Re: [PATCH/RFC] range-diff: make output format more useful for "rebase --onto"

On Fri, Sep 17, 2021 at 04:04:01PM -0700, Junio C Hamano wrote:

In a range-diff output, we often see an early part of an updated
series having no changes since the previous iteration.  After
applying an updated patch submission to the same base as the
previous round,

    $ git range-diff master..topic@{1} master..topic

to view the differences since the previous edition, we might see
something like this:

[...]

Hmm... looks familiar :).

The updated output format looks like this instead:

    1 = 1 a05f02b1d9 a05f02b1d9 : t/helper/test-bitmap.c: add  ...
    2 = 2 8de300e1f7 8de300e1f7 : pack-bitmap.c: propagate nam ...
    3 = 3 caca3c9f07 caca3c9f07 : midx.c: respect 'pack.writeB ...
    4 = 4 2082224f17 2082224f17 : p5326: create missing 'perf- ...
    5 = 5 97b89c8150 97b89c8150 : p5326: don't set core.multiP ...
    6 < - a1dd4c97b9 ---------- : p5326: generate pack bitmaps ...
    - > 6 ---------- bf4a60874a : p5326: generate pack bitmaps ...
    7 ! 7 2b909ebad3 54156af0d6 : t5326: test propagating hash ...
	@@ t/t5326-multi-pack-bitmaps.sh: test_expect_success 'pack.preferBitmapTips' '
	 +	(
	 +		cd repo &&

I do not feel strongly about either format, and can understand an appeal
for both. I think that I find the existing format a little clearer to
read, but it's possible that I'm just biased since it's always been
that way.

But I can equally understand why it would be nicer to have both hashes
next to each other separated by whitespace.

I also don't care to make it configurable, or to work on more flexible
formatting options for range-diff. I would be fine with the proposed
format. As an alternative, something like this could work:

  perl -ale 'if (@F[2] ne "=") { print "$l $r"; exit }; $l = @F[1]; $r = @F[4]'

when piped the output of range-diff with --no-patch. That gives you
empty output when the two series aren't related, which may or may not be
what you want.

Thanks,
Taylor