On Fri, Apr 29, 2022 at 11:08:41AM +0200, Omar Polo wrote:
> A recent mail on ports@ (the net/znc diff) reminded me that there's
> still a corner case of git-style rename diffs that I forgot to handle:
> renames without edits.  Such diffs looks like this
> 
> > diff --git a/net/znc/pkg/DESCR b/net/znc/pkg/DESCR-main
> > similarity index 100%
> > rename from net/znc/pkg/DESCR
> > rename to net/znc/pkg/DESCR-main
> 
> and don't have any hunks.  I don't have any idea why they decided to
> avoid the +++/--- syntax and include a "@@ -0,0 +0,0 @@" since
> compatibility doesn't seem to be a concern...
> 
> Here's a straightforward diff to handle this situation.  I intend to
> rename that "ok" variable in read_patch in a follow-up diff (needs the
> same rename also in parse_hunk and parse_hdr.)  I've also simplified the
> test_patch_rename function other than adding a test case for this.
> 
> comments/ok?

Makes sense, ok.

No idea yet what kind of output 'got diff' will eventually produce
for renames. But we need to handle Git-style diffs in any case.

> P.S.: i forgot that we supported the rename with an empty hunk "@@ -0,0
> +0,0 @@".  Since git doesn't seem to handle it, and neither patch(1), I
> intend to remove the support for it, but again, in a follow-up commit :)

Agreed. If nobody else uses such empty hunks as a rename marker, then we
shouldn't rely on this either.