Hi Ben,
Ben Pfaff wrote:
> PSPP would like to use the isnan macro portably. Gnulib, through
> its isnanf-nolibm, isnand-nolibm, and isnanl modules, almost has
> what it needs. I've written up a change set that adds isnanf and
> isnand modules, which correspond to isnanl, and then builds an
> isnan module on top of them.
Correct, these are exactly the pieces that are missing for complete isnan()
support.
> + * lib/frexp.c: Use new isnand-nolibm.h instead of isnand.h.
> + * lib/isfinite.c: Likewise, and isnanf-nolibm.h instead of
> + isnanf.h.
> + * lib/signbitd.c: Likewise.
> + * lib/signbitf.c: Likewise.
> + * lib/vasnprintf.c: Likewise.
I think this is all right, but it would convince me more fully if you could
send two separate patches:
1) one patch that renames isnand.h to isnand-nolibm.h and likewise for
isnanf.h,
2) one patch that introduces the new modules.
The tricky point is that the *printf-posix replacements must not depend on
libm, because the system's printf() does not require libm and hence programs
don't expect to need to link with -lm just for printf. During testing, this
means that test-vasnprintf-posix must not produce link errors.
> + * m4/isnanl.h (gl_BUILD_ISNANL): New macro used by gl_FUNC_ISNANL,
> + gl_FUNC_ISNANL_NO_LIBM, and gl_FUNC_ISNAN.
Typo: that should be m4/isnanl.m4.
Good work, thank you!
Bruno
