From: Shaoxuan Yuan <shaoxuan.yuan02@gmail.com>
Date: Sat, 06 Aug 2022 03:18:14 GMT
Subject: Re: [PATCH v1 4/4] rm: integrate with sparse-index
Message-ID: <12afcbe9-218b-528d-6d81-f39628388ba9@gmail.com>
In-Reply-To: <999169c6-a727-af2a-3361-51ac7b1f1d80@github.com>

On 8/4/2022 10:48 PM, Derrick Stolee wrote:
 > On 8/3/2022 12:51 AM, Shaoxuan Yuan wrote:
 >> Enable the sparse index within the `git-rm` command.
 >>
 >> The `p2000` tests demonstrate a ~96% execution time reduction for
 >> 'git rm' using a sparse index.
 >
 > Sorry that I got sidetracked yesterday when I was reviewing this
 > series, but I noticed something looking at these results:
 >
 >> Test                                     before  after
 >> -------------------------------------------------------------
 >> 2000.74: git rm -f f2/f4/a (full-v3)     0.66    0.88 +33.0%
 >> 2000.75: git rm -f f2/f4/a (full-v4)     0.67    0.75 +12.0%
 >
 > The range of _growth_ here seemed odd, so I wanted to check if this was
 > due to a small sample size or not.

Yes, I do feel they are odd, as I've been checking pervious
integrations and p2000 results, they usuallly fall below 10% range.
But I was not discerning enough to name a problem :-(

 >> 2000.76: git rm -f f2/f4/a (sparse-v3)   1.99    0.08 -96.0%
 >> 2000.77: git rm -f f2/f4/a (sparse-v4)   2.06    0.07 -96.6%
 >
 > These numbers are as expected.
 >
 >>  test_perf_on_all git read-tree -mu HEAD
 >>  test_perf_on_all git checkout-index -f --all
 >>  test_perf_on_all git update-index --add --remove $SPARSE_CONE/a
 >> +test_perf_on_all git rm -f $SPARSE_CONE/a
 >
 > At first, I was confused why we needed '-f' and thought that maybe
 > this was turning into a no-op after the first deletion. However, the
 > test_perf_on_all helper does an "echo >>$SPARSE_CONE/a" before hand,
 > so the file exists _in the worktree_ every time. That requires '-f'
 > since otherwise Git complains that we have modifications.

Yeah, it took me some time to find out.

 > However, after the first instance the file no longer exists in the
 > index, so we are losing some testing of the index modification.

So true, I didn't realize at all.

 > We can fix this by resetting the index in each test loop:
 >
 >   test_perf_on_all "git rm -f $SPARSE_CONE/a && git checkout HEAD -- 
$SPARSE_CONE/a"
 >
 > Running this version of the test with GIT_PERF_REPEAT_COUNT=10 and
 > using the Git repository itself, I get these numbers:
 >
 > Test                              HEAD~1            HEAD
 > 
--------------------------------------------------------------------------
 > 2000.74: git rm ... (full-v3)     0.41(0.37+0.05) 0.43(0.36+0.07) +4.9%
 > 2000.75: git rm ... (full-v4)     0.38(0.34+0.05) 0.39(0.35+0.05) +2.6%
 > 2000.76: git rm ... (sparse-v3)   0.57(0.56+0.01) 0.05(0.05+0.00) -91.2%
 > 2000.77: git rm ... (sparse-v4)   0.57(0.55+0.02) 0.03(0.03+0.00) -94.7%
 >
 > Yes, the 'git checkout' command is contributing to the overall
 > numbers, but it also already has the performance improvements of
 > the sparse-index, so it contributes only a little to the performance
 > on the left.
 >
 > (Also note that the full index cases change only by amounts within
 > reasonable noise. The repeat count helps there.)

New thing learned, repeat to average out noise.

--
Thanks,
Shaoxuan