Re: More problems...

On Mon, 2 May 2005, Anton Altaparmakov wrote:

I do agree.

I think hardlinks are wonderful for

 - "git farms" (ie something like what kernel.org does, but in a more 
   controlled manner - right now kernel.org is really just a standard
   location for different people putting their own files in).

   In this environment, doing hard-linking should also imply 

	- mounting the filesystem "noatime"
	- using a different UID for the hardlinked objects

   ie the "farm administrator" does the hardlinking automatically, and 
   chown()'s them to himself, so that different git trees cannot screw 
   each other up. The "noatime" thing is there because having different 
   users means that git's internal "O_NOATIME" optimization no longer 
   works, and you really want to avoi dgetting lots of write-backs just 
   for "atime".

 - people who have lots of trees. I think Jeff Garzik has something like
   20+ BK trees. At that point, hardlinking just makes sense, and your 
   work patterns are likely to be aware of the different trees anyway.

But for "normal" situations, where you have a tree or two, the hardlinking win might not be big enough to warrant the maintenance headache. With hardlinking, you _do_ need to "trust" the other trees to some degree.

		Linus