Re: [PATCH 7/3] Make index preloading check the whole path to the file

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Jul 11, 2009, 03:04 UTC
Message-ID: <alpine.LFD.2.01.0907101957200.3552@localhost.localdomain>
In-Reply-To: <7veisorkux.fsf@alter.siamese.dyndns.org>

On Fri, 10 Jul 2009, Junio C Hamano wrote:

> 
> I actually was hoping to hear "Didn't you notice that this is the first
> function run by the pthread and its stack is zeroed by thread creation" or
> something clever like that ;-)

It's probably true that it is often zero in practice. I certainly saw no problems in my testing, even though I do have preloading on (partly for testing, partly because it actually helps a bit on my machine).

I also suspect that the way the whole 'cache_def' thing works, even if it's initialized with random crud, you'll probably never notice. There are all those safety rules that check that 'cache->track_flags' has to match the new value etc in order for the cache to be used. And even when it is used, it has no pointers in it, it has that static array and the lengths.

So I don't think you really even need to have the "it was zeroed by accident" explanation. It's probably as simple as "even if it is totally uninitialized, that will basically never trigger anything odd in practice".

Not to mention that the whole new index preloading addition was just a new safety feature that we didn't even use to have before - and one that only impacted an _optimization_ that didn't change semantics. So in the end: even in the really unlikely situation that the cache would have triggered, and returned an incorrect return value, the worst that would have happened would be that the preloading wasn't quite as efficient.

End result: you did well by noticing the lack of initializers, but I _really_ don't think it could probably ever possibly have mattered in practice.

			Linus