Re: [PATCH] Speedup recursive by flushing index only once for all entries

From: Alex Riesen <fork0@t-online.de>
Date: Jan 11, 2007, 21:48 UTC
Message-ID: <20070111214826.GC6058@steel.home>
In-Reply-To: <Pine.LNX.4.64.0701111001150.3594@woody.osdl.org>

Linus Torvalds, Thu, Jan 11, 2007 19:02:52 +0100:

Show 10 quoted lines

> > > That said, I think we actually have another problem entirely:
> > > 
> > > Look at "write_cache()", Junio: isn't it leaking memory like mad?
> > 
> > Unless it is used in some corner case not covered by tests - it
> > looks like it does leak memory like mad. With the patch the
> > memory usage for 44k-merge is more than halved!
> 
> Is that halving on _top_ of your and Junio's fixes to not flush 
> unnecessarily?

Yes.

> And it does pass all the tests, although I don't know how much coverage 
> they have in this area..

Quite a bit: criss-cross, renames, simple. My 44k-merge is just a primitive two-head merge useful only as a stress test (it doesn't even has any renames).