bug#25410: 26.0.50; Refine an unified diff hunk only if adds lines

[Tino Calancha <tino.calancha@gmail.com>]

On Tue, 10 Jan 2017, npostavs@users.sourceforge.net wrote:

> Tino Calancha <tino.calancha@gmail.com> writes:
>
>> After deletion of a large file from CVS, a diff shows
>> a very large hunk with just deleted lines.  Then, for unified diffs, a call
>> to `diff-refine-hunk' on that hunk takes a huge time.
>> Instead, it's better to first check if the hunk adds new lines: only when
>> this is true, then proceed with the hunk refinement.
>
> What about a diff that adds a very large file?  Perhaps we should only
> refine if there added lines *and* deleted lines?
On Tue, 10 Jan 2017, npostavs@users.sourceforge.net wrote:

>What about a diff that adds a very large file?  Perhaps we should only
>refine if there added lines *and* deleted lines?
That's logical; at the end neither a hunk just deleting nor one
just adding lines need to be refined.  We might do that if you like.  It 
would be more symmetrical.

From a performance point of view, current code in the case where
the hunk just adds lines is not as patological as the opposite one.
For instance:
emacs -Q
M-! git diff ef8c9f8^ ef8c9f8 RET
C-x o C-x C-q
M-x diff-mode RET
R ; Reverse the direction of the diffs.
C-c C-b ; Refine hunk.
;; Perform reasonably fast.