On Sat, May 12, 2018 at 1:00 AM, Nguyễn Thái Ngọc Duy <pclouds@xxxxxxxxx> wrote: > There's not much to write here. It's basically a copy from 12/12: > > This 'util' pointer can be used for many different purposes, > controlled in different ways. Some are not even contained in a command > code, but buried deep in common code with no clue who will use it and > how. For example, if revs.show_source is set, then it's used for > storing path name, but if you happen to call get_merge_parent() then > some 'util' may end up storing another thing. This is cool! It makes life easy when it comes to clearing up commits in the object store (as it is unclear what the util pointer is used for we leaked that memory so far in the object store conversion series). > The move to using commit-slab gives us a much better picture of how > some piece of data is associated with a commit and what for. Since > nobody uses 'util' pointer anymore, we can retire it so that nobody will > abuse it again. commit-slab will be the way forward for associating > data to a commit. > > As a side benefit, this shrinks struct commit by 8 bytes (on 64-bit > architecture) which should help reduce memory usage for reachability > test a bit. This is also what commit-slab is invented for [1]. The object store series (called -lookup-final) is at a point where I have to move one of the slabs (buffer_slab in commit.c) into the repository or rather into the parsed_object_pool in object.h. So I think I'll build a series on top of this one first, which allows for not just defining the slabs all over the place, but actually allows them to be embedded into other structs. Then I merge that series with origin/sb/object-store-graft and build the -final series on top of that. Stefan
