On Tue, 3 Apr 2007, Nicolas Pitre wrote: > > Are hard numbers like 7% overhead (because right now that's all we have) > really worth it against bad _perceptions_? If it actually stays at just 7% even with large repos (and the numbers from Chris seem to say that it doesn't get worse - in fact, it may be that the lookup gets relatively more efficient for a large repo thanks to the log(n) costs), I agree that 7% probably isn't worth worrying about when weighed against "guaranteed no SHA1 collision". Especially as long as you'd normally only hit it when your real performance issue is going to be the network. So especially if we can make sure that the *local* case is ok, where the network isn't going to be the bottleneck, I think we can/should do the paranoia. That's especially true as it is also the local case where the 7% has already been shown to be just the best case, with the worst case being many hundred percent (and memory use going up from 55M to 280M in one example), thanks to us actually *finding* the objects. Linus - To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html