negative infinity. -- Keith Thompson (The_Other_Keith) kst-u@mib.org <http://www.ghoti.net/~kst> Nokia "We must do something. This is something. Therefore, we must do this." -- Antony Jay and Jonathan Lynn, "Yes Minister"

Richard Tobin wrote: > > The GNU Autoconf manual provides a work-around for missing isinf() > and isnan() functions, which exist in C99 and many, but not all, > pre-C99 implementations: > > http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html > > The given definition of isinf(x) boils down to return isnan (x - x), > which will return true for NaNs. Surely this is wrong? Think about it. If x is not a number, what can you say about x-x? I fail to see any way that expression could become a number. -- [mail]: Chuck F (cbfalconer at maineline dot net) [page]: <http://cbfalconer.home.att.net> Try the download section.

On 8 Oct 2008 14:11:29 GMT, Richard Tobin <richard@cogsci.ed.ac.uk> wrote: > The GNU Autoconf manual provides a work-around for missing isinf() > and isnan() functions, which exist in C99 and many, but not all, > pre-C99 implementations: > > http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html > > The given definition of isinf(x) boils down to return isnan (x - x), > which will return true for NaNs. Surely this is wrong? If isnan(x) is true, then x - x is also NaN, and isnan(x - x) would also be true. Check Annex F to the C99 standard. Specifically, in F8.2 it is made clear that the equivalence x - x == 0.0 does not hold when x can be NaN or Inf. Martien -- | Martien Verbruggen | I took an IQ test and the results were | negative. |

CBFalconer <cbfalconer@yahoo.com> writes: > Richard Tobin wrote: >> >> The GNU Autoconf manual provides a work-around for missing isinf() >> and isnan() functions, which exist in C99 and many, but not all, >> pre-C99 implementations: >> >> http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html >> >> The given definition of isinf(x) boils down to return isnan (x - x), >> which will return true for NaNs. Surely this is wrong? > > Think about it. If x is not a number, what can you say about x-x? > I fail to see any way that expression could become a number. Indeed, if x is NaN then so is x-x, so isnan(x-x) would be true. But 7.12.3.3 says "The isinf macro returns a nonzero value if and only if its argument has an infinite value." That makes it sound as if isinf(x) should return zero if x is NaN, since NaN does not have an infinite value.

In article <o9ujcg.aa1.ln@news.heliotrope.home>, Martien Verbruggen <mgjv@tradingpost.com.au> wrote: >> The given definition of isinf(x) boils down to return isnan (x - x), >> which will return true for NaNs. Surely this is wrong? >If isnan(x) is true, then x - x is also NaN, and isnan(x - x) would also >be true. Exactly. And that would be wrong, because isinf(x) should be false if x is a NaN. Anyway, I eventually discovered that this mistake has already been reported to the autoconf project. -- Richard -- Please remember to mention me / in tapes you leave behind.

Martien Verbruggen wrote: > On 8 Oct 2008 14:11:29 GMT, > Richard Tobin <richard@cogsci.ed.ac.uk> wrote: >> The GNU Autoconf manual provides a work-around for missing isinf() >> and isnan() functions, which exist in C99 and many, but not all, >> pre-C99 implementations: >> >> http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html >> >> The given definition of isinf(x) boils down to return isnan (x - x), >> which will return true for NaNs. Surely this is wrong? > > If isnan(x) is true, then x - x is also NaN, and isnan(x - x) would also > be true. > > Check Annex F to the C99 standard. Specifically, in F8.2 it is made > clear that the equivalence x - x == 0.0 does not hold when x can be NaN > or Inf. Exactly. Which is why isnan(x-x) doesn't make sense as an implementation of ifinf(x). isinf(x) isn't supposed to return true when x is a NaN; it's only supposed to return true when x is infinite.

On Thu, 09 Oct 2008 12:57:20 GMT, James Kuyper <jameskuyper@verizon.net> wrote: > Martien Verbruggen wrote: >> On 8 Oct 2008 14:11:29 GMT, >> Richard Tobin <richard@cogsci.ed.ac.uk> wrote: >>> The GNU Autoconf manual provides a work-around for missing isinf() >>> and isnan() functions, which exist in C99 and many, but not all, >>> pre-C99 implementations: >>> >>> http://www.gnu.org/software/libtool/manual/autoconf/Function-Portability.html >>> >>> The given definition of isinf(x) boils down to return isnan (x - x), >>> which will return true for NaNs. Surely this is wrong? >> >> If isnan(x) is true, then x - x is also NaN, and isnan(x - x) would also >> be true. >> >> Check Annex F to the C99 standard. Specifically, in F8.2 it is made >> clear that the equivalence x - x == 0.0 does not hold when x can be NaN >> or Inf. > > Exactly. Which is why isnan(x-x) doesn't make sense as an implementation > of ifinf(x). isinf(x) isn't supposed to return true when x is a NaN; > it's only supposed to return true when x is infinite. Sorry all. I misread the isinf() as isnan(). I thought the OP was confused about why isnan(x-x) returned true. Martien -- | Yes; Windows is great for running & Martien Verbruggen | developing viruses, for instance. It's also | very popular, but then again, so is the | common cold. -- Dave Hinz