Re: [PATCH 0/2] A new merge algorithm, take 3

Linus Torvalds [off-list ref] writes:

So the data structure is optimized for a different case than reading in 
trees. Big deal. That optimization is definitely worth it: it allows us to 
do the read_cache() with the actual index entries being totally read-only 
(a linked list would have to add a "next" pointer to the cache entries and 
not allow the in-place thing that read_cache() does).

Yes, you are right.  What is being discussed is to help
read-tree.c while keeping that nice property.

I think Daniel's patch is going in the right direction.  It can
be told to populate the resulting cache from scratch by reading
the current cache and the trees being read, instead of inserting
and removing the current cache as it does right now.  Once that
is done, we would not have to do repeated memmove to insert and
delete an entry one at a time anymore.