Re: [RFC PATCH 2/5] format-patch: add --range-diff option to embed diff in cover letter

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi Eric,

On Thu, 26 Jul 2018, Eric Sunshine wrote:

> On Thu, Jul 26, 2018 at 6:56 AM Johannes Schindelin
> <Johannes.Schindelin@gmx.de> wrote:
> > On Tue, 17 Jul 2018, Eric Sunshine wrote:
> > > On Tue, Jul 17, 2018 at 6:31 AM Johannes Schindelin
> > > <Johannes.Schindelin@gmx.de> wrote:
> > > > BTW I like to have an extra space in front of all the range-diff
> > > > lines, to make it easier to discern them from the rest.
> > >
> > > I'm not sure what you mean. Perhaps I'm misreading your comment.
> >
> > Sorry, I was really unclear.
> >
> > In the cover letters sent out by GitGitGadget (or earlier, my
> > mail-patch-series.sh command), I took pains to indent the entire
> > range-diff (or interdiff) with a single space. That is, the footer
> > "Range-diff vs v<n>:" is not indented at all, but all subsequent lines
> > of the range-diff have a leading space.
> >
> > The original reason was to stop confusing `git apply` when sending an
> > interdiff as part of a single patch without a cover letter (in which
> > case mail-patch-series.sh inserted the interdiff below the `---`
> > marker, and the interdiff would have looked like the start of the real
> > diff otherwise).
> 
> The new version[1] likewise indents the interdiff to avoid confusing
> git-am / git-apply.
> 
> [1]: https://public-inbox.org/git/20180722095717.17912-1-sunshine@sunshineco.com/

Great!

> > In the meantime, I got used to this indentation so much that I do not
> > want to miss it, it is a relatively easy and intuitive visual marker.
> >
> > This, however, will be harder to achieve now that you are using the
> > libified range-diff.
> 
> I toyed with indenting the range-diff in both the cover letter and
> below the "---" line in a patch. With the libified range-diff, doing
> so involves modifying the range-diff implementation (rather than
> having the consumer of the range-diff manage the indentation locally),
> so it adds a bit of complexity to show_range_diff(), though perhaps
> not too much.
> 
> However, I opted against it for a few reasons. First, "header" lines
> apart, all lines of the range-diff are already indented, and the
> existing indentation was sufficient (for me, at least) as a visual
> marker. Second, range-diffs tend to be _wide_, especially the header
> lines, and I was loath to make it wider by indenting more. Third, due
> to the existing indentation of the diff proper, a range-diff won't
> confuse git-am / git-apply, nor will the unindented header lines, so
> extra indentation seemed superfluous.

Totally understandable. For some reason, I never thought about that fact
(a range-diff is *not* a diff) when changing mail-patch-series.ts to use
range-diffs instead of interdiffs.

> > > > > @@ -1438,6 +1480,7 @@ int cmd_format_patch(int argc, const char **argv, const char *prefix)
> > > > > +     const char *range_diff = NULL;
> > > >
> > > > Maybe `range_diff_opt`? It's not exactly the range diff that is
> > > > contained in this variable.
> > >
> > > I could, though I was trying to keep it shorter rather than longer.
> > > This is still the same in the re-roll, but I can rename it if you
> > > insist.
> >
> > I think it will confuse me in the future if I read `range_diff` and
> > even the data type suggests that it could hold the output of a `git
> > range-diff <options>` run.
> >
> > So I would like to insist.
> 
> In the new version[1], this variable is named 'rdiff_prev' (the
> "previous" version against which the range-diff is to be generated).

Thank you.

> > > > > +format_patch () {
> > > > > +     title=$1 &&
> > > > > +     range=$2 &&
> > > > > +     test_expect_success "format-patch --range-diff ($title)" '
> > > > > +             git format-patch --stdout --cover-letter --range-diff=$range \
> > > > > +                     master..unmodified >actual &&
> > > > > +             grep "= 1: .* s/5/A" actual &&
> > > > > +             grep "= 2: .* s/4/A" actual &&
> > > > > +             grep "= 3: .* s/11/B" actual &&
> > > > > +             grep "= 4: .* s/12/B" actual
> > > >
> > > > I guess this might make sense if `format_patch` was not a
> > > > function, but it is specifically marked as a function... so...
> > > > shouldn't these `grep`s also be using function parameters?
> > >
> > > A later patch adds a second test which specifies the same ranges but
> > > in a different way, so the result will be the same, hence the
> > > hard-coded grep'ing. The function avoids repetition across the two
> > > tests. I suppose I could do this a bit differently, though, to avoid
> > > pretending it's a general-purpose function.
> >
> > If you can think of a way that would make this easier to read for,
> > say, myself if I ever find myself debugging a regression caught by
> > this test, I would appreciate that.
> 
> In the new version, the function is gone; it looks like this:
> 
> --- >8 ---
> for prev in topic master..topic
> do
>     test_expect_success "format-patch --range-diff=$prev" '
>         git format-patch --stdout --cover-letter --range-diff=$prev \
>             master..unmodified >actual &&
>         grep "= 1: .* s/5/A" actual &&
>         grep "= 2: .* s/4/A" actual &&
>         grep "= 3: .* s/11/B" actual &&
>         grep "= 4: .* s/12/B" actual
>     '
> done
> --- >8 ---

Looks good.

Thank you so much!
Dscho