On Wed, Aug 16, 2023 at 12:22:33PM +0100, Simon McVittie wrote: > I personally think the extent to which git has "won", both upstream and in > Debian package maintenance, means that the Patch Tagging Guidelines should > be encouraging the git-like style as primary, with From/Date/Subject > in a header and all Debian-specific fields in a trailing pseudo-header, > the same layout that's familiar for Signed-off-by and similar tags used > by many upstream developers: > > From <anything here, optional, ignored> > From: ... > Date: ... > Subject: <short description> > > <long description> > > <more long description> > > Signed-off-by: ... > Bug: https://... > Origin: vendor, Debian > Forwarded: https://... > --- > <the actual diff> While I completely agree with everything that you've said, I will additionally point out that a notable use of patches is backporting upstream fixes to older package releases. In that context, it's generally useful (best practice?) to call 'git cherry-pick -x' on the upstream commit, which will result in git generating a line like (cherry-picked from commit ...) In that case, the patch metadata should IMO go *after* that line, leaving the upstream commit message completely unchanged. See [1] for an example of how that looks like in practice. Incidentally, notice how I've had to include Forwarded: not-needed Origin: https://gitlab.com/libvirt/libvirt/-/commit/... in order for tracker.d.o not to report this backport as a patch that needs forwarding upstream. I feel that the presence of a "(cherry-picked from commit ...)" line should imply at least the first bit, but maybe there are scenarios that I haven't considered and that make being explicit about it a necessity. [1] https://salsa.debian.org/libvirt-team/libvirt/-/blob/26c4e495779fc9e7649956983818735bbc7300fd/debian/patches/backport/src-fix-max-file-limits-in-systemd-services.patch -- Andrea Bolognani <eof@kiyuko.org> Resistance is futile, you will be garbage collected.
Attachment:
signature.asc
Description: PGP signature