The old "tree.h" parsing logic (before I did the "tree-walk" tree buffer parser) would generate a list of "struct tree_entry" objects, which was basically a simpler-to-walk version of the raw tree buffer. Now, the problem with that simpler-to-walk version is that it's a lot less efficient than just walking the raw tree buffer in-place, and a lot of the core git tree walkers already use the much more efficient in-place tree-walker interfaces. Which means that we have _two_ independent ways of walking the tree. Very annoying. This series of four patches (on top of my trivial memory leak fix for git rev-list) gets rid of the tree-entry list from tree parsing, and converts most users over to use the tree-walker, and has a small compatibility helper function to keep some users that are hard to convert using what is now obviously just a temporary list. Apart from being one more step in the direction of cleaning up tree walking, it also makes things more efficient, by not needlessly creating that silly list that most users don't actually want. I used the same test vector as I did for the git rev-list leaker: doing a git-rev-list --all --objects on the historical kernel archive (that one has 566k objects and is more than twice the size of the current kernel git tree). The series brings that down from 74.67user 0.45system 1:15.43elapsed to 62.50user 0.41system 1:02.93elapsed for me. NOTE! I think the git-rev-list leak fix can go into the main tree immediately, but this series should probably be kept in "next" for a while, just to verify that it all is good.. Linus - : send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html