Bug 3500

Laurentiu,

Seems your recent patch should address this issue.

Thanks,
jessica

This looks like a pseudo issue. In the do_install function of the automake recipe we have the following loop [1].

* aclocal-1.12 is hardlink to aclocal
* automake-1.12 is a hardlink to automake

As you can see in the [2] listing, after the loop completes, aclocal-1.12 and automake-1.12 have another ownership. That's because 'sed -i' creates a temporary file first and then does a "mv tmp_file aclocal". This basically deletes the old file and creates another one with a new inode. The hardlink will point to the old inode, which is fine, but the ownership is changed to 'test'! After the loop, we will end up with 4 different files actually.

So, the main question here is: why does the ownerchip changes?

[1]
for i in aclocal aclocal-1.12 automake automake-1.12; do
  if [ -f ${D}${bindir}/$i ]; then
     sed -i -e '##NOT_IMPORTANT##' ${D}${bindir}/$i
  fi
done

[2]
before loop:
6338159 drwxr-xr-x 2 root root   4096 Dec  7 16:12 .
6338158 drwxr-xr-x 4 root root   4096 Dec  7 16:12 ..
6338445 -rwxr-xr-x 2 root root  31882 Dec  7 16:12 aclocal
6338445 -rwxr-xr-x 2 root root  31882 Dec  7 16:12 aclocal-1.12
6338444 -rwxr-xr-x 2 root root 256888 Dec  7 16:12 automake
6338444 -rwxr-xr-x 2 root root 256888 Dec  7 16:12 automake-1.12

after loop:
6338159 drwxr-xr-x 2 root root   4096 Dec  7 16:12 .
6338158 drwxr-xr-x 4 root root   4096 Dec  7 16:12 ..
6340801 -rwxr-xr-x 1 root root  31883 Dec  7 16:12 aclocal
6340802 -rwxr-xr-x 1 test test  31883 Dec  7 16:12 aclocal-1.12
6338445 -rwxr-xr-x 1 root root 256889 Dec  7 16:12 automake
6340803 -rwxr-xr-x 1 test test 256889 Dec  7 16:12 automake-1.12

I found the root cause. It looks like ln, in newer versions of coreutils, uses linkat() system call in order to create a hardlink, instead of link(). Unfortunately, pseudo does not have a linkat() wrapper function and the hard link is not recorded in the database. Hence, when the file the link points to is deleted and the inode dissapears from the DB, the hard link ownership swtiches to the current user.

I will work on implementing a wrapper script for linkat() and give it a try.

Agreed that I need a linkat() wrapper. This is... way harder than I anticipated, partially because the obvious implementation would break strangely on Darwin. Hoping to get a patch done "soon".

I have a possible fix for this up on the SEEBS_TESTING branch of pseudo.

Seebs,

I tested your fixes and the results are good. Hardlink ownership is maintained after the original file is deleted.

Let me know when is the next pseudo release or, if you're not planning a release any time soon, let me know when you merge the changes in master so I can switch to pseudo_git recipe.

Pseudo has been upgraded to 1.4.3. Closing this now.