Re: [PATCH v11 2/5] p0006-read-tree-checkout: perf test to time read-tree

From: Jeff King <peff@peff.net>
Date: Apr 19, 2017, 01:25 UTC
Message-ID: <20170419012528.2xjnwrpgvlsgzszj@sigill.intra.peff.net>
In-Reply-To: <20170418214025.GA4989@hank>

On Tue, Apr 18, 2017 at 10:40:25PM +0100, Thomas Gummerer wrote:

Show 8 quoted lines

> > +test_perf_default_repo
> 
> I like that it's possible to use a real world repository now instead
> of forcing the use of a synthetic repository :)
> 
> Is there a reason for this being test_perf_default_repo instead of
> test_perf_large_repo?  It seems like generating a large repo is what
> you are doing with repos/many-files.sh.

I'm actually of the opinion that the default/large repo thing should go away. I think the original premise was that you could pick a default/large pair and run the whole suite against them. But in reality, I have always been confused about which one I should use when writing a perf test, and what I should use when running the suite.

I think it would be more useful for the perf tests to just respect a single repo parameter. Then you could run the whole suite against each repo in turn. And we could get results for git.git, linux.git, some synthetic cases, the gigantic Windows repo, etc.

-Peff