Re: git format.from (was: 9p: Fix file ID collisions)

[Jeff King <peff@peff.net>]

On Tue, Sep 24, 2019 at 11:03:38AM +0200, Christian Schoenebeck wrote:

> > Yes, the resulting mail would be correct, in the sense that it could be
> > applied just fine by git-am. But I think it would be uglier. IOW, I
> > consider the presence of the in-body From to be a clue that something
> > interesting is going on (like forwarding somebody else's patch). So from
> > my perspective, it would just be useless noise. Other communities may
> > have different opinions, though (I think I have seen some kernel folks
> > always including all of the possible in-body headers, including Date).
> > But it seems like it makes sense to keep both possibilities.
> 
> Exactly, current git behaviour is solely "prettier" (at first thought only 
> though), but does not address anything useful in real life.

I wouldn't agree with that. By being pretty, it also is functionally
more useful (I can tell at a glance whether somebody is sending a patch
from another author).

> Current git behaviour does cause real life problems though: Many email lists 
> are munging emails of patch senders whose domain is configured for requiring 
> domain's emails being DKIM signed and/or being subject to SPF rules (a.k.a 
> DMARC). So original sender's From: header is then automatically replaced by an 
> alias (by e.g. mailman): https://en.wikipedia.org/wiki/DMARC#From:_rewriting
> 
> For instance the email header:
> 
> From: "Bob Bold" <bold@foo.com>
> 
> is automatically replaced by lists by something like
> 
> From: "Bob Bold via Somelist" <somelist@gnu.org>
> 
> And since git currently always drops the From: line from the email's body if
> sender == author, as a consequence maintainers applying patches from such 
> lists, always need to rewrite git history subsequently and have to replace 
> patch author's identity manually for each commit to have their correct, real 
> email address and real name in git history instead of something like
> "Bob Bold via Somelist" <somelist@gnu.org>
> 
> So what do you find "uglier"? I prefer key info not being lost as default 
> behaviour. :-)

Sure, for your list that munges From headers, always including an
in-body From is way better. But for those of us _not_ on such lists, I'd
much prefer not to force the in-body version on them.

-Peff