Re: lib/49240: Some corner cases for pow() fail to work correctly

[Havard Eidnes <he%NetBSD.org@localhost>]

The following reply was made to PR lib/49240; it has been noted by GNATS.

From: Havard Eidnes <he%NetBSD.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: lib/49240: Some corner cases for pow() fail to work correctly
Date: Sat, 04 Oct 2014 16:24:18 +0200 (CEST)

 Hmm.
 
 It appears that we build a "multi-standard fdlibm as libm", and that
 we by default implement _POSIX_MODE (ref. src/lib/lim/Makefile).
 
 So ... the test program, if it should test the IEEE compliance for the
 corner cases, needs to tell libm to do the IEEE operations instead of
 the POSIX operations via
 
         _LIB_VERSION =3D _IEEE_;
 
 (is that the officially sanctioned method, BTW?)
 
 and the only failures flagged by the test program I posted earlier
 which our original libm then fail with is
 
 Test failure: 1.0 =3D=3D pow(-1, Infinity)
    a =3D=3D 1.000000
    b =3D=3D nan
 Test failure: 1.0 =3D=3D pow(-1, -Infinity)
    a =3D=3D 1.000000
    b =3D=3D nan
 
 which correspond to this change:
 
  * Add a case for -1**+-Inf, since C99 and IEEE 754 revised 2008(?)
    reportedly state that this should be 1
 
 Further, it looks like the reason for the "yy1" variable name is that
 a use of "y1" appears to mask a global "y1", at least on netbsd-6.
 
 Test cases to justify the other changes should be found.
 
 Regards,
 
 - H=E5vard