Re: [PATCH 2/3] Consistently use BFD's time

[Tom Tromey <tromey@adacore.com>]

Pedro> I disagree.  It's not only GDB's stat usage that counts, it's the
Pedro> other gnulib modules that depend on the stat module fixes too.  Also,
Pedro> that approach disables the stat module for all systems, not just
Pedro> mingw.  I think it's just bad policy to disable a module like that.

The reason I came to this conclusion is that the list of fixes doesn't
seem to include anything super important; while using it introduces bugs
that are difficult to work around, or even audit for.  In particular,
we'll have to be careful to never compare mtime-from-gnulib against
mtime-from-BFD.

Here's the list from the version of gnulib we're using.

   Portability problems fixed by Gnulib:
   * On platforms where 'off_t' is a 32-bit type, 'stat' may not
     correctly report the size of files or block devices larger than 2
     GB. (Cf.  'AC_SYS_LARGEFILE'.)
   * On Linux/x86 and Linux/x86_64, applications compiled in 32-bit mode
     cannot access files that happen to have a 64-bit inode number.
     This can occur with file systems such as XFS (typically on large
     disks) and NFS. (Cf.  'AC_SYS_LARGEFILE'.)
   * The 'st_atime', 'st_ctime', 'st_mtime' fields are affected by the
     current time zone and by the DST flag of the current time zone on
     some platforms: mingw, MSVC 14 (when the environment variable 'TZ'
     is set).
   * On MSVC 14, this function fails with error 'ENOENT' on files such
     as 'C:\pagefile.sys' and on directories such as 'C:\System Volume
     Information'.
   * On some platforms, 'stat("link-to-file/",buf)' succeeds instead of
     failing with 'ENOTDIR'.  FreeBSD 7.2, AIX 7.1, Solaris 9, mingw64.
   * On some platforms, 'stat(".",buf)' and 'stat("./",buf)' give
     different results: mingw, MSVC 14.
   * On Solaris 11.4, when this function yields a timestamp with a
     nonpositive 'tv_sec' value, 'tv_nsec' might be in the range
     -1000000000..-1, representing a negative nanoseconds offset from
     'tv_sec'.

I think the first one is probably not important to gdb; the second is
already working; and the third is a bug.  The rest just don't seem very
relevant.

Tom