On Mon, Dec 15, 2008 at 05:15:44PM -0800, Junio C Hamano wrote:
Perhaps the two runs are seeing different number of CPUs (hence threads)
available? That would then change the distribution of the work itself
(i.e. what slice of obj-list goes as a single chunk to be processed) and
would affect the outcome.
I am seeing the same failure as Mike on one of my boxen. Each run sees
the same number of threads (4 in my case).
Does the second test this patch adds fail?
With your patch, the added "both should match" test fails. So it seems
to be about doing a second run at all, not the difference in index
versions.
-Peff