Re: Lockless Refs?

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 4, 2013, 21:28 UTC
Message-ID: <7vip7ctz7j.fsf@alter.siamese.dyndns.org>
In-Reply-To: <201301031652.44982.mfick@codeaurora.org>

Martin Fick <mfick@codeaurora.org> writes:

> Any thoughts on this idea?  Is it flawed?  I am trying to 
> write it up in a more formal generalized manner and was 
> hoping to get at least one "it seems sane" before I do.

The general impression I have been getting was that this isn't even worth the effort and the resulting complexity of the code, given Peff's observations earlier in the thread that ref update conflicts and leftover locks are reasonably rare in practice. But perhaps I has been mis-reading the discussion.

I also have this suspicion that if you really want to shoot for multi-repository transactions in an massively scaled repository hosting environment, you would rather want to not rely on hacks based on filesystem semantics, but instead want to RPC with a dedicated "ref management service" that knows the transaction semantics you want, but that could become a much larger change.

I dunno.