Re: [PATCH 2/2] xdiff: implement empty line chunk heuristic

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 18, 2016, 22:24 UTC
Message-ID: <xmqq7ffu4ktu.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <CA+P7+xrisA0qqQ01GoSUdNm+O85NN9H7arovzqDD2e5GUv2GAw@mail.gmail.com>

Jacob Keller <jacob.keller@gmail.com> writes:

> Thanks Stephan and Junio, this looks pretty good. I think before it's
> merged we'd probably want to implement some sort of attributes which
> allows per-path configuration, incase it needs to be configured at
> all.

My take on it is that we'd want to make sure that the shift with blank line heuristics is "good enough", i.e. there is no need for end-user configuration or attributes, and then remove the tentative option, configuration and its documentation, before this is merged.

If we really want to add knobs to handle different kind of payloads in vastly different way, the right place to do so is to add a set of bits "use this and that heuristics" to userdiff driver, I would say, but in the compaction codepath it does not seem to be enough room to have that many knobs to be tweaked in the first place to me.

> I've got it applied to my local git, and I'm going to try to run a
> diff between enabled vs disabled on a large section of the Linux
> kernel history and a few other projects to see if I spot anything odd.

Thanks.