todo

Linux下文件工具。

Tasks for GNU fileutils:

From Paul Eggert:
>> ... GNU `ls' violates the POSIX standard in its handling of
>> options.  For example, POSIX requires `ls -x -C' and `ls -C -x' to
>> have the same effect, and this works correctly with Solaris 8 `ls',
>> but it doesn't work with GNU `ls'.
======================

Pull the separate paragraphs in sort and ls describing effect of
locale (LC_ALL etc.) into a new section and make ls and sort xref that.

======================
 dd skip=N vs linux tape devices
 df vs ext3 listed as auto (and hence ignored)
 rm's HAVE_WORKING_READDIR: run test in more directories: /tmp /var/tmp
======================

  shred.c (isaac_seed_machdep): rewrite so it doesn't need signal handlers

  [From Paul Eggert]
  One of these days we really should change all those 'unsigned's in
  hash.[ch] to 'size_t', as the current code has an arbitrary 4GB limit
  on many 64-bit hosts.

  poison functions like fseeko that have been entirely removed from these
    packages, so they can't be accidentally reintroduced later.

  Apply a few minor patches to ls.

  Add a -g option to df so it dumps the entire statvfs structure.
  Solaris' df already does that.  From Karl Berry.

  ------------------
  Something that I would really appreciate is if someone would run the
  Open Group's VSC-lite test suite against the fileutils and textutils
  and report the failures.

    http://www.opengroup.org/testing/downloads/vsclite.html

  I've been meaning to do it myself for months, but haven't found the time.
  There's a bit of set-up required, some of which requires root access, e.g.,
  to create a few test user accounts and some test groups.
  ------------------

  remove support for `touch DATE FILE' once POSIX 1003.1-200x becomes official,
    (see Paul's 2001-03-17 doc change)

  ls: fix ls and then tweak the last two tests here: tests/ls-2/tests
    (I already have patches for both of them)

  m4/inttypes_h.m4 (jm_AC_HEADER_INTTYPES_H): Don't define HAVE_INTTYPES_H.
    That conflicts with the symbol you'd get from AC_CHECK_HEADERS(inttypes.h)

  Currently `ls -Fd ./' prints this: .//
  It should probably print this instead: ./    (reported by Fran鏾is Pinard)

  Consider making `chmod MODE DIR' and `mkdir -m MODE DIR' work the
  same way those programs work on on non-GNU systems when the containing
  directory has the set-GID bit set.  Currently, GNU chmod and mkdir set
  the mode of DIR to MODE.  At least Solaris and AIX versions of those
  programs set the mode to MODE,g+s.  The only exception seems to be that
  `chmod g-s DIR' removes the setgid attribute.  POSIX allows both
  implementations.  See tests/chmod/setgid for a little more info.
  In any case, document the behavior.

  integrate patches for Linux ACL support

  ============================
  From Stallman:

  THE DIRECTORY TREE.
  -------------------
  In DOS the command `tree' displays the directory tree. This is very
  helpful in grasping the morphology of the directory tree. I would be
  most desirable to have this in G-L console mode too. In the first
  PC-s the directory tree was so small that it could be displayed on
  one screen. Now it is necessary to pipe a full tree into a file to
  see it by scrolling it up and down. A G-L tree must be piped to be
  displayed and scroll-able.
  ============================

  ls: test suite to exercize -l and -l --full-time with various LANG settings

  ls: recognize --format="%a %b..." strftime formats
  ls: add new % formats to get the 10 bytes of the permissions string
    (both individually and as groups)

  Don't strip slashes unless configure
      detects that e.g. lstat can't deal with them.

  From: [e-mail] (Colin Plumb)
  | A very nice feature I just wanted, and I assumed a GNU tool would have,
  | is the mv analogue to cp -s, that is, move the file, but leave a symlink
  | to the new home.
  |
  | E.g. mv -s very_big_file /dir/in/fs/with/lots/of/space
  |
  | (Okay, I should make the change and submit a patch.  I just wish I
  | didn't have to deal with that bizarre brace style when doing it.)

These are mainly notes to myself.
If something piques your interest and you want to contribute,
let me know and I can send you details.

  chown, cp, etc: don't use chown on symlinks

  new option: failure of `cp -p' to restore perm bits causes failure;
     POSIX says default should be no.  Stallman +u.cp

  cp, mv, install: factor out duplicated hash_init cruft -- put it in copy.c?

  cp.c: remove clone of make_path:  see FIXME

  copy.c (is_ancestor): See FIXME comment.

  cp -f (and install, etc.) fail when the destination is a dangling symlink

  make ls's size field wider, so files of size > 9,999,999 don't cause
    misalignment

  From Noel L Yap:
    Add a new option (or extend --preserve) to allow preserving some
    subset of a file's attributed.  E.g.,
       cp --preserve=timestamp,owner,group,mode
    where any of "timestamp", "owner", "group", or "mode" may or may not
    be specified.  The default behaviour of "cp --preserve" would be the
    way it is now (ie preserve everything that can be).