[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

dpkg 0.93.66: overwriting right so that later removal works

The intended behaviour of dpkg when overwriting files from another
package is to remove them from the old package's list, so that if
the old package is removed the overwritten file will not be.

As Bill Mitchell discovered, dpkg 0.93.64 gets this wrong.  I think
the bug was introduced in 0.93.59, though it may have been later.
(The old Perl dpkg doesn't do anything nearly so sophisticated.)

This is usually not critical, but if you're doing updates of packages
which are exchanging files or changing names you should upgrade.

Also, in this version --purge is an action in its own right, so you
don't have to say `dpkg --purge --remove' any more.  The -P alias for
--purge has been withdrawn.  `dpkg --auto --purge' does the same as
`dpkg --auto --remove': they both remove or purge files as selected by
dselect, regardless of which action is specified.

dpkg (0.93.66) ALPHA; priority=MEDIUM

  * dpkg will correctly remove overwritten files from the lists of
    the package(s) that used to contain them.

  * dpkg --purge is now an action, rather than a modifier for --remove,
    and the -P alias for it is withdrawn.

  * dpkg --unpack/--install filenames in messages are now more sensible
    about when to use .../ (show as many trailing components as possible
    in 40 characters, or the whole path if that the last component is
    longer than that).

 -- Ian Jackson <iwj10@cus.cam.ac.uk>  Thu, 3 Aug 1995 02:11:03 +0100

38a4e5549174de1ff0255d3effb37127  dpkg-0.93.66.deb
a2018efd99114beaba5a20ea9eaac5ce  dpkg-0.93.66.nondebbin.tar.gz
badec1d80d7b52bd596f71c85154f1a5  dpkg-0.93.66.tar.gz
-rw-r--r--   1 root     root       369950 Aug  3 02:27 dpkg-0.93.66.deb
-rw-rw-r--   1 root     ian        367541 Aug  3 02:24 dpkg-0.93.66.nondebbin.tar.gz
-rw-rw-r--   1 ian      ian        219884 Aug  3 02:14 dpkg-0.93.66.tar.gz

This is the new C dpkg, available for alpha testing.

To be a candidate alpha tester you need:
 1. To be competent.  You need to know at least roughly how dpkg
    is supposed to work, and you need to be able to edit
 2. Backups of your system.  This version of dpkg has code that can
    call `rm -rf -- <something>' !  I have hedged it with checks,
    but something might go wrong.
 3. Coredumps enabled.

Features in the new release include:
 * Full support for virtual packages.
 * Completely rewritten in C - much faster.
 * No longer vulnerable to bugs in `tar' (better handling of
   symlinks, etc).
 * Better conffiles handling - tends to prompt less about
   files the user has never heard of, and also doesn't routinely
   displace the users' files while unpacking the archive.
 * Better handling of files that always need to be present -
   no more need to use `ln' and `mv' in the maintainer scripts.
 * Better handling of files from one package which overwrite
   some from another.
 * More maintainable.

IMPORTANT: there is a problem with conffiles that will only happen
once, but it will happen to ordinary users as well as you alpha

The conffile naming scheme for unpacked but not configured packages
has changed in this version.  It used to be <foo>.dpkg-tmp for the
user's version and <foo> for the package maintainer's new one.  These
are now <foo> and <foo>.dpkg-new respectively.

This means that the users files are not messed about with
unnecessarily, but it does mean that if you use the new dpkg to
configure a package unpacked with an earlier version, but not yet

You should be aware of this, and use the old Perl dpkg to get packages
in an appropriate state before installing dpkg 0.93.50.

for the same reason - if anything goes wrong, you may find yourself
with dpkg 0.93.50 on the system and some unconfigured packages that
you can't fix.

The preinst script will notice that you're upgrading from pre-0.93.50,
warn you and give you the option to back out of the installation.  It
will also spot packages which are in `bad' states and offer to abort
the installation, and let you know what to do after you've aborted.

The Perl version (equivalent to is also installed, as
`perl-dpkg'.  Don't use it unless absolutely necessary.

Known issues outstanding are:
 * Dependency and conflict processing needs better testing.  If you
   notice odd behaviour please report it, with a copy of your
   /var/lib/dpkg/status file and details of what you did.  Use
   --force-depends, or --ignore-depends=<package>, if necessary.
 * There is no support for reassembling split packages.
 * There are no options yet to set conffiles update behaviour.
 * The binary is currently statically linked, unstripped and contains
   debugging symbols.  It has some malloc-paranoia enabled too.
   I'll change this when the new dpkg is beta.

When you report a bug whose cause isn't immediately obvious, bear in
mind that the following information may make it much easier for me to
find the problem:

(i) If the bug is reproducible, provide the output of running dpkg
   with some useful combination of debugging flags:
              dbg_general=           00001,
              dbg_scripts=           00002,
              dbg_eachfile=          00010,
              dbg_eachfiledetail=    00100,
              dbg_conff=             00020,
              dbg_conffdetail=       00200,
              dbg_depcon=            00040,
              dbg_depcondetail=      00400,
              dbg_veryverbose=       01000,
              dbg_stupidlyverbose=   02000,
   Specify any combination of these using -D<octal>.  -D7777 is likely
   to be very large - if I need that much detail I'll mail you and ask
   for it.

(ii) If you get a segfault: ensure you get a core file, and at the
   very least run gdb on it and mail me the output of `where'.  If
   it's not too large (500Kb, say) gzip it, uuencode it and mail it to
   me in a separate message.  Don't delete it, I may ask for it.

(iii) Provide a copy of your /var/lib/dpkg/status file, preferably
   before and after the problem manifests itself.  gzip and uuencode
   it if it is large.

(iv) Tell me where I can get any .deb files required to reproduce the
   problem.  Provide their MD5 checksums.

Ian Jackson, at home.          ijackson@nyx.cs.du.edu or iwj10@cus.cam.ac.uk
+44 1223 575512    Escoerea on IRC.    http://www.cl.cam.ac.uk/users/iwj10/
2 Lexington Close, Cambridge, CB4 3LS, England.  Urgent: iwj10@thor.cam.co.uk

Reply to: