Phillip Wood [off-list ref] writes:

With hindsight I should have been clearer here that the advice given
is based on the user's config settings.

Ahh, OK.  If the "hint" advice message gets generated with custom
sequence of commands, that explains why the sample looked so uneven.
Disregard what I said about clearing every variant from every scope.

The advice will recommend a command that updates commentChar in the
scope where it is currently set so if it is set globally it will not
prompt you to set it locally in each repository and if it is set
locally it will prompt you to update it there.

Again, I misunderstood the set-up that would lead to the sample
output.  If the user has "auto" in ~/.gitconfig, replacing it at the
same place may make sense.

If the "auto" comes from /etc/gitconfig then we'd recommend
changing it there, instead of overriding it per-user in ~/.gitconfig?

quoted

It would be necessary to special case "auto" after 3.0 boundary
anyway, whether we (1) die when we notice the value is set to
"auto", and refuse to work until the user chooses a comment char, or
(2) use "#" or something hardcoded.  Either would be better than
using literal string "auto" as comment char.

We can do that if you've changed your view from
<xmqqfrj6vfsn.fsf@gitster.g>

Yeah, I think using "auto " as comment line prefix is simply a
nonsense.  Thanks.

quoted

So, a simpler approach might be to treat literal string "auto" as if
"#" was specified under WITH_BREAKING_CHANGES so that the end-user
does not have to do anything when they want to "revert" to the
default comment string.  Then we do not have to give any large text
like the above.  We can instead say something like
	The 'auto' setting of core.commentChar (or core.commentString)
	will change its meaning in Git 3.0 and later and will always
	use the default '#'.

That's certainly simpler for us but it does not help the user to
update their config. Presumably they're using the auto commentchar
because '#' does not work for them.

OK.  But those with "auto" because '#' did not work for them are
setting "auto" not because '#' does not work, but because none of
these "#;@!$%^&|:" work for them, no?

As you said earlier, the "auto" setting cannot fundamentally work at
all if we let a third-party inject any commented material into the
editor buffer.  The comment we inject ourselves we can control (and
notice), and perhaps back in the simpler days when "auto" setting
was invented, it was sufficient.  But that may be no longer true, so
it may not be just "tricky to fix" but simply "unworkable".  From
that point of view, as long as the reason clearly is explained to
end-users, I am fine with "'auto' stops Git and you'd need to unset
or set it to something else at the 3.0 boundary".

Thanks.