[OE-core] [PATCH 37/61] lib/oe/package: Ensure strip breaks hardlinks

[Richard Purdie]

On Wed, 2015-02-18 at 19:25 +0100, Enrico Scholz wrote:
> Armin Kuster <akuster808-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org>
> writes:
> 
> > Normally, strip preserves hardlinks which in the case of the way our
> > hardlink rather than copy functionality works,
> 
> This breaks stripping of hardlinked files.  'package.bbclass' contains
> logic to run 'strip' only once per hardlinked file.  With this patch,
> only one instance gets stripped.
> 
> E.g. the 'e2fsprogs-mke2fs' package contains now
> 
> -rwxr-xr-x    4 root     root        350025 Feb 18 01:42 /sbin/mke2fs
> -rwxr-xr-x    4 root     root        350025 Feb 18 01:42 /sbin/mkfs.ext2
> -rwxr-xr-x    4 root     root        350025 Feb 18 01:42 /sbin/mkfs.ext3
> -rwxr-xr-x    4 root     root        350025 Feb 18 01:42 /sbin/mkfs.ext4
> -rwxr-xr-x    1 root     root         84700 Feb 18 01:42 /sbin/mkfs.ext4dev
> 
> where the first 4 programs (which are one inode with 4 hardlinks) are
> not stripped.

There is a problem here for sure. The issue is that for
space/performance reasons we hardlink the files in the sysroot with the
ones in the packaging process amongst other places. If one set gets
stripped, before this patch they'd all get stripped and the process was
not deterministic. Were the files in the sysroot stripped or unstripped
for example?

So the patch addresses a real issue.


> > -    stripcmd = "'%s' %s '%s'" % (strip, extraflags, file)
> > +    # Use mv to break hardlinks
> > +    stripcmd = "'%s' %s '%s' -o '%s.tmp' && mv '%s.tmp' '%s'" % (strip, extraflags, file, file, file, file)
> 
> How are these files then merged again?

They're not :(. So we have a different problem introduced by the patch.
I'll give this a bit more thought overnight. We might need to implement
some code to merge these back given we already maintain the lists of
files.

Cheers,

Richard