Bruno Haible <[EMAIL PROTECTED]> wrote: > On i386, x86_64, ia64, the 'long double' format has an explicitly represented > most significant mantissa bit. This has the consequence that some bit patterns > in a 'long double' are not covered by IEEE 754. *printf better prints these > numbers as NaN (rather than printing some of them as finite numbers, or > crashing on some of them, like glibc does [1]). > > This changes the gnulib *printf-posix modules to recognize this case and > provide a workaround. > > Jim is interested in using the *printf-posix modules but wants no big code > imports on glibc-2.6 systems. Therefore I have tied the recognition of this > workaround to a modules named 'printf-safe'. If you don't want it, simply > pass the option --avoid=print-safe to gnulib-tool. > > Bruno > > [1] http://sourceware.org/bugzilla/show_bug.cgi?id=4586 > The report is for ia64, but the same test program crashes also on i386.
Thanks for doing that. But when you put it that way, perhaps some glibc-post-2.6 release would be a more reasonable target.
