On Thu, Aug 29, 2019 at 08:41:58AM -0700, Darrick J. Wong wrote: > A signed s64 comparison would just break the diff_two_keys function > again. The reason for the big dorky comment is to point out that the > signed comparison doesn't work for xfs_btree_query_all, because it does: Even more reason to have a good helper encapsulating and documenting all the caveats :) > I wouldn't mind you cooking up a patch (I think I'm going to be busy > for a few hours digging through all of Dave's patches) but the helper > needs to be cmp_u64. Though ... I also think the logic in the patched > bmbt diff_two_keys is easy enough to follow along. > > (Personally I find the subtraction logic harder to follow, though it > generates less asm code on x64...) The subtraction is a little weird, but very efficient if it works. I'm not sure any of our users is worth the micro-optimization, though. I'll cook up a series over the weekend.
