Pádraig Brady <p...@draigbrady.com> writes: > Jim Meyering wrote: >> Pádraig Brady wrote: >> >>> Jim Meyering wrote: >>>> Eric Blake wrote: >>>>> According to Pádraig Brady on 10/5/2009 3:53 PM: >>>>>>>>> This is a new test, but FC5 is soooo old, >>>>>>>>> that I'm not sure it's worth worrying about. >>>>>>>> March 2006? >>>>>>> The failure is probably a function of the kernel. >>>>>>> Which is it? >>>>>> In summary this is what fails: >>>>>> >>>>>> $ touch a >>>>>> $ ln -s a symlink >>>>>> $ ln -L symlink hardlink >>>>>> ln: creating hard link `hardlink' => `symlink': Invalid argument >>>>>> >>>>>> `man linkat` says that AT_SYMLINK_FOLLOW is only supported since 2.6.18 >>>>>> and my FC5 system is 2.6.17 >>>>> This should fix it. I don't have access to FC5, but I tested the new code >>>>> path by priming the cache (gl_cv_func_linkat_follow=runtime ./configure) >>>>> along with a temporary setting of have_follow_really=-1 in linkat.c. I >>>>> also verified that the replacement is not picked up on cygwin 1.7, where >>>>> AT_SYMLINK_FOLLOW was implemented at the same time as linkat. >>>>> >>>>> The patch copies from areadlink.c, as well as link_follow earlier in >>>>> linkat.c, to create two new fd-relative helpers. For now, I didn't see >>>>> any reason to expose them, but areadlinkat may someday be worth making >>>>> into a full-blown module. >>>> Wow, that was quick. Thanks. >>>> I should have read this first. >>>> >>>> I was just reviewing the changes in gnulib and >>>> see a few that should be included in the imminent coreutils >>>> beta release, so will probably take this one, too. >>> Needs a couple of tweaks.. >>> >>> This needs to be added to linkat.c >>> (seems like it should be refactored somewhere?) >>> >>> #ifndef SIZE_MAX >>> # define SIZE_MAX ((size_t) -1) >>> #endif >>> #ifndef SSIZE_MAX >>> # define SSIZE_MAX ((ssize_t) (SIZE_MAX / 2)) >>> #endif >> >> This should do it: >> >>>From 6f6420cc9705dcfa545a28c674fddf5703e72c86 Mon Sep 17 00:00:00 2001 >> From: Jim Meyering <meyer...@redhat.com> >> Date: Tue, 6 Oct 2009 11:11:39 +0200 >> Subject: [PATCH] linkat: avoid compilation failure >> >> * lib/linkat.c: Include <stdint.h> for use of SIZE_MAX. > > That works thanks. > > I suppose these should include stdint.h also? > > areadlink.c:# define SIZE_MAX ((size_t) -1) > areadlink-with-size.c:# define SIZE_MAX ((size_t) -1) > backupfile.c:# define SIZE_MAX ((size_t) -1) > fnmatch.c:# define SIZE_MAX ((size_t) -1) > quotearg.c:# define SIZE_MAX ((size_t) -1) > striconv.c:# define SIZE_MAX ((size_t) -1)
Note that stdint.h may not be sufficient to get SIZE_MAX, quoting size_max.h: /* Get SIZE_MAX declaration on systems like Solaris 7/8/9. */ # include <limits.h> /* Get SIZE_MAX declaration on systems like glibc 2. */ # if HAVE_STDINT_H # include <stdint.h> # endif /* On systems where these include files don't define it, SIZE_MAX is defined in config.h. */ However given that SIZE_MAX should be in stdint.h according to POSIX, maybe it makes more sense to make sure gnulib's stdint.h replacement is enabled when SIZE_MAX is not provided by the system's stdint.h? And then deprecate size_max.h in favor of stdint. > While these already include stdint.h so should probably not redefine > > fts.c:# define SIZE_MAX ((size_t) -1) > getdelim.c:# define SSIZE_MAX ((ssize_t) (SIZE_MAX / 2)) > getndelim2.c:# define SSIZE_MAX ((ssize_t) (SIZE_MAX / 2)) SSIZE_MAX should be provided by limit.h, see: http://www.opengroup.org/onlinepubs/009695399/basedefs/limits.h.html The stdint.h documentation doesn't mention SSIZE_MAX: http://www.opengroup.org/onlinepubs/009695399/basedefs/stdint.h.html /Simon
