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

[Tom Lord]

    > From: Chris Mason <address@hidden>

    > 1) Do inode signatures actually help performance in the current form?  I
    > think they make most uses slower (except revision libraries).  


A while back, briefly, they improved performance for implicit or
tagline trees.  Currently, they do impose a new cost -- for the return
of increasing safety.  The safety concern is worth it, I think --
those costs aren't likely to go away.

(The other pending issue is that they hork some NFS implementations.
That has to be fixed.)

    > 2) Does a reverse mapping safely allow arch_apply_changeset to skip
    > whole tree inventories?  

Unlikely but I haven't dug into it yet.

    > > Doing a binary comparison before diffing is a solution to this
    > > problem,

And one that tla universally applies.  The only performance issue
that's been identified so far is that tla sometimes creates a tmp file
to hold diff output _before_ doing the binary comparison -- only to
delete the tmp file afterwords if there are no differences (which is
the expected case).  In environments where file creation and deletion
are expensive, that's a needless expense.


-t