pbin archives should restore hard links efficiently (instead of making multiple copies)

[chutz]

Packages installed via a binary do not preserve hard links while installing the package directly from the source repository does preserve the hard links.

I see that gnu tar, and even bsdtar are capable of maintaining hard links, so it seems that Paludis should be able to do the same, too.

Demonstration

1. Git installed from a source repo to the root system. /usr/bin/git-receive-pack has 109 links

   > stat /usr/bin/git-receive-pack
     File: `/usr/bin/git-receive-pack'
     Size: 1314360         Blocks: 2568       IO Block: 4096   regular file
   Device: fd01h/64769d    Inode: 3171604     Links: 109
   Access: (0755/-rwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
   Access: 2012-02-25 21:27:48.539721714 +0900
   Modify: 2012-02-25 21:27:50.959721299 +0900
   Change: 2012-02-25 21:27:55.429720538 +0900
    Birth: -
   

2. A PBIN of git, which contains no hard link information

   > tar tvf /tmp/desktop-bin--dev-vcs--git-1.7.8.4--C.18361.1330169606.850618.C.tar.bz2  | grep -e bin/git-receive -e bin/git-upload-archive
   -rwxr-xr-x root/root   1314360 2012-02-25 20:33 usr/bin/git-receive-pack
   -rwxr-xr-x root/root   1314360 2012-02-25 20:33 usr/bin/git-upload-archive
   -rw-r--r-- root/root    171923 2012-02-25 20:33 usr/lib64/debug/usr/bin/git-receive-pack.debug
   

3. libarchive *can* archive hardlinks efficiently (version 3.0.3 here)

   > bsdtar  -cf - --format gnutar -C / usr/bin/git usr/bin/git-receive-pack usr/bin/git-upload-archive  | tar tv
   -rwxr-xr-x root/root   1314360 2012-02-25 21:27 usr/bin/git
   hrwxr-xr-x root/root         0 2012-02-25 21:27 usr/bin/git-receive-pack link to usr/bin/git
   hrwxr-xr-x root/root         0 2012-02-25 21:27 usr/bin/git-upload-archive link to usr/bin/git
   

[ciaranm]

Can you figure out how to get libarchive to do that when using the library?

[chutz]

Much to my surprise it was easier than expected.

I installed git from a binary package and now my /usr/bin/git has 109 links, as it should. And the binary package itself is considerably smaller:

$ ls -hgG desktop-bin--dev-vcs--git-1.7.8.4--*
-rw-r--r--+ 1  69M Feb 25 15:14 desktop-bin--dev-vcs--git-1.7.8.4--C.4199.1330150437.946996.C.tar.bz2
-rw-r--r--+ 1 4.3M Feb 26 18:30 desktop-bin--dev-vcs--git-1.7.8.4--C.4320.1330248606.623596.C.tar.bz2

All I had to update was the archiver. On extraction, tar can handle the hardlinks itself.

I have not tried this with the other archive formats, though I would expect it to still work. I did try the following, though, and it seems okay:

$ touch f1; ln f1 f2; mkdir out
$ bsdtar -cf - --format pax f? | /usr/libexec/paludis/utils/unpaxinate img /dev/stdin out
f1
f2
$ stat -c %i:%n out/f1 out/f2
2107974:out/f1
2107974:out/f2

Here is the patch:

The following changes since commit 8f4f01deaf9fad78ec0a6c9cd5928855109a054c:

  Don't try to dodoc directories. (2012-02-21 15:05:28 +0000)

are available in the git repository at:
  git://github.com/chutzimir/paludis.git pbin-hardlinks

Georgi Georgiev (1):
      Support hardlinks in pbins

 paludis/tar_extras.cc |   35 +++++++++++++++++++++++++----------
 1 files changed, 25 insertions(+), 10 deletions(-)

[ciaranm]

Thanks. I added in a missing archive_entry_linkresolver_free.