[PATCH 0/7] arm64 / x86-64: low-level code generation issues

[Linus Torvalds]

On Tue, 11 Jun 2024 at 10:24, David Woodhouse <dwmw2 at infradead.org> wrote:
>
> I've removed the footer.

Thanks.

> Even with the footer though, the messages should have had valid DKIM
> signatures for the @lists.infradead.org address which is actually the
> Sender: of the message which comes via the list.

Ahh. I only looked at the body differences, not the header differences
(which will obviously be much bigger for other reasons).

> I'm a bit surprised that lore screws it up too; mailing lists always
> used to add a footer (mostly because humans are too stupid to remember
> how to unsubscribe otherwise); not doing so is a relatively recent
> phenomenon. So it really *ought* to cope.

So lore (well, technically the infra it uses: public-inbox) doesn't
care about the DKIM bit, but lore basically indexes emails by message
ID - like you are supposed to.

And if it sees two different emails that have the same message ID, it
doesn't just decide to keep one of them randomly. Instead if indexes
them both, and makes the thread very hard to read, and makes unhappy
noises.

Normally it's not that noticeable, but it's visible when the same
email gets to lore because it was cc'd to two different mailing lists,
and one of them messed with the message and the other did not (or
messed with it differently).

So for this particular thread, see the results at

   https://lore.kernel.org/all/20240610204821.230388-1-torvalds@linux-foundation.org/

and scroll down to see the threading thing - all my emails show up
twice, and with insane threading, because the extra one added by this
"two different emails with the same message ID" issue.

You also see the message twice, with a line

   WARNING: multiple messages have this Message-ID (diff)

by lore / public-inbox.

Which is annoying in situations like this, but is actually quite
reasonable from a technical angle. People (and mailing lists) should
*not* send a new email with an old message ID. When you see two
messages with the same ID, which one is the "real" one?

Now, a mailing list could obviously just make up a new message ID when
it changes the message - the same way you apparently just re-DKIM it -
but that causes even more problems for threading, so no, that's not
the answer either.

The answer is for mailing lists to not mess with the body, and to not
mess with standard headers. Instead, just add message-list headers. So
for example, the normal kernel.org message lists add headers like
this:

  X-Mailing-List: linux-kernel at vger.kernel.org
  List-Id: <linux-kernel.vger.kernel.org>
  List-Subscribe: <mailto:linux-kernel+subscribe at vger.kernel.org>
  List-Unsubscribe: <mailto:linux-kernel+unsubscribe at vger.kernel.org>

and that doesn't mess with DKIM _or_ the Message-ID rules, so the
kernel.org lists don't need to make up a DKIM thing either, so it
actually validates properly as being from the original sender.

                   Linus