bug#81375: diff-refine-hunk over-refines large unrelated replacement blocks

Stefan Monnier [16/Jul  4:27pm -04] wrote:
> Thanks for your demo, it's nice to see it in more concrete terms.
> In your examples, the only cases where I find the refinement distracting
> are case-05-total-rewrite add case-12-repeated-punctuation.  Of course,
> my opinion is biased from looking at such refinements for years, so I've
> trained my brain to "ignore the noise".
>
> Looking at them and thinking about the problem again, I wonder if it
> wouldn't be better to take a more local look at the problem, i.e. rather
> than try to decide whether to refine this hunk or not, look at whether
> some subhunks (by which I mean the hunks returned by the `diff` called
> by `smerge-refine-region`, as opposed to the hunks to which we apply
> `smerge-refine-region`) should be coalesced.
>
> E.g. for case-05-total-rewrite coalescing those subhunks that are separated
> by a single-char token would remove most of the distraction.
>
> Taking such a more local approach could also help in "mixed" cases like
> when a hunk is the result of a complete-rewrite of half a function and
> reindentation of the other half.
>
> WDYT?

Joining up hunks seems like something that ought to be diff(1)'s
business, not Emacs?

-- 
Sean Whitton