Re: [Gnu-arch-users] Re: tla get --hard-links available

[Andrea Arcangeli]

On Fri, Oct 03, 2003 at 07:10:27PM +0900, Miles Bader wrote:
> Karel Gardas <address@hidden> writes:
> > I'm sorry for asking silly question, but before going into deep-details
> > with hard links hacking, have you done any performance test? i.e. as an
> > Arch user I'm quite worried with this direction of development...
> 
> Note that many people (including me) already use this style of
> development with no problems, not with arch, but by maintaining many
> hard-linked source trees (made with e.g. `cp -al') and relying on
> programs breaking hard-links to COW modified files.  It works fine as

exactly. And that's the only way to handle huge trees IMHO. that's also
why all other SCM are hardly usable with huge trees during checkout and
diff, they don't work that way. The cool thing of arch is that it *can*
work that same way ;) Comparing the inode number instead of the data
payload (that doesn't even need to be loaded in cache from disk for
equal inodes, the directory metadata is enough), and sharing all the
pagecache and disk space is an *huge* boost.

I did a mistake only once, no idea how or when (that's why it's a
mistake after all, if I noticed about it it wouldn't happen in the first
place ;), and luckily I noticed it before the corruption spread to deep
affecting my patches too.

the LD_PRELOAD lib from Davide sounds like the final solution to this problem.

However it segfaults and it even spawns illegal instructions on 64bit
;), I need to look at it, it may not be an issue in the library but in
the linker.

Andrea - If you prefer relying on open source software, check these links:
            rsync.kernel.org::pub/scm/linux/kernel/bkcvs/linux-2.[45]/
            http://www.cobite.com/cvsps/