Re: [PATCH 3/6] t/perf: add infrastructure for measuring sizes

From: Jeff King <peff@peff.net>
Date: Aug 22, 2018, 15:31 UTC
Message-ID: <20180822153104.GE32630@sigill.intra.peff.net>
In-Reply-To: <b071f6a2-2403-0bbb-72a5-695f8c4e724d@gmail.com>

On Wed, Aug 22, 2018 at 09:40:55AM -0400, Derrick Stolee wrote:

Show 14 quoted lines

> On 8/21/2018 3:06 PM, Jeff King wrote:
> > The main objective of scripts in the perf framework is to
> > run "test_perf", which measures the time it takes to run
> > some operation. However, it can also be interesting to see
> > the change in the output size of certain operations.
> > 
> > This patch introduces test_size, which records a single
> > numeric output from the test and shows it in the aggregated
> > output (with pretty printing and relative size comparison).
> 
> I'm interested in exploring this test_size mechanism. The other area that
> could benefit from size testing is 'git repack', but I don't have any plans
> to change our compression or delta strategies. If we _did_ look into that,
> then using test_size would be a natural fit.

Yeah, I agree it would be a good tool for showing off improvements there.

It may also be useful for catching regressions in topics that are trying to speed things up, but _don't_ intend to change the size. We could even do that proactively now. I.e., something like:

  test_perf 'repack' 'git repack -adf'
  test_size 'pack size' 'wc -c <.git/objects/pack/*.pack'

just to see if it ever changes. But I suspect its usefulness may depend on how you are packing (e.g., is "-f" more likely to catch issues than without?).

The new tests I added in this series cover packs created for fetches. There's no guarantee that will overlap with the behavior of an on-disk repack, but at least we have some generic coverage of pack-objects output sizes now. Absent any suspicion of a regression for a particular case, that's probably an acceptable canary in the coal mine.

-Peff