'Minimal' diff-algorithm producing a different result than 'myers', 'patience' and 'histogram' ones

Dmitry Malikov· May 12, 2015, 13:07 UTC · lore

Good day,

I'm trying to compare 4 different git-diff algorithms and the 'minimal' one is the most vague and non-obvious. The documentation says "Spend extra time to make sure the smallest possible diff is produced." - that's all.

By any chance, is there any example of diff when 'minimal' algorithm produces a different result than a 'myers', 'patience' and 'histogram' ones?

PS: in case of a question like 'why do you need that?' - it's for
testing purposes of some git wrapper library.

Jeff King· May 24, 2015, 07:33 UTC · re: Dmitry Malikov · lore

Re: 'Minimal' diff-algorithm producing a different result than 'myers', 'patience' and 'histogram' ones

On Tue, May 12, 2015 at 03:07:46PM +0200, Dmitry Malikov wrote:

Show 8 quoted lines

> I'm trying to compare 4 different git-diff algorithms and the
> 'minimal' one is the most vague and non-obvious. The documentation
> says "Spend extra time to make sure the smallest possible diff is
> produced." - that's all.
> 
> By any chance, is there any example of diff when 'minimal' algorithm
> produces a different result than a 'myers', 'patience' and 'histogram'
> ones?

I don't know of a simple example offhand, but you can easily generate all of the patches for a repository with each type by doing:

  for i in myers minimal patience histogram; do
    git log --diff-algorithm=$i -p >$i
  done

In git.git, the output for each type is distinct. You might also find this thread interesting:

  http://thread.gmane.org/gmane.comp.version-control.git/143426

It mentions 717b83117 from git.git, whose minimal diff is over 300 lines shorter than the non-minimal one. It's not exactly a simple example, but I suspect you won't find a short case; "minimal" only matters in the complicated ones.

-Peff