Re: Revision walking, commit dates, slop

From: Derrick Stolee <stolee@gmail.com>
Date: May 20, 2019, 01:33 UTC
Message-ID: <f14799c3-e343-eb41-3536-65de7e38fbd9@gmail.com>
In-Reply-To: <20190518041706.ct6ie5trvxgdhjar@glandium.org>

On 5/18/2019 12:17 AM, Mike Hommey wrote:

Show 23 quoted lines

> On Sat, May 18, 2019 at 12:58:28PM +0900, Mike Hommey wrote:
>> On Sat, May 18, 2019 at 03:50:05AM +0200, SZEDER Gábor wrote:
>>>
>>> All the above is without commit-graph, I presume?  If so, then you
>>> should give it a try, as it might bring immediate help in your
>>> pathological repo.  With 5k commit in the same second (enforced via
>>> 'export GIT_COMMITTER_DATE=$(date); for i in {1..5000} ...') I get:
>>>
>>>   $ best-of-five -q git rev-list HEAD~..HEAD
>>>   0.069
>>>   $ git commit-graph write --reachableComputing commit graph generation
>>>   numbers: 100% (5000/5000), done.
>>>   $ best-of-five -q git rev-list HEAD~..HEAD
>>>   0.004
>>
>> I'm not observing any difference from using commit-graph, whether in
>> time or in the number of commits that are looked at in limit_list().
> 
> -c core.commitGraph=true does make a difference in time, but not in the
> number of commits looked at in limit_list(). So it's only faster because
> each iteration of the loop is faster. It means it's still dependent on
> the depth of the dag, and the larger the repo will grow, the slower it
> will get.

The plan is to use the commit-graph's generation numbers for these A..B queries, but due to some cases when commit date is a _better_ heuristic than generation numbers, we have not enabled them for A..B. You'll see that 'git rev-list --topo-order -n 1 HEAD` will be much faster with the commit-graph, but adding '--topo-order' to your 'HEAD~1..HEAD' query should not change the time at all.

See [1] for the discussion about "generation number v2" which will allow us to use a better heuristic in these cases.

Thanks, -Stolee

[1] https://public-inbox.org/git/6367e30a-1b3a-4fe9-611b-d931f51effef@gmail.com/