pagaltzis at gmx.de
Thu Apr 14 23:06:45 PDT 2011
* Matthew Wickline <m_pm_pdx at wickline.org> [2011-04-14 07:35]:
> I've added some examples. If you diff -u the before and after
> files you should get something like the png at
> Note that the bottom two block edits didn't render as block
> edits because a rows was added or inserted. The script does not
> try to merge lines when presented with N minus lines followed
> by other-than-N plus lines. It only wants to merge N minus
> lines with N plus lines.
Ahh. I see. That was the case I was hitting at least in several
of the cases it seemed to not be doing anything.
It finally worked fine on other diffs since, so it *is* working.
> Another case you might be hitting is that if too many edits are
> done in a line, it decides that it would look better to show
> you "this whole line changed into this other whole line" rather
> than trying to merge the two lines.
No. But there was one case where I had a single-line change where
into which I had inserted a short substring, that diff_painter
was not rendering as I expected, even though it seemed that it
should be able to. Unfortunately I didn’t write down the commit
in question so I can’t re-check now.
Was this possibly because `git diff` merged the one-line change
into a single diff hunk together with another change where one
line was split into three?
In that case it means diff_painter would be able to sprinkle its
magic on more of the diff if the context is reduced (`-U0` in the
extreme) – right?
If so then it would be better to split the diff hunks on its own
(ie collect +/- stretches between unchanged lines) instead of
only relying on the numbers in the hunk head, and then operate
on these sub-hunks. (Otherwise it’s useless for complex changes
best read using `-U999` to get a before/after of the whole file,
where it could do the most to help the reader.)
> As for input, basically anything that looks like diff or diff
> -u. That would include svn diff and git diff and git log -p.
Aristotle Pagaltzis // <http://plasmasturm.org/>
More information about the Pdx-pm-list