Re: [Gnu-arch-users] [PATCH] arch speedups on big trees

[Tom Lord]

    > From: Chris Mason <address@hidden>

    > The part we won't catch is if the merge source creates a file at
    > one location and that file already exists at a different
    > location with the same id.  Looking at the the changeset code, I
    > don't see how that gets handled right now.

asuffield's measurements suggest that:

(a) a significant part of what your hack does is redundent with
    another optimization already planned for 1.2

(b) after that hack, it's still worth ~10% (relative to current
    performance, in unscientific measurements, on one platform at
    least, but this is still significant) to just prune your hack
    back to apply to only those cases where we know we're
    exact-patching.   Those are some important common cases (e.g.,
    library-add, get, and update).

-t